Compare commits

..
13 Commits
Author SHA1 Message Date
cflmyandCursor bfd468d5e3 release: Marqdo 1.4.5
Streamable HTTP MCP, web invoke Bearer/401, openai/agent json.get hardening.
Bump CLI/Ext/wasm and debian changelog.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 17:36:07 +08:00
cflmyandCursor 8353ba38fe Add Streamable HTTP MCP and harden web /v1 invoke auth.
agent_mcp_serve gains http transport; invoke injects Bearer/X-Api-Key and maps
unauthorized to 401; openai/agent field access prefers json.get over footnotes.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 17:33:24 +08:00
cflmyandCursor b43a08b2e2 release: Marqdo 1.4.4
Soft LLM transport errors and table-first ext/ai; CLI + Ext + wasm SemVer bump.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 14:38:16 +08:00
cflmyandCursor 1908bf1a53 fix(ai): soft LLM transport errors and strip ext/ai json glue
Host/plugin HTTP failures return structured finish|status=error instead of
aborting the VM; agent/llm/openai/智能体 result maps use GFM tables and
table.put. SKILL gains positive copy-paste examples so agents stop mirroring
legacy json.set chains.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 14:27:45 +08:00
cflmyandCursor c44ea38ae2 release: Ext pack 1.4.3 — agent.run cold EFI / resolve llm_free
Bump ext/VERSION only (CLI stays 1.4.2). Ships agent facade + native miss-field fix for qdagent cold OKF.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 12:56:40 +08:00
cflmyandCursor d489f7c088 fix(ext/ai): cold agent.run EFI — self.preflight and resolve llm_free
Miss resolve omitted llm_free and run called bare preflight, which crashed
qdagent on empty OKF. Align miss fields, use self.preflight + json.get, add
regression gold and research note.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 12:54:00 +08:00
cflmyandCursor a993cfd76a chore: sync Cargo.lock package versions to 1.4.2
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 12:21:23 +08:00
cflmyandCursor d5f826a8c8 release: Marqdo 1.4.2
Bump CLI/wasm/ext to 1.4.2 for Agent/LLM optimize O1–O5 (sandbox trial,
promote proof, step_route/EFI, lint, quarantine).

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 11:51:41 +08:00
cflmyandCursor 0f72173d08 feat(ext/ai): land Agent/LLM optimize waves O1–O5
Wire opt-in sandbox trial into agent, require sandbox proof before OKF
promote, add step_route/EFI/cost surfaces, style lint + quarantine, and
offline policy optimize script—keeping GFM tables over json glue.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 11:47:28 +08:00
cflmyandCursor a707758d47 docs: research and plan Agent/LLM optimization with OKF and sandbox
Capture the 2026-10 landscape (skills, Skill Lift, routing, GEPA) against Marqdo’s code-as-docs assets and list feasible optimize directions without editing doc/next.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 11:13:48 +08:00
cflmyandCursor 63bcb882a6 feat(ext/ai): add CoW sandbox for Marqdo-only agent trials
Give agents a capability-capped run path that stages workspace edits until commit, so generated .mq.md can be verified without polluting the real tree.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-16 11:11:54 +08:00
cflmyandCursor 1bb16dd6db release: Marqdo 1.4.1
View annot leaders mid-to-mid SVG, frame chip id unify; bump CLI/wasm/ext.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-15 20:37:24 +08:00
cflmyandCursor 58ec2172d6 fix(view): mid-to-mid annot leaders and prose chip density
Unify frame chip ids with marks, reorder chips under anchors, and draw
SVG leaders from underline mid-bottom to chip mid-top so crowded bands
stay readable.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-10-15 20:36:12 +08:00
77 changed files with 5370 additions and 750 deletions

No files matched your search

+55 -3
View File
@@ -38,7 +38,7 @@ A `.mq.md` file must stay readable as Markdown. If a human cannot skim the data
Cells are expressions (`doc/design/table-cell-expressions.md`): bare words = text, `` `var` `` = variable, quote paths/`/` ratios.
### 2. Good patterns
### 2. Good patterns (copy these — not `json.set`)
**Wire / steps / commands** — `@` record table:
@@ -60,7 +60,53 @@ Cells are expressions (`doc/design/table-cell-expressions.md`): bare words = tex
| `wire` | `set_text` |
```
**Single dynamic key** — one `table.put` (or a helper), not a json chain:
**Agent / LLM result map** — one readable row, not a `json.parse` + `json.set` chain:
```markdown
`out` =
| status | task | result | llm_calls | tokens |
|--------|------|--------|-----------|--------|
| ok | `task` | `reply` | `llm_calls` | `tokens` |
*out*
```
**Soft error pack** (transport / gate failure) — still a table humans can skim:
```markdown
`pack` =
| finish | text | error | usage | tool_calls |
|--------|------|-------|-------|------------|
| error | "" | `msg` | `usage0` | `empty_tc` |
*pack*
```
**Optional / conditional key** — sparse `table.put` after the static table (one dynamic key only):
```markdown
`out` =
| status | task |
|--------|------|
| error | `task` |
1. `err`
**out = > table.put in=`out` at="error" value=`err`**
2. *
**_ = 1**
```
**Empty list / empty map** — prefer stdlib seeds over `json.parse text=[]` / `{}`:
```markdown
**empty_tc = > table.empty_list**
**empty_map = > table.empty_map**
```
**Single dynamic key from scratch** — one `table.put` (or a helper), not a json chain:
```markdown
**`set_text` = table.put in=None at="#log" value=msg**
@@ -73,14 +119,20 @@ Better still — call a named helper:
**browser.set_text sel="#log" text=msg**
```
### 3. Bad pattern (do not write)
**When `json.*` is OK:** `json.parse` / `json.stringify` / `json.quote` on **wire text**; `json.get` to read an opaque map. Never use `json.set` / `json.append` to *build* author-facing structures.
### 3. Bad pattern (do not write — and do not copy from old `ext/ai`)
```markdown
**`ret` = json.set map=None key="set_text" value=…**
**`ret` = json.set map=ret key="canvas" value=…**
**`cmds` = json.append list=cmds item=c0**
**`out` = json.parse text={"status":"ok"}**
**`out` = json.set map=out key="task" value=task**
```
Matching surrounding **legacy** `json.set` in `ext/ai/*.mq.md` is **not** compliance. Refactor the call site to tables / `table.put` instead of extending the glue.
## Writing application `.mq.md`
1. Frontmatter: title + imports (`table`, domain libs). **Do not import `json` unless parsing strings.**
+51 -163
View File
@@ -4,95 +4,40 @@ description: >-
Author and edit Marqdo programs (.mq.md): Markdown markers are executable
syntax (functions, calls, statements, returns, control flow). Use when writing
or modifying .mq.md files, teaching Marqdo, generating Marqdo from prose,
importing lib/* or ext/web or ext/quantum or ext/linalg, running marqdo CLI,
building dynamic sites with ext/web, quantum circuits with ext/quantum,
formula-first linear algebra with ext/linalg, or when the user mentions
Marqdo, mq.md, markup-as-syntax, code-as-documentation, web/网页, quantum/量子,
or linalg/线性代数 extension.
importing lib/* or ext/web or ext/quantum, running marqdo CLI, building dynamic
sites with ext/web, quantum circuits with ext/quantum, or when the user
mentions Marqdo, mq.md, markup-as-syntax, code-as-documentation, web/网页, or
quantum/量子 extension.
---
# Marqdo development
Marqdo = **Markdown markers as syntax**. A `.mq.md` file is both documentation and a runnable program. Read this skill **before** inventing syntax from Markdown habits or Python.
Canonical design (repo): [`markdown-mapping-v0.3.md`](../../doc/design/markdown-mapping-v0.3.md) (**v0.3 · runtime**), `doc/design/keywords.md`. Legacy v0.2 notes: `doc/design/markdown-mapping.md` (historical). Deeper tables: [reference.md](reference.md). Copy-paste patterns: [examples.md](examples.md).
Canonical design (repo): `doc/design/markdown-mapping.md`, `doc/design/keywords.md`. Deeper tables: [reference.md](reference.md). Copy-paste patterns: [examples.md](examples.md).
**Authoring / library craft** (GFM-first, no json glue, how to develop Marqdo): use **[marqdo-dev](../marqdo-dev/SKILL.md)** whenever writing or refactoring `.mq.md`, `lib/*`, or `ext/*`.
**Product release** (tag, CHANGELOG, README, public docs, VSIX, GitHub assets, **ext → R2+GitHub**, **CLI → Ubuntu PPA**): use [marqdo-release](../marqdo-release/SKILL.md) — detect versions, **ask for SemVer + mode (`full` / `cli` / `ext`)**, then follow that playbook. Ext-only releases are allowed. If sudo/GPG/PPA steps need the user, print the exact commands.
**Product release** (tag, CHANGELOG, README, public docs, VSIX, GitHub assets): use [marqdo-release](../marqdo-release/SKILL.md) — detect latest version, **ask the user for the next SemVer**, then follow that playbook. Do not invent a version number.
## Hard rules (do not violate)
1. **File suffix** must be `.mq.md` (never plain `.md` for executable sources).
2. **`**…**` = code** (assign / call). **`*…*` = return value**. Output is **not** either marker — use `> print text=…` / `**print text="…"**` / `> 打印 内容=…`.
3. **Prose:** unmarked narrative may contain `` `名` `` (declare/ref), inline `**code**`, and `*return*`. Soft Markdown emphasis that is **not** code-shaped (`**说明**`, mid-line `*only*`) is ignored — see [code-as-docs-gaps.md](../../doc/design/code-as-docs-gaps.md). Dead `` `名` `` (never read) is OK and does not become a param.
4. **Do not invent keywords** `if` / `else` / `while` / `for` / `def` / `return`. Control flow is Markdown: **`1.` `2.` … branches**, **`-` loops**, arm `N. *` = else, `#` = **object/type**, `##`+ = **function/method**. Params: prefer prose `` `名` `` / `` `名`=默认 `` (inferred); `` + `名` `` still accepted.
5. **Identifiers:** `` `名` `` in prose; inside `*…*` / `**…**`, bare ids are **variables**. **Text literals must be quoted** in bold/italic: `**print text="hi"**`. Standalone `>` calls: bare words = text; vars need ticks: `` > str `n` ``.
6. **Structure lines** (`#` `>` `+` `-` `|` `1.`) need not be wrapped in bold.
7. **Paths:** In bare expressions `/` is division. In call args / defaults / table cells, unspaced `a/b` and quoted paths are text.
8. Prefer ending side-effect-only bodies with `*None*` / `*无*` / whole-line `**` / `****` when the next lines must belong to an outer function. Lone `---` / `***` are Markdown thematic breaks (skipped everywhere), not function end.
2. **Output is not `**bold**`.** Print with `> print text=…` (or `> 打印 内容=…`). Bold `**…**` is **return value only**.
3. **Prose vs code:** after a blank line, an unmarked first line starts a **comment paragraph**; every following non-blank line stays comment until the next blank line. Put a **blank line** between narration and executable lines.
4. **Do not invent keywords** `if` / `else` / `while` / `for` / `def` / `return`. Control flow is Markdown: **`+` params** (under heading), **`1.` `2.` … branches**, **`-` loops**, arm `N. *` = else, `#` = **object/type**, `##`+ = **function/method**, `**…**` = return.
5. **Identifiers use backticks** — params `` + `name` ``. **Exemptions:** foreach `` - [item](coll) ``; footnote `` name[^key] ``; **inside italic `*…*` / bold `**…**`**, bare ids are **variables** (Python-style unified namespace) and method-call receivers are bare too. **Drop the backticks on value-expression variable names and method receivers inside `*…*` / `**…**`** — they are redundant once the `*…*`/`**…**` markers mark the segment as code: `*分类 = 分类[^苹果]*`, `**n * 2**`, `*p = > page.主体装配 组件=home*`. **Text literals must be quoted**: `*a = > text.split value=src sep=","*`. Backticks **still required** where a bare word is *not* a variable in **standalone `>` calls** (bare = text there): `` > str `n` ``, `` > `obj`.method ``. **No trailing space** inside `*…*` / `**…**` wrapped code: the closing marker must touch the last token directly (`*a = 1*`, `**n**`), never `` *a = 1 * ``.
6. **Structure lines are not wrapped in italics.** `#` `>` `+` `-` `|` lines stand alone; use `*…*` only for general statements (bindings / expressions).
7. **Paths:** In bare *expressions* `/` is division. In **call args / param defaults / table cells**, unspaced `a/b` and quoted `".marqdo/agent-kb"` are path text — no `json.parse` needed. Numeric ratios that must stay text use quotes (`"1/5"`, `"16/9"`); bare `1/5` is division.
8. Prefer ending side-effect-only function bodies with `*None*` / `*无*` / whole-line `**` / `****` when later lines must belong to an outer function. Lone `---` / `***` are Markdown thematic breaks (skipped), not function end.
9. **`ext/**` never calls `host_*`.** Agent/OKF helpers are plugin names (`agent_kb_*`, …) after `plugin.load`. Do **not** add agent/OKF domain code to `src/host/` (core bloat). See `doc/design/ext-agent.md` §4.
10. **Browser (route C/D/E/F):** client logic is Marqdo on WASM; official bridge is host glue and **may** implement lists, routing, storage, WebSocket, canvas, file read, observers — authors **must not hand-write business JS**. Use `web.client_embed` / `data-mq-source-url` auto-mount. Client effects: prefer **GFM tables + `lib/browser`** ([marqdo-dev](../marqdo-dev/SKILL.md)).
11. **Code-as-documentation / no bag glue:** Prefer **GFM tables** for maps, lists, wire, commands. Prefer **`table.put` / named helpers** over `json.set` / `json.append` chains. `lib/json` is for parse/stringify/quote only — not a dict builder. Unreadable json pipelines are a style bug.
12. **Bracket-marked calls:** Function name may be marked with `[函数名]`; bare words before `[` are boolean modifiers (`礼貌 [问候] x` → `礼貌=True`). `]` immediately followed by `(` is **index** `[key](coll)`, not a call. Prefer `[k](c)` over footnote index in new code. See [bracket-call-modifiers.md](../../doc/design/bracket-call-modifiers.md).
11. **Code-as-documentation / no bag glue:** Prefer **GFM tables** for maps, lists, wire, commands. Prefer **`table.put` / named helpers** over `json.set` / `json.append` chains. `lib/json` is for parse/stringify/quote only — not a dict builder. Unreadable json pipelines are a style bug. Read collections with **`[key](map)`**, not footnote indexes for new code. **Positive copy-paste shapes** (result map, soft-error pack, sparse `table.put`): [marqdo-dev](../marqdo-dev/SKILL.md) §2 · [examples.md](examples.md) §6e. Do **not** extend legacy `json.set` in `ext/ai` — refactor to tables.
12. **One language per file:** Do not mix EN and ZH API names in the same `.mq.md` (imports, calls, params). Pick English *or* Chinese mirrors for the whole program.
13. **Sandbox trial habit (Agent / generated `.mq.md`):** Prefer `agent.trial` / `run sandbox=True` / `sandbox.open`→`run`→`diff`→`commit` before promoting to the host tree. Host files must stay unchanged until a successful commit. Caps live in `sandbox.profile` GFM tables — do not invent a second dialect.
## 记不住语法细节?查 MLSP(AI 必读)
**不必背 Marqdo 语法细节**——语法可查、程序可验、修复有界。写码前后走**查询式**工作流:
```bash
printf '%s\n' '{"id":1,"method":"syntax","params":{"query":"返回"}}' | marqdo mlsp
printf '%s\n' '{"id":2,"method":"locate","params":{"path":"a.mq.md","symbol":"加一"}}' | marqdo mlsp
printf '%s\n' '{"id":3,"method":"validate","params":{"path":"a.mq.md"}}' | marqdo mlsp
printf '%s\n' '{"id":4,"method":"schema","params":{"path":"a.mq.md","unit":"加一"}}' | marqdo mlsp
printf '%s\n' '{"id":5,"method":"repair_targets","params":{"code":"contract.arg_mismatch","doc_anchor":"a.mq.md#L9-L15"}}' | marqdo mlsp
```
| method | 何时用 |
|--------|--------|
| `syntax` | 不确定某个语法怎么写(关键字/别名 → 构造卡片:「返回」「表」「调用」「循环」…) |
| `locate` | 找符号定义位置 + 契约;或反查构造卡片 |
| `validate` | 提交前校验(结构化诊断数组;`ok:false` 就修) |
| `schema` | 查单元契约(形参/返回/字段) |
| `repair_targets` | 拿错误的**有界**修复靶点:只改锚点范围内行,越界必弃权(abstain) |
行分隔 JSON:每行请求 `{"id","method","params"}` → 每行响应 `{"id","ok","result"|"error"}`。
语法唯一事实来源 = `doc/design/core-surface.md`(CI 守卫同步);卡片带 `manual` 指回文档行。
所有错误诊断都是结构化 JSON:`code` / `severity` / `span` / `suggestion` / `doc_anchor` / `doc_quote` / `contract_ref`。
### 渐进式契约(可选标注;schema 就长在文档表里)
契约是**未绑定文档表**(绑定表是数据,永远不是契约)。三种形状、严格位置:
| 形状(须含 `类型` 列) | 允许位置 | 附着 |
|--------------------------|----------|------|
| `参数`/`类型`/`说明` 表 + `返回`/`类型`/`说明` 表 | 函数体首个可执行行之前 | 函数 |
| `字段`/`类型`/`可空`/`说明` 表 | 紧邻 `` `名` = `` 绑定之前 | 集合(变量) |
| `字段`/`类型`/`可空`/`说明` 表 | 对象(`#`)体首成员之前 | 对象类型 |
````text
## 加一
对于输入变量`n`,期望是数字。
| 参数 | 类型 | 说明 |
|------|------|------|
| n | number | 输入值 |
| 返回 | 类型 | 说明 |
|------|------|------|
| | number | 加一结果 |
*n+1*
````
- 类型词汇:`text` / `number` / `bool` / `list` / `map` / `any` / `#` 对象名(拼错 ⇒ `contract.unknown_type` + 最近建议)。
- 契约表**永不执行**、不污染升参推断;无契约文件行为逐字节不变(动态默认,零摩擦)。
- 运行时只在四个边界校验:调用实参 / 字段访问 `[键](集合)` / 表绑定 / 返回。
- `marqdo check [FILE] [--json]` 静态互查:契约行 vs 升参推断形参**双向**漂移(错行=Error;新升参未覆盖=Warning)、未知类型、错位契约;有 Error 非零退出。
- 契约诊断带 `doc_anchor` + `doc_quote`(表原文)——修复只照锚点局部改。
## Markup → meaning (v0.3)
## Markup → meaning (v0.2)
| Marker | Meaning |
|--------|---------|
@@ -103,20 +48,17 @@ printf '%s\n' '{"id":5,"method":"repair_targets","params":{"code":"contract.arg_
| `- …` inside body | Loop (`while` or `` [item](coll) `` / `` [`item`](`coll`) ``) |
| Line `N. *` | Else arm |
| `> fn args` | Call (named `k=v` or positional) |
| `修饰 [fn] args` | Bracket-marked call; leading words → `名=True` |
| `` > `obj`.method args `` | Method call (`obj` must be a map with `_type`) |
| Frontmatter `import bind:path.mq.md` / `导入` | Import file (bind library name) |
| Frontmatter `import bind:lib.member` / `导入` | Short name for a library member (same keyword; no separate `use`) |
| `> lib.member …` / `> lib.Type.member …` | Call via bare dotted path (instance methods need `` `var`.m ``) |
| `**…**` | **Code** (assign / call) — closing `**` touches last token; bare ids are variables; quote text literals |
| `*…*` | **Return value** — closing `*` touches last token; bare ids are variables |
| `****` / whole-line `**` / `*None*` | Return `None` and end function body |
| `*…*` | Statement (bind / expr) — closing `*` touches last token, **no trailing space**; bare ids are variables |
| `**…**` | Return value — closing `**` touches last token, **no trailing space**; bare ids are variables |
| `****` or `**` + spaces + `**` | Return `None` and end function body |
| Lone `---` / `***` | Markdown thematic break (skip; not function end) |
| GFM table after empty RHS bind | Collection; get with `[key](coll)` (preferred) or legacy `` `m`[^key] `` |
| `` ```lang `` | Foreign code block (via `lib/foreign`); bind with `` `name` = `` then fence |
| `$$` + `% marqdo: var=name` | Block formula bind (TeX comment; prefer over empty RHS) |
| `- [item](coll)` | Foreach (**preferred**, nestable). Sugar: `- 对`coll`的每一项`item`` / `- for each `item` in `coll`` |
| `![marqdo-out](path)` | Image writeback form (validated stem.ext / URL); text writeback stays HTML comment |
| `- [item](coll)` | Foreach (unchanged) |
| `` ```lang `` | Foreign code block (via `lib/foreign`) |
| `` `"text"` `` | Quoted string (`\n` `\t` `\\` `\"`; `` `var` `` inside); bare tokens unescaped |
| Unmarked prose | Comment |
@@ -140,19 +82,20 @@ Chinese builtins (same functions, no import):
## Functions, calls, end-body
Prefer **body-inferred params** and **prose-unified** wording (v0.3). `` + `name` `` still works, but teach the inferred form first:
```markdown
# main
**greet who="Marqdo"**
> greet who=Marqdo
## greet
+ `who`
When we meet`who`, we **print text="Hello, `who`!"** and close with*None*.
> print text=Hello, `who`!
*None*
```
- Call: `**fn key="text"**` (bold; quote text literals) or `> fn key=text` (standalone `>`; bare words are text).
- Call: `> name key=value` or `> name value`.
- Nested helpers use deeper `#` (`##`, `###`, …).
- After a helper with only side effects, end with `*None*` / whole-line `**` / `****` when later lines must stay in the outer function. Lone `---` is only a thematic break (skipped).
- In bold assigns, prefer bare calls: `**摘要 = 结账摘要 …**` (omit `>`).
@@ -160,16 +103,17 @@ When we meet`who`, we **print text="Hello, `who`!"** and close with*None*.
## Statements, returns, branches
```markdown
**x = 1**
**y = x + 2**
*x = 1*
*y = x + 2*
## add_one
+ `n`
For input`n`, do **n = n + 1** and return*n*.
**n + 1**
```
```markdown
**n = 0**
*n = 0*
1. `n` > 0
> print text=positive
@@ -192,16 +136,16 @@ import clock:lib/time.mq.md
# main
**xs = > text.str_split s="a,b,c" sep=","**
*xs = > text.split value="a,b,c" sep=","*
> print text=`xs`
**t = > clock.now_unix**
*t = > clock.now_unix*
```
- Import **English** or **Chinese** library file; call that file’s API names via `lib.member` (do not mix languages).
- Instance methods stay `` > `obj`.method `` (backticks on the receiver only).
- `lib/…` resolves via `MARQDO_LIB`, cwd `lib/`, or `lib/` next to the `marqdo` binary.
- Design: [module-namespace.md](../../doc/design/module-namespace.md).
- Official optional extensions (`ext/`, not stdlib): install with `marqdo ext add llm|agent|web|quantum|linalg` (see `doc/design/ext-cli.md`). **`ext/llm`** — intelligence primitive (`ask` / `stream` / `collect`; `complete` is compat); **`ext/agent`** — document-driven agents (see **ext/agent** below); **`ext/web`** — dynamic sites (see below); **`ext/quantum`** — circuits + Q7/Q8 (see below); **`ext/linalg`** — formula-first linear algebra (see below).
- Official optional extensions (`ext/`, not stdlib): install with `marqdo ext add llm|agent|web|quantum` (see `doc/design/ext-cli.md`). `ext/llm` — chat; **`ext/agent`** — document-driven agents (see **ext/agent** below); **`ext/web`** — dynamic sites (see below); **`ext/quantum`** — circuits + Q7/Q8 (see below).
- Native plugins: `lib/plugin` — `## load` / `unload` / `list`. C ABI: `include/marqdo_abi.h`.
- Prefer `table.put` / `表.改` for list/map element updates; keep `json` for parse/stringify/quote (see `doc/design/stdlib-table.md`).
- Builtins (no import): `print`/`打印`, `input`/`输入`, `len`/`长度`, `str`/`文本`, `int`/`整数`; literals `True`/`真`, `False`/`假`, `None`/`空`; logic `and`/`且`, `or`/`或`, `not`/`非`.
@@ -216,61 +160,24 @@ Install: `marqdo ext add agent` (ZH: `智能体`). Native: `cargo build --releas
1. **Code is documentation** — the runbook `.mq.md` is the model’s ground truth, not a hidden system prompt bag.
2. **Call site is first-class** — `build_step_context` injects path/function/line so the model knows where it is in the program.
3. **Document is the knowledge base (OKF)** — writeback → promote/solidify → reuse / `llm_free`; MCP/corpus are evidence only. **Engineering Knowledge(EKC)**:`marqdo knowledge|find|reuse|…` + `agent.preflight`(REUSE→ADAPT→CREATE)——见 [engineering-knowledge.md](../../doc/design/engineering-knowledge.md)。
4. **Skill Compilation (v2)** — successful runs become `.mq.md` episodes; after enough evidence, `maybe_learn` compiles an executable skill (`llm_free`). Prefer `## run` over bare `## plan` when you want learning.
5. **Adaptive Routing (P4)** — router is abstract (`marqdo` / `small-llm` / `llm` / `auto`); `jev` is a compat alias of `small-llm`. Decision via `router_model` (any llm handle). Policy Compilation writes `policies/router.mq.md`.
3. **Document is the knowledge base (OKF)** — writeback → promote/solidify → reuse / `llm_free`; MCP/corpus are evidence only.
APIs: `# agent` / `# 智能体` → `## run` / `## route` / `## step` / `## plan` (+ `resolve` / `record_episode` / `maybe_learn` / `compile` / `compile_policy` / `metrics`). Context: standing + task + **call site** + tools + budgeted source/skill + act protocol (`CALL:` / `READ:`). Inspect without LLM: `dump_step_context`. Inside `writeback=` / `写回=` branches use `host_writeback_record` (param name shadows the writeback import).
APIs: `# agent` / `# 智能体` → `## step` / `## plan`. Context: standing + task + **call site** + tools + budgeted source/skill + act protocol (`CALL:` / `READ:`).
Examples: [agent-pong](../../examples/agent-pong/) · [agent-okf-flywheel](../../examples/agent-okf-flywheel/). Harness: `scripts/agent-harness.sh`. Design: [ext-agent.md](../../doc/design/ext-agent.md) · Wave B: [agent-framework-2026-09.md](../../doc/research/agent-framework-2026-09.md) · v2: [doc/next/003.md](../../doc/next/003.md) · P4: [doc/next/004.md](../../doc/next/004.md). Dev LLM: copy [.env.example](../../.env.example) → `.env` (gitignored).
Examples: [agent-pong](../../examples/agent-pong/) · [agent-okf-flywheel](../../examples/agent-okf-flywheel/) · [agent-sandbox-okf](../../examples/agent-sandbox-okf/). Harness: `scripts/agent-harness.sh`. Design: [ext-agent.md](../../doc/design/ext-agent.md) · optimize O1–O5: [ext-ai-optimize.md](../../doc/roadmap/ext-ai-optimize.md). Dev LLM: copy [.env.example](../../.env.example) → `.env` (gitignored).
**Sandbox (O1):** `import sandbox:ext/ai/sandbox.mq.md` · `agent.trial` / `智能体.试跑` · `run`/`step` take `sandbox=` / `沙盒=` (default `False`). Profile caps are tables via `sandbox.profile`. Gold: `tests/ext/agent-sandbox-optin.mq.md`.
## Official extension: `ext/web` (Document / Endpoint / Resource)
Install: `marqdo ext add web` (ZH id: `网页`). Build native plugin: `scripts/build-web-plugin.sh` / `marqdo ext add web`.
Install: `marqdo ext add web` (ZH id: `网页`). See ADR 0007 · [ext-web-artifact.md](../../doc/design/ext-web-artifact.md).
**Constitution (ADR 0007 — do not degrade):**
**Do not degrade:** Document/Endpoint metadata; facade `web.page|route|serve|render|inspect`; View `|type|slot|value|attrs|style|` only; no `compose_*` / `configure` / author `ensure_plugin`. DB → `ext/data`; auth → `ext/security`; net → `ext/net`.
1. **Document / Endpoint / Resource** — `.mq.md` with `type: web|endpoint` (ZH `类型: 网页|端点`) is the site surface; not a Flask-style class pile.
2. **Code is documentation** — the file is docs + route + handler.
3. **Table = data** — View schema only `|type|slot|value|attrs|style|`. No `compose_*` dialects, no `json.set` glue.
4. **Facade only** — `web.page` · `web.route` · `web.serve` · `web.render` · `web.inspect`. DB → `ext/data`; auth → `ext/security`; ws/http → `ext/net`.
5. **No author `ensure_plugin` / `host_*`**; import web = capability (plugin auto-load).
6. **Declared vs system routes**; `web.inspect` lists both. No silent user routes.
Import **one language** per file:
- EN: `import web:ext/web/web.mq.md` (+ `import data:ext/data/db.mq.md` as needed)
- EN: `import web:ext/web/web.mq.md`
- ZH: `导入 网页:ext/web/网页.mq.md`
```markdown
---
type: web
route: /
title: Hello
import web:ext/web/web.mq.md
---
# Hello
Executable document as a page.
```
```markdown
---
type: endpoint
method: POST
path: /api/ping
response: json
---
*{"ok": true}*
```
Site entry may `# main` → `web.serve root="."`. Client WASM: `ext/web/client.mq.md` / `lib/browser` — no author JS.
Design: [ext-web-artifact.md](../../doc/design/ext-web-artifact.md) · ADR [0007](../../doc/adr/0007-web-document-endpoint.md) · rule: `.cursor/rules/web-artifact-no-degrade.mdc` · example: [examples/marqdo-blog/](../../examples/marqdo-blog/).
Web patterns: [examples.md](examples.md) · API index: [reference.md](reference.md).
Design: [ext-web-artifact.md](../../doc/design/ext-web-artifact.md) · example: [examples/marqdo-blog/](../../examples/marqdo-blog/).
## Official extension: `ext/quantum` (circuits + Q7/Q8)
@@ -296,29 +203,13 @@ Install: `marqdo ext add quantum` (ZH id: `量子`). Build: `cargo build --relea
Design: [ext-quantum.md](../../doc/design/ext-quantum.md) · Q7: [ext-quantum-q7.md](../../doc/design/ext-quantum-q7.md) · Q8 viz: [ext-quantum-viz-style.md](../../doc/design/ext-quantum-viz-style.md) · examples: [quantum-bell](../../examples/quantum-bell/) · [quantum-entanglement](../../examples/quantum-entanglement/).
## Official extension: `ext/linalg` (formula-first LA)
Install: `marqdo ext add linalg` (ZH id: `线性代数`). Build: `cargo build --release -p marqdo_plugin_linalg`, then **`marqdo ext add linalg` again** so view loads the new native lib.
- EN: `import la:ext/linalg/linalg.mq.md`
- ZH: `导入 la:ext/linalg/线性代数.mq.md`
**Hard rules:** `ext/**` never calls `host_*`; hot path is ABI plugin; prefer top-level `la.symbol` (not `matrix.symbol` without a receiver); quote `kind=` strings; dense `explicit` / factorize obey size guards.
APIs (EN): `symbol` → `# matrix`;中缀 `+`/`-`/`*`;方法 `` `A`.T `` / ``.simplify`` / ``.ascii`` · `from_formula` / `explicit` / `factorize` / `draw` · `lstsq` / `norm` / `cond` / `rank`.
Examples: [linalg-svd](../../examples/linalg-svd/) · [linalg-least-squares](../../examples/linalg-least-squares/). Design: [ext-linalg.md](../../doc/design/ext-linalg.md) · formula-doc: [ext-linalg-formula-doc.md](../../doc/design/ext-linalg-formula-doc.md).
## AI authoring workflow
1. **查再写**:不确定语法就 `marqdo mlsp`(`syntax` / `locate`)——不必背语法细节。
2. Decide language surface (English builtins + `lib/text.mq.md`, or Chinese + `lib/文本.mq.md`).
3. Write `.mq.md` with `# main`, blank lines around prose, correct markers.
4. 可选:给函数/集合写**渐进式契约**(上节三种表格形状;写全或不写,别写一半)。
5. Run: `marqdo run path/to/file.mq.md` (cwd = project root so `lib/` resolves).
6. On errors:结构化诊断(含 `doc_anchor` / `suggestion`)→ `repair_targets` 取有界靶点,**只改锚点范围内**的行。
7. 提交前:`marqdo check path/to/file.mq.md`(或 `mlsp validate`)清到零 Error。
8. Optional: `marqdo view .` / `marqdo debug .` for structure and breakpoints.
1. Decide language surface (English builtins + `lib/text.mq.md`, or Chinese + `lib/文本.mq.md`).
2. Write `.mq.md` with `# main`, blank lines around prose, correct markers.
3. Run: `marqdo run path/to/file.mq.md` (cwd = project root so `lib/` resolves).
4. On `path:line:col: message`, fix that span; re-run until exit 0.
5. Optional: `marqdo view .` / `marqdo debug .` for structure and breakpoints.
## Anti-patterns (common model mistakes)
@@ -339,8 +230,7 @@ Examples: [linalg-svd](../../examples/linalg-svd/) · [linalg-least-squares](../
| Mix `web.page` and `网页.页面` in one file | One import language per `.mq.md` |
| Call `host_web_*` from `ext/web` | Use `# app` / `# db` methods; plugin ABI only |
| Call `host_*` from `ext/quantum` | Use `# circuit` / `# density` methods; plugin ABI only |
| Call `host_*` from `ext/linalg` | Use `la.*` L1 wrappers; plugin ABI only |
| Bare `kind=eig` inside `*…*` | `kind="eig"` — quote text literals |
| Bare `kind=hinton` inside `*…*` | `kind="hinton"` — quote text literals |
| Bare `theme=dark` inside `*…*` | `theme="dark"` — quote; ZH `主题="dark"` |
| Rebuild plugin but skip `ext add` | `cargo build -p marqdo_plugin_quantum` then `marqdo ext add quantum` (or set `MARQDO_QUANTUM_PLUGIN`) |
@@ -357,5 +247,3 @@ Examples: [linalg-svd](../../examples/linalg-svd/) · [linalg-least-squares](../
- [ ] Browser client: `lib/browser` + tables when returning effects
- [ ] Quantum draw: quote `kind=` / `theme=`; rebuild plugin → `ext add quantum` before view
- [ ] `marqdo run` succeeds (or errors addressed)
- [ ] 语法拿不准的地方已查 MLSP(`syntax`/`locate`),而不是凭记忆发明
- [ ] 契约(如有)与升参一致:`marqdo check` 零 Error
+50
View File
@@ -162,6 +162,56 @@ import browser:lib/browser.mq.md
Do **not** build effect maps with `json.set` chains. See [marqdo-dev](../marqdo-dev/SKILL.md).
## 6e. Agent / LLM result maps — GFM first (positive)
Return and soft-error packs are **tables**, not `json.set` pipelines. Full style contract: [marqdo-dev](../marqdo-dev/SKILL.md) §2.
```markdown
---
import table:lib/table.mq.md
---
## step_ok
+ `task`
+ `reply`
`out` =
| status | task | result |
|--------|------|--------|
| ok | `task` | `reply` |
*out*
## step_err
+ `task`
+ `err`
`out` =
| status | task | error |
|--------|------|-------|
| error | `task` | `err` |
*out*
## soft_http_pack
+ `msg`
+ `usage0`
**empty_tc = > table.empty_list**
`pack` =
| finish | text | error | usage | tool_calls |
|--------|------|-------|-------|------------|
| error | "" | `msg` | `usage0` | `empty_tc` |
*pack*
```
Do **not** mirror legacy `ext/ai/agent.mq.md` `json.parse`+`json.set` chains when adding new arms — convert the shape to a table.
## 7. Import text stdlib
```markdown
+7 -5
View File
@@ -126,6 +126,12 @@ jobs:
## Marqdo ${{ github.ref_name }}
### 亮点 / Highlights
- **v1.4.5**:Streamable HTTP MCP + web invoke Bearer/401 · openai/agent json.get 硬化
- **v1.4.5**: Streamable HTTP MCP + web invoke Bearer/401; openai/agent json.get hardening
- **v1.4.2**:Agent/LLM 优化 O1–O5——sandbox trial · promote proof · step_route/EFI/cost · lint · quarantine
- **v1.4.2**: Agent/LLM optimize O1–O5 — sandbox trial, promote proof, step_route/EFI/cost, lint, quarantine
- **v1.4.1**:view 标注连线——SVG mid→mid(underline 底中 → chip 顶中)· 帧 chip id 对齐 · L→R 密度
- **v1.4.1**: View annot leaders — SVG mid→mid (mark bottom-center → chip top-center), frame chip id unify, L→R density
- **v1.4.0**:叙述统一——`% marqdo: var=` · 臂/循环体叙述 · foreach 中英糖 · 图片自写回 · view 推断形参
- **v1.4.0**: Prose-unified control flow — formula meta bind, arm/loop narrative, foreach sugar, image writeback, inferred-param view
- **v1.3.3**:view 文档面 Structure——叙述保真 · 标记面 · 调用/循环帧 · `#mq-view-trace` · 静态站同源
@@ -134,12 +140,8 @@ jobs:
- **v1.3.0**: Engineering Knowledge Compiler — L0–L4 catalog, reuse/find/duplicate, Agent preflight (REUSE→ADAPT→CREATE)
- **v1.2.0**:Web Artifact(ADR 0007)——Document/Endpoint · `web.serve`/`render`/`inspect` · `ext/data|security|net` · 移除 `compose_*`/`configure`
- **v1.2.0**: Web Artifact (ADR 0007) — Document/Endpoint, web.serve/render/inspect, ext/data|security|net, compose_*/configure removed
- **v1.1.1**:AI 栈收口——llm `ask`/`stream`/`collect` · Metadata Binding · Skill Compilation + Adaptive Routing · MCP 客户端 + resume · 可选 `plugins/llm`
- **v1.1.1**: AI stack — llm ask/stream/collect, Metadata Binding, Skill Compilation + Adaptive Routing, MCP client + resume, optional plugins/llm
- **v1.0.7**:WS/网络边界加固;`url_encode` `%20`
- **v1.0.3**:Ubuntu PPA + 扩展 CDN(ext.marqdo.com)
- **Linux 包必出**:`*-native-x86_64-unknown-linux-gnu.zip` + Linux CLI bundle
- Details: [CHANGELOG](https://github.com/cflmy/marqdo/blob/main/CHANGELOG.md) · [mapping v0.3](https://github.com/cflmy/marqdo/blob/main/doc/design/markdown-mapping-v0.3.md)
- Details: [CHANGELOG](https://github.com/cflmy/marqdo/blob/main/CHANGELOG.md) · [ext-ai-optimize](https://github.com/cflmy/marqdo/blob/main/doc/roadmap/ext-ai-optimize.md)
### Downloads
+6
View File
@@ -87,3 +87,9 @@ tests/ext/:memory:
# local secrets
.env
# sandbox gold leftovers
tests/ext/_sbx_cow/
tests/ext/_sbx_*/
examples/**/_example_sbx/
+133
View File
@@ -8,6 +8,139 @@
### Changed
## v1.4.5 — 2026-10-16
### Highlights
**Streamable HTTP MCP + web `/v1` 鉴权硬化**:`agent_mcp_serve` 支持 `transport=http`(MCP Streamable HTTP 2025-03-26);web invoke 注入 `Authorization` / `X-Api-Key`,`ok=false`+`unauthorized` → HTTP 401;`ext/ai` openai/agent 字段访问改 `json.get`,消除 footnote 硬崩。CLI · Ext · wasm **1.4.5**。
```bash
git checkout v1.4.5 && cargo build --release
cargo build --release -p marqdo_plugin_agent
# HTTP MCP:
# QDAGENT_MCP_TRANSPORT=http marqdo run cli/求道-mcp.mq.md
marqdo ext add agent # CDN → 1.4.5
```
**核心表面 Δ(Core surface Δ)**:**新增** MCP Streamable HTTP 传输(`transport=http|streamable-http`);web invoke 契约扩展(header 注入 · 401);**无** MQ 关键词 / 表几何变更。
### Added
- **`agent_mcp_serve` Streamable HTTP**:`transport=http` · `host`/`port`/`path`;POST JSON-RPC + 可选 GET SSE;`Mcp-Session-Id`。facade:`ext/ai/agent` `## serve`。
### Fixed
- **web invoke**:`Authorization` / `X-Api-Key` 注入 payload;`jsonWrapInvokeValue` 对 `unauthorized` 返回 401;`{sse}` 自动 event-stream。
- **`ext/ai/llm/openai` · `agent` · `llm`**:字段访问改 `json.get` / `at`,避免 `[field](x)` 在非 map 上硬崩。
- **explore workbook**:子任务 `stream=False`(网关 SSE 524 更稳)。
### Changed
- CLI / Ext / wasm SemVer **1.4.5**。
- 设计:[doc/design/ext-web-artifact.md](doc/design/ext-web-artifact.md)(sse / bearer)。
## v1.4.4 — 2026-10-16
### Highlights
**LLM 传输软失败 + `ext/ai` 表优先重构**:上游连接拒绝/超时/HTTP≠200 时,Runtime 返回结构化 `finish|status=error`(不再硬中断 VM);`agent` / `智能体` / `llm/openai` 去掉 `json.set`/`json.append` 袋胶水。含 CLI host · `plugins/llm` · Ext pack **1.4.4**。
```bash
git checkout v1.4.4 && cargo build --release
cargo build --release -p marqdo_plugin_llm
marqdo run tests/ext/agent-llm-transport-soft.mq.md
marqdo ext add agent # CDN → 1.4.4
```
**核心表面 Δ(Core surface Δ)**:**无增删**——HTTP 宿主软失败契约 + Extension 风格重构;不改 MQ 关键词 / 表几何。
### Fixed
- **LLM / Agent 传输软失败**:`host_http_post` / SSE · `plugins/llm` · `ext/ai/llm/openai` 结构化 `finish=error`;`agent.step` / `plan` / `run` 收成 `status=error`。金样 `tests/ext/agent-llm-transport-soft.mq.md`。说明:[doc/research/agent-llm-transport-soft-2026-10.md](doc/research/agent-llm-transport-soft-2026-10.md)。
- **`agent.run`**:`cache == "soft-hit"` 引号比较(裸 `soft-hit` 曾被解析为减法)。
### Changed
- **`ext/ai` 去 json 胶水**:结果 map / 事件用 GFM 表 + `table.put` / `table.append` / `table.empty_*`。SKILL 正面示例:`.cursor/skills/marqdo-dev` §2 · `marqdo/examples.md` §6e。
- CLI / Ext / wasm SemVer **1.4.4**。
## Ext pack v1.4.3 — 2026-10-16
### Highlights
**Agent cold EFI / resolve 硬崩修复(CLI 仍为 v1.4.2)**:qdagent 接通 `agent.run` 时空 OKF 触发 `unknown function preflight` 与 `missing map key llm_free`。本包修复 `ext/ai/agent` facade 与 `agent` native;金样 `tests/ext/agent-run-cold-efi.mq.md`。说明:[`doc/research/agent-run-efi-resolve-bugs-2026-10.md`](doc/research/agent-run-efi-resolve-bugs-2026-10.md)。
```bash
marqdo ext add agent # CDN latest → 1.4.3
# 或:MARQDO_EXT_VERSION=1.4.3 marqdo ext add agent
```
**核心表面 Δ(Core surface Δ)**:**无增删**——仅 Extension / Agent Runtime 契约修补。
### Fixed
- **`agent.run`**:`efi=True` 改为 `` `self`.preflight ``(不再裸调全局 `preflight`)。
- **`agent_resolve` miss**:返回字段对齐命中路径,含 `llm_free: false`;facade 对 `llm_free` / route 字段改用 `json.get`。
### Changed
- 扩展包 SemVer **1.4.3**(`ext/VERSION`);与 CLI **v1.4.2** 可独立安装。
## v1.4.2 — 2026-10-16
### Highlights
**Agent/LLM 优化 O1–O5**:`ext/ai/sandbox` CoW 试跑接入 `agent.trial` / `run sandbox=`(默认关);promote 强制 sandbox proof;步骤路由 `step_route` + EFI preflight + `usage`/`token_cost`;风格 lint 与 `sandbox.report`;技能 `quarantine` 与离线 standing/policy 优化脚本。表优先、少 json 胶水。路线:[ext-ai-optimize.md](doc/roadmap/ext-ai-optimize.md)。
```bash
git checkout v1.4.2 && cargo build --release
cargo build --release -p marqdo_plugin_agent
marqdo run tests/ext/agent-sandbox-optin.mq.md
marqdo run examples/agent-sandbox-okf/index.mq.md
# Ubuntu: sudo apt update && sudo apt install marqdo
marqdo ext add agent # CDN latest → 1.4.2
```
**核心表面 Δ(Core surface Δ)**:**无增删**——19 个核心构造不变。本版为 **Extension / Agent Runtime**(`ext/ai` 沙盒与晋升闸门),不触碰语言标记。
### Added
- **`ext/ai/sandbox`**:CoW 工作区(`open` / `run` / `diff` / `commit` / `discard` / `inspect` / `profile` / `report`);经 `plugins/agent` `sandbox_*` → `host_query`;默认拒 `plugin`/`exec`/`net`。
- **O1**:`agent.trial` · `run`/`step` `sandbox=` · 中文 `试跑`/`沙盒=` · 示例 `examples/agent-sandbox-okf/` · 金样 `agent-sandbox-optin`。
- **O2**:`agent_kb_promote` 默认 sandbox proof;OKF `applies_when`/`fixtures`/`sandbox_profile`/`proof`;harness Skill Lift。
- **O3**:`agent.step_route` · `ext/ai/policies/step-route.mq.md` · `run efi=`(缺工程知识时软放行)· 结果 `usage`/`token_cost`。
- **O4**:`scripts/mq-style-lint.sh` · `sandbox.report` markdown 审阅面。
- **O5**:`agent_kb_quarantine` / `agent.quarantine` · `scripts/optimize-standing-policy.sh`。
### Fixed
### Changed
- CLI / wasm / `ext/VERSION` **同步 1.4.2**。
- Skill:强制表/链接取元、禁同文件混 EN/ZH、点名 sandbox 试跑习惯。
## v1.4.1 — 2026-10-15
### Highlights
**view 叙述标注连线**:Browse / `view output` 中 underline→chip 改为 SVG 正交连线(描述底边中点 → 轨道 → 短竖线 → 值顶边中点);同跨度多帧 chip 共用 annot id;可见 chip 按锚点 L→R;帧切换后重算密度。设计:[view-prose-annot-gaps.md](doc/design/view-prose-annot-gaps.md)。
```bash
git checkout v1.4.1 && cargo build --release
marqdo view public/00-welcome.mq.md --no-open
marqdo view output public -o /tmp/mq-public
# Ubuntu: sudo apt update && sudo apt install marqdo
marqdo ext add web # CDN latest → 1.4.1
```
**核心表面 Δ(Core surface Δ)**:**无增删**——19 个核心构造不变。本版为 **Runtime / Tooling**(view 标注几何与帧 chip 对齐),不触碰语言标记。
### Added
- 设计缺口文档 [view-prose-annot-gaps.md](doc/design/view-prose-annot-gaps.md):叙述 fence 卡、帧感知 chip 密度、L-leader 几何约定。
### Fixed
- view L-leader:id 与 mark 错配、chip 逆序交叉、border 无色变黑碎段、横线穿过 chip 文字;现为 mid→mid SVG path。
- 叙述 fence / 代码卡单层渲染;静态页隐藏 Run;主文函数引用 call chip + leader。
### Changed
- CLI / wasm / `ext/VERSION` **同步 1.4.1**。
## v1.4.0 — 2026-10-15
### Highlights
Generated
+2 -2
View File
@@ -572,7 +572,7 @@ checksum = "0ceec5bc11778974d1bcb055b18002eba7f4b3518b6a0081b3af5f21666da9ad"
[[package]]
name = "marqdo"
version = "1.4.0"
version = "1.4.5"
dependencies = [
"anyhow",
"assert_cmd",
@@ -603,7 +603,7 @@ dependencies = [
[[package]]
name = "marqdo-wasm"
version = "1.4.0"
version = "1.4.5"
dependencies = [
"marqdo",
"serde_json",
+1 -1
View File
@@ -1,6 +1,6 @@
[package]
name = "marqdo"
version = "1.4.0"
version = "1.4.5"
edition = "2021"
license = "Apache-2.0"
description = "Marqdo — Markdown markers as an interpreted programming language"
+10 -5
View File
@@ -149,8 +149,13 @@ marqdo catalog public -o .marqdo
---
## 现状(v1.4.0)
## 现状(v1.4.5)
- **Streamable HTTP MCP + web invoke 鉴权(v1.4.5)**:`agent_mcp_serve transport=http` · Bearer/`X-Api-Key` 注入 · `unauthorized`→401 · openai/agent `json.get` 硬化
- **LLM 传输软失败 + `ext/ai` 表优先(v1.4.4)**:上游不可达 → `status=error`(不 abort);`agent`/`llm` 去 `json.set` 胶水 · [agent-llm-transport-soft-2026-10.md](doc/research/agent-llm-transport-soft-2026-10.md)
- **Ext pack 1.4.3**:修复 `agent.run` cold EFI(`` `self`.preflight `` + miss resolve `llm_free`);已并入 **≥1.4.4** · [agent-run-efi-resolve-bugs-2026-10.md](doc/research/agent-run-efi-resolve-bugs-2026-10.md)
- **Agent/LLM 优化 O1–O5(v1.4.2)**:`ext/ai/sandbox` CoW · `agent.trial` / `sandbox=` · promote sandbox proof · `step_route` / EFI / cost · style lint · quarantine · [ext-ai-optimize.md](doc/roadmap/ext-ai-optimize.md)
- **view 标注连线(v1.4.1)**:underline→chip SVG mid→mid(底边中点→轨道→短竖线→顶边中点)· 帧 chip 共用 id · L→R 密度 · [view-prose-annot-gaps.md](doc/design/view-prose-annot-gaps.md)
- **叙述统一 / 公式 meta / 图片自写回(v1.4.0)**:`% marqdo: var=` 块公式绑定 · 臂/循环体叙述与 foreach 中英糖 · `![marqdo-out](…)` 图片自写回 · view 推断形参实值 · [fence-meta-and-control-prose.md](doc/design/fence-meta-and-control-prose.md) · [stdlib-writeback.md](doc/design/stdlib-writeback.md)
- **view 文档面(v1.3.3)**:Browse Structure 叙述保真 · 标记面 · 分支/循环帧下拉 · AI `#mq-view-trace` · 静态 `view output` / gh-pages 同源 · [view-document-annotations.md](doc/design/view-document-annotations.md)
- **Engineering Knowledge(v1.3.0)**:EKC L0–L4 · `marqdo knowledge|find|reuse|duplicate|verify|…` · Agent `preflight` / REUSE→ADAPT→CREATE · [engineering-knowledge.md](doc/design/engineering-knowledge.md)
@@ -169,7 +174,7 @@ marqdo catalog public -o .marqdo
- **`web`**:**Artifact 模型(ADR [0007](doc/adr/0007-web-document-endpoint.md))** — `type: web|endpoint` Document/Endpoint;facade `page`/`route`/`serve`/`render`/`inspect`/`use`;Resource 在 `ext/data` · `ext/security` · `ext/net`;设计 [ext-web-artifact.md](doc/design/ext-web-artifact.md);原生 **Go `libweb`**([ADR 0004](doc/adr/0004-web-plugin-go.md));示例 [marqdo-blog](examples/marqdo-blog/) · [web-client-site](examples/web-client-site/) · [anlian-mq](examples/anlian-mq/)
- **`quantum`**:Q0–Q7 + Q8a/b 主题 SVG;示例 [quantum-entanglement](examples/quantum-entanglement/)
- **`linalg`**:L0–L6 + 公式文档面(中缀 / `declare` / 展示化简);示例 [linalg-transpose](examples/linalg-transpose/) · [linalg-svd](examples/linalg-svd/) · [linalg-least-squares](examples/linalg-least-squares/)
- **`agent`**:A1–A4 + **MCP Server/Client stdio** + Skill Compilation + Adaptive Routing + resume + **Engineering preflight**(`agent.preflight` / `eng_preflight`);宿主缺口见 [ext-hosting.md](doc/roadmap/ext-hosting.md) · [agent-framework-gaps-after-a4.md](doc/research/agent-framework-gaps-after-a4.md) · [engineering-knowledge.md](doc/design/engineering-knowledge.md)
- **`agent`**:A1–A4 + **MCP Server/Client**(stdio + **Streamable HTTP**)+ Skill Compilation + Adaptive Routing + resume + **Engineering preflight**(`agent.preflight` / `eng_preflight`);宿主缺口见 [ext-hosting.md](doc/roadmap/ext-hosting.md) · [agent-framework-gaps-after-a4.md](doc/research/agent-framework-gaps-after-a4.md) · [engineering-knowledge.md](doc/design/engineering-knowledge.md)
- **`llm`**:语义 `ask`/`stream`/`collect`(OpenAI/Ollama;可选原生 `plugins/llm`)
- 安装:`marqdo ext list` / `add …` / `remove`([ext-cli.md](doc/design/ext-cli.md);默认 `~/.marqdo/ext`)。原生插件先 `cargo build -p marqdo_plugin_*` 再 `ext add`
- **原生插件 ABI**:[`include/marqdo_abi.h`](include/marqdo_abi.h) · [ext-abi.md](doc/design/ext-abi.md);`plugins/{demo,agent,web,quantum,linalg,llm}`(`web` 为 Go;其余为 Rust)
@@ -178,7 +183,7 @@ marqdo catalog public -o .marqdo
- **浏览器 Marqdo(WASM)**:`marqdo wasm build` + 官方 bridge 自启(作者零业务 JS;桥内可含列表/路由/storage/ws/文件/Canvas/音频/Observer/拖放);`lib/browser` + GFM;`web.client_embed`;示例 [browser-hello](examples/browser-hello/) · [browser-app](examples/browser-app/) · [browser-media](examples/browser-media/) · [web-client-site](examples/web-client-site/)([ADR 0002](doc/adr/0002-browser-marqdo-wasm.md) · [D](doc/roadmap/browser-wasm-d.md) · [E](doc/roadmap/browser-wasm-e.md) · [F](doc/roadmap/browser-wasm-f.md))
- 选型:[ADR 0001 — Rust](doc/adr/0001-implementation-language.md) · [ADR 0002 — 浏览器 WASM](doc/adr/0002-browser-marqdo-wasm.md)(C0–C5 完结,见 [roadmap/browser-wasm.md](doc/roadmap/browser-wasm.md))· [ADR 0003 — 异步效应](doc/adr/0003-browser-async-effects.md)
### 如何使用最新 Marqdo(v1.4.0)
### 如何使用最新 Marqdo(v1.4.5)
**安装解释器(任选其一)**
@@ -193,7 +198,7 @@ marqdo catalog public -o .marqdo
```bash
# 1) 源码安装(跟 tag 或 main)
git clone https://github.com/cflmy/marqdo.git && cd marqdo
git checkout v1.4.0 # 或留在 main
git checkout v1.4.5 # 或留在 main
cargo build --release
export PATH="$PWD/target/release:$PATH"
marqdo version
@@ -232,7 +237,7 @@ marqdo wasm build
# → dist/wasm/ … 见 examples/browser-hello/
# 6) Releases:Windows exe/zip/vsix;Linux CLI zip + native `.so` zip
# https://github.com/cflmy/marqdo/releases/tag/v1.4.0
# https://github.com/cflmy/marqdo/releases/tag/v1.4.5
```
开发期也可用 `cargo run -- …` 代替已安装的 `marqdo`:
+1 -1
View File
@@ -1,6 +1,6 @@
[package]
name = "marqdo-wasm"
version = "1.4.0"
version = "1.4.5"
edition = "2021"
license = "Apache-2.0"
description = "Marqdo interpreter as WebAssembly (browser host bridge)"
+29
View File
@@ -1,3 +1,32 @@
marqdo (1.4.5-1ppa1) noble; urgency=medium
* Marqdo 1.4.5: Streamable HTTP MCP (agent_mcp_serve transport=http);
web invoke Bearer/X-Api-Key inject + unauthorized→401; openai/agent
field access via json.get (no footnote hard-fail).
-- cflmy <pingan@cflmy.cn> Fri, 16 Oct 2026 17:40:00 +0800
marqdo (1.4.4-1ppa1) noble; urgency=medium
* Marqdo 1.4.4: Soft LLM transport errors (structured status=error);
ext/ai table-first refactor (no json.set glue).
-- cflmy <pingan@cflmy.cn> Fri, 16 Oct 2026 14:40:00 +0800
marqdo (1.4.2-1ppa1) noble; urgency=medium
* Marqdo 1.4.2: Agent/LLM optimize O1–O5 — sandbox trial opt-in,
promote sandbox proof, step_route/EFI/cost, style lint, quarantine.
-- cflmy <pingan@cflmy.cn> Fri, 16 Oct 2026 11:51:05 +0800
marqdo (1.4.1-1ppa1) noble; urgency=medium
* Marqdo 1.4.1: View annot leaders — SVG mid-to-mid (mark bottom-center
to chip top-center), unify frame chip ids, L-to-R chip density.
-- cflmy <pingan@cflmy.cn> Thu, 15 Oct 2026 20:40:00 +0800
marqdo (1.4.0-1ppa1) noble; urgency=medium
* Marqdo 1.4.0: Prose-unified control flow — block formula % marqdo: var=,
+7 -1
View File
@@ -90,10 +90,16 @@
| [research/agent-frameworks-and-marqdo.md](research/agent-frameworks-and-marqdo.md) | 主流智能体框架 vs Marqdo 优势 |
| [research/agent-framework-gaps-after-a4.md](research/agent-framework-gaps-after-a4.md) | **`ext/agent` A0–A4 之后缺口盘点** |
| [research/agent-framework-2026-09.md](research/agent-framework-2026-09.md) | **2026-09 刷新:宪法优先 Wave B** |
| [research/ai-agent-sandbox.md](research/ai-agent-sandbox.md) | **AI 扩展沙盒调研**:Wasm / OS / CoW / symlink 对比与 Marqdo 分层推荐 |
| [design/ext-ai-sandbox.md](design/ext-ai-sandbox.md) | **`ext/ai/sandbox` 设计锁定**:CoW · 能力表 · agent ABI(非 stdlib) |
| [roadmap/ext-ai-sandbox.md](roadmap/ext-ai-sandbox.md) | **`ext/ai/sandbox` 实现里程碑** S0–S4 |
| [research/ai-agent-llm-optimize-2026-10.md](research/ai-agent-llm-optimize-2026-10.md) | **2026-10 调研**:代码即文档×OKF×沙盒下的 Agent/LLM 优化与问题分析 |
| [design/ext-ai-optimize.md](design/ext-ai-optimize.md) | **Accepted**:Agent/LLM 优化 O1–O5 locked |
| [roadmap/ext-ai-optimize.md](roadmap/ext-ai-optimize.md) | **O1–O5 实现路线**(沙盒接入→证明晋升→路由/成本→lint→隔离) |
| [research/ext-hosting-gaps.md](research/ext-hosting-gaps.md) | **宿主集成缺口**:同域 SSE 中继 · HTTP→`##` · MCP Server · 副端口 · 跨模块 db/corpus |
| [design/ext-hosting-fill.md](design/ext-hosting-fill.md) | **宿主缺口补齐设计**(`ext/web` proxy/invoke · `ext/ai` mcp_server) |
| [roadmap/ext-hosting.md](roadmap/ext-hosting.md) | **宿主集成 H0–H4**(规划) |
| [roadmap/ext-agent-optimize.md](roadmap/ext-agent-optimize.md) | **`ext/agent` A0–A4 done · Wave B Active** |
| [roadmap/ext-agent-optimize.md](roadmap/ext-agent-optimize.md) | **`ext/agent` A0–A4 done · Wave B Active** · 后续优化见 [ext-ai-optimize](roadmap/ext-ai-optimize.md) O 波次 |
金样例与集成测试:[`tests/`](../tests/)。面向访客的可执行文档:[`public/`](../public/)。Spike 存档:[`spike/`](../spike/)。
+142
View File
@@ -0,0 +1,142 @@
# 方案:智能体 / 大模型优化方向(代码即文档 · OKF · 沙盒)
| | |
|---|---|
| 状态 | **Accepted · O1–O5 locked** |
| 日期 | 2026-10-16 |
| 调研 | [ai-agent-llm-optimize-2026-10.md](../research/ai-agent-llm-optimize-2026-10.md) |
| 路线 | [ext-ai-optimize.md](../roadmap/ext-ai-optimize.md) |
| 相关 | [ext-ai-sandbox.md](ext-ai-sandbox.md) · [ext-agent.md](ext-agent.md) · [okf.md](okf.md) · [engineering-knowledge.md](engineering-knowledge.md) · [ext-agent-optimize.md](../roadmap/ext-agent-optimize.md) · [ai-stack.md](../roadmap/ai-stack.md) |
---
## 0. 方案立场
1. **强化差异化**:可执行 `.mq.md` 为权威;OKF 为复用;沙盒为证明;LLM 为探索。
2. **不向隐藏 TOOL: 循环 / 向量权威回退。**
3. 下列方向均为**可行**;推荐波次见 §3,最终裁剪由你选型。
4. **不改** `doc/next/*`;落地时另开 roadmap 切片。
---
## 1. 架构愿景(闭环)
```text
目标 / 任务
│
▼
EFI / OKF lookup ──hit──► llm_free / marqdo 执行 ──► writeback
│ miss
▼
small-llm 路由 / 父分解
│
▼
生成或修补 .mq.md(代码即文档风格)
│
▼
sandbox.open → sandbox.run(caps)→ diff
│
├─ 失败 → writeback error / 隔离,不 promote
└─ 成功 → 人审或策略 → commit → solidify / OKF promote
│
▼
下次 REUSE(降本)
```
---
## 2. 可行方向全集
每个方向标注:解决的问题(Pn)、依赖、风险、粗粒度成本。
### A. 沙盒 × Agent 闭环
| ID | 方向 | 解决 | 做什么 | 依赖 | 风险 | 成本 |
|----|------|------|--------|------|------|------|
| **A1** | `step`/`plan` 默认经 sandbox 试跑再写实树 | P1 | 生成落 upper;绿才 `commit` | sandbox | 破坏现有金样默认 | 中 |
| **A2** | Promote / solidify 强制 sandbox proof | P1 | OKF 写入前要求 proof fixture 绿 | sandbox+OKF | 过严挡复用 | 中 |
| **A3** | Sandbox profile 表(caps 宣告) | P1/P3 | GFM 表声明 net/plugin/fs | sandbox | 表方言膨胀 | 低 |
| **A4** | Guest 风格门禁(表优先、禁 json 胶水) | P5 | commit 前 lint / preflight | sandbox+parser | 误杀合法 json.parse | 中 |
| **A5** | 会话 diff → view 审阅卡 | P1/P4 | HITL 看 CoW diff | view+sandbox | UI 工时 | 中高 |
| **A6** | 多会话并行 CoW(fan-out 实验) | — | 每假设一 session | sandbox | 隔离/合并复杂 | 高 |
### B. OKF / EKC / 技能层
| ID | 方向 | 解决 | 做什么 | 依赖 | 风险 | 成本 |
|----|------|------|--------|------|------|------|
| **B1** | OKF 技能元数据:适用条件 / fixtures / sandbox_profile | P2 | frontmatter + 目录约定 | okf | 与 catalog 字段冲突 | 中 |
| **B2** | Progressive disclosure(索引轻、正文按需 READ) | P2/P6 | 对齐 Anthropic Skills 层级 | agent A3 | 实现重复 | 中 |
| **B3** | Skill Lift 评测(有/无技能配对) | P4 | harness 指标 | B1 评测 | 样本少噪声大 | 中 |
| **B4** | 失败技能隔离(quarantine)+ 回归用例跟随血缘 | P1 | ForgeAgent 式 | OKF | 状态机复杂 | 中高 |
| **B5** | near-match / soft_match 产品化默认开 | P6 | 已有 roadmap | okf-near | 误复用 | 低中 |
| **B6** | EKC preflight 默认硬门槛(REUSE→ADAPT→CREATE) | P6 | step 前强制 EFI | EKC | 拖慢探索 | 中 |
| **B7** | Capability 包与 `llm_free` 资源双向链接 | P2 | catalog L3↔任务包 | EKC+OKF | 图维护 | 中 |
### C. 大模型 / 路由 / 成本
| ID | 方向 | 解决 | 做什么 | 依赖 | 风险 | 成本 |
|----|------|------|--------|------|------|------|
| **C1** | 步骤级路由(lookup / patch / sandbox_run / net) | P3 | 扩展 route policy 表 | routing | 策略过拟合 | 中 |
| **C2** | 总成本计量(含推理 token 膨胀)写回文档 | P7 | usage→writeback/metrics | llm result | 噪声 | 低中 |
| **C3** | 用强模型「反思」优化 standing / policy `.mq.md`(类 GEPA) | P3/P5 | 离线优化环,产物仍是文件 | llm | 需评测集 | 高 |
| **C4** | 高频 OKF 技能蒸馏/微调小模型(编译进权重) | — | 长期研究 | 数据+算力 | 运维重;偏离「源码权威」 | 很高 |
| **C5** | Named model 绑定任务类型(fast=路由,reasoning=分解) | P3 | 已有 named handles 加深 | llm | 配置漂移 | 低 |
| **C6** | Prompt document 强制:无参 ask 必须来自 type:prompt 文件 | P5 | 硬化 005 | llm | 破坏脚本式调用 | 低 |
### D. 代码即文档(产出质量)
| ID | 方向 | 解决 | 做什么 | 依赖 | 风险 | 成本 |
|----|------|------|--------|------|------|------|
| **D1** | 官方「可读 Marqdo」Skill 强化(表/链接取元/禁混名) | P5 | skills/marqdo 加厚 | skill | 模型仍违规 | 低 |
| **D2** | AST/结构 lint:json.set 链、混 EN/ZH API | P5 | `marqdo lint` 或 sandbox 钩子 | parser | 误报 | 中 |
| **D3** | 工作簿补丁只允许表驱动 / 命名槽写回 | P5 | 收紧 A0 补丁面 | agent | 表达力下降 | 中 |
| **D4** | 示例库:sandbox+OKF+表风格 canonical demos | P4/P5 | examples/ | 文档 | 维护 | 低 |
### E. 评测与产品化
| ID | 方向 | 解决 | 做什么 | 依赖 | 风险 | 成本 |
|----|------|------|--------|------|------|------|
| **E1** | 完成 Wave B1 harness(constitution / writeback / call_site) | P4 | 已规划 | agent | — | 中 |
| **E2** | 指标面板:ReuseRate / ProofRate / TokenCost | P4/P7 | metrics 写回或 catalog | agent | 虚荣指标 | 中 |
| **E3** | Live 金样矩阵(sandbox 开/关 × 路由后端) | P4 | CI optional | CI | flaky | 中 |
| **E4** | 教程:「只读一份 .mq.md 理解 Agent」 | — | doc/tutorial | 文档 | — | 低 |
### F. 明确不优先 / 需单独论证
| ID | 方向 | 理由 |
|----|------|------|
| **F1** | 第二套 Agent Framework(Python/TS 对象图权威) | 违宪 |
| **F2** | 向量库升为记忆权威 | 违宪;可作证据工具 |
| **F3** | 默认通用 shell Agent | 与 Marqdo-only 沙盒战略冲突 |
| **F4** | 整库 `ext/web` 进默认沙盒 | 已否决 |
| **F5** | 人人 microVM 本地默认 | DX 过重;留给 hosted |
---
## 3. 执行顺序(O1→O5 locked)
| 波次 | 方向组合 | 为何先做 |
|------|----------|----------|
| **O1** | A1 轻量(opt-in `sandbox=True`)+ A3 + D1 + D4 | 立刻接上刚落地的沙盒,风险可控 |
| **O2** | A2 + B1 + B3 + E1 | 验证→晋升→可度量 Skill Lift |
| **O3** | C1 + C2 + B6 | 降本与 EFI 硬门槛 |
| **O4** | A4 + D2 + A5 | 质量门禁与人审 |
| **O5** | B4 + C3 | 失败隔离与离线提示优化 |
| **O∞** | C4 / A6 | 研究项,单独立项 |
切片实现见 [ext-ai-optimize.md](../roadmap/ext-ai-optimize.md)。
---
## 4. 验收原则(任一波次)
- 金样:有/无沙盒、有/无 OKF hit、路由后端切换。
- 指标:至少报告 ReuseRate 或 TokenCost 之一相对基线变化。
- 文档:`.mq.md` 示例仍可读为 Markdown(表/链接取元,少 json 胶水)。
- 红线:不改 `doc/next/*`;ext 不调 `host_*`。
---
## 5. 落地
O1–O5 已锁定;实现见 [roadmap/ext-ai-optimize.md](../roadmap/ext-ai-optimize.md)。**仍不改** `doc/next/*`。
+104
View File
@@ -0,0 +1,104 @@
# Official AI sandbox (`ext/ai/sandbox`)
| | |
|---|---|
| Status | **Accepted · implementing** |
| Date | 2026-10-16 |
| Related | [research/ai-agent-sandbox.md](../research/ai-agent-sandbox.md) · [roadmap/ext-ai-sandbox.md](../roadmap/ext-ai-sandbox.md) · [ext-agent.md](ext-agent.md) · [ext-abi.md](ext-abi.md) · [browser-marqdo-wasm.md](browser-marqdo-wasm.md) · [ai-stack.md](../roadmap/ai-stack.md) · [stdlib-writeback.md](stdlib-writeback.md) |
---
## 0. One-liner
**Agent-only Marqdo trial runtime:** capability-isomorphic to browser Wasm defaults, CoW workspace until explicit `commit`, author surface under `ext/ai` (not stdlib).
---
## 1. Goals / non-goals
### Goals
| # | Goal |
|---|------|
| G1 | Open a CoW session over a workspace root; edits stay in upper until `commit` |
| G2 | `run` executes `.mq.md` (or in-memory source) under deny-by-default caps with core `lib/*` usable |
| G3 | `diff` / `discard` / `inspect` for review gates |
| G4 | Stable `unavailable in sandbox` for plugin.load, exec foreign/subtask path|code|lang, net (default), `ext/web` server APIs |
| G5 | Author API in `ext/ai/sandbox` via `plugins/agent` — **no** public `lib/sandbox` |
### Non-goals (this wave)
| # | Non-goal |
|---|----------|
| N1 | Embed wasmtime in CLI |
| N2 | True Wasm ABI plugins (linalg/quantum) |
| N3 | bwrap / microVM |
| N4 | Full `ext/web` listen/DB/WS inside sandbox |
| N5 | Symlink-forest workspace isolation |
| N6 | Default host-proxy of all native ABI plugins |
| N7 | Wire `agent.step`/`plan` through sandbox automatically (later) |
---
## 2. Layering
```text
ext/ai/sandbox.mq.md → plugin.load agent → sandbox_* ABI
↓ host_query
HostContext + CowSession
(fs / writeback see upper)
```
| Layer | May call `host_*` | Role |
|-------|-------------------|------|
| `ext/ai/sandbox` | **No** | Author facade |
| `plugins/agent` | via `host_query` only | Register `sandbox_*`; forward to host |
| `src/host` | — | CoW session store; fs/writeback intercept; nested `run` |
---
## 3. Author API
EN: `ext/ai/sandbox.mq.md` · ZH: `ext/ai/沙盒.mq.md` (separate files, no mixed names in one module).
| Method | Params | Result |
|--------|--------|--------|
| `open` | optional `root=` (default entry dir / cwd) | map `{ id, root }` |
| `write` | `id=`, `path=`, `text=` | none |
| `read` | `id=`, `path=` | text |
| `remove` | `id=`, `path=` | none (whiteout) |
| `diff` | `id=` | list of `{ path, kind: write\|remove }` |
| `commit` | `id=` | `{ ok, applied }` |
| `discard` | `id=` | `{ ok }` |
| `run` | `id=`, `path=` **or** `source=` | `{ ok, value, stdout?, error? }` |
| `inspect` | `id=` | caps, root, change count, notes |
Default `run` caps: `fs_write=true` (CoW only), `exec=false`, `net=false`, `plugin=false`.
---
## 4. CoW semantics
- **lower** = real workspace root (read-through).
- **upper** = session map path → bytes.
- **whiteout** = logical deletes; reads miss; commit unlinks lower.
- No symlink-as-isolation; resolve paths under root; reject `..` escape.
- Until `commit`, host tree unchanged (verifiable with `git status` / file mtime).
---
## 5. Hard bans
1. No public `lib/sandbox`.
2. No symlink forest as the isolation story.
3. No silent success when capability missing — diagnose `unavailable in sandbox`.
4. `ext/**` must not call `host_*`.
5. Do not treat empty `# main` as sandbox completeness.
---
## 6. ABI / host_query
Plugin exports (registered on agent plugin): `sandbox_open`, `sandbox_write`, `sandbox_read`, `sandbox_remove`, `sandbox_diff`, `sandbox_commit`, `sandbox_discard`, `sandbox_run`, `sandbox_inspect`.
Implementation: each forwards to allowlisted `host_query` names of the same stem (e.g. `sandbox_open`), which mutate the process-local CoW store and (for `run`) invoke nested capture with a sandboxed `HostContext`.
+2 -2
View File
@@ -120,8 +120,8 @@ import data:ext/data/db.mq.md
| `path` | `路径` | URL |
| `method` | `方法` | GET/POST/… |
| `request` | `请求` | `json` / `form` / `none` |
| `response` | `响应` | `json` / `text` / `html` |
| `auth` | `鉴权` | 角色名或 `none` |
| `response` | `响应` | `json` / `text` / `html` / `sse`(`text/event-stream`) |
| `auth` | `鉴权` | 角色名或 `none`;`bearer` / `api_key` = 不要求 cookie,把 `Authorization` / `X-Api-Key` 注入 payload 由 `# main` 校验 |
正文在请求时执行;返回值按 `response` 序列化。Result 协议见 §8。
+18 -2
View File
@@ -314,10 +314,15 @@ type: Marqdo Agent Skill
title: <短名>
sig: <12hex>
resource: ../../resources/<slug>.mq.md
status: stable | candidate
status: stable | candidate | quarantine
llm_free: true | false
quality: <int>
hits: <int>
# O2 optional metadata (from workbook frontmatter; see §7.5):
applies_when: <text condition or tag>
fixtures: <relative .mq.md proof path>
sandbox_profile: default | readonly | <name>
proof: <relative .mq.md> # alias of fixtures for sandbox gate
generated:
by: marqdo-agent/plan
at: <ISO8601>
@@ -331,7 +336,7 @@ verified:
Prefer spawning `resource` over re-planning.
```
`status`:`llm_free` 的 resource 晋升为 **stable**(零父 LLM 快路径);否则为 **candidate**(可复用,但按 `improve_every` 偶发再优化,每轮最多 1 次改进)。失败运行不晋升;质量更差不覆盖。
`status`:`llm_free` 的 resource 晋升为 **stable**(零父 LLM 快路径);否则为 **candidate**(可复用,但按 `improve_every` 偶发再优化,每轮最多 1 次改进)。失败运行不晋升;质量更差不覆盖。**O2:** `agent_kb_promote` 默认要求 sandbox proof(`proof` / `fixtures` / workbook 本身经 `sandbox.run` 成功)才写入;失败返回 `promoted: false` / `reason: sandbox_proof_failed`。传 `require_proof=false` 仅用于明确豁免。
### 7.3 任务签名 → Concept 路径
@@ -401,6 +406,17 @@ Prefer spawning `resource` over re-planning.
| 迭代 | 同 Concept slug/`sig` 覆盖 resource + 刷新 `generated`/`verified` |
| 成长 | 命中快路径;`candidate`/`hits`/`improve_every` 驱动再优化;`log.md` 可审计 |
| 分发 | 整包 `agent-kb/` 可复制(格式即契约) |
| **证明晋升(O2)** | `applies_when` / `fixtures` / `sandbox_profile` / `proof` 写在技能与 workbook frontmatter;promote 前 `sandbox.run` 绿才 `llm_free` |
| **隔离(O5)** | 失败技能 `status: quarantine`;回归 fixture 跟随 slug |
### 7.5a 技能 frontmatter 约定(O2 B1)
| 字段 | 含义 |
|------|------|
| `applies_when` | 适用条件(短文本 / 标签);路由与 harness 可读 |
| `fixtures` | 相对路径:sandbox 证明用 `.mq.md` |
| `sandbox_profile` | `default` / `readonly` / 其它 `sandbox.profile` 名 |
| `proof` | 与 `fixtures` 同义;promote 优先读 `proof` 再 `fixtures`,否则跑 workbook 自身 |
---
+1 -1
View File
@@ -4,7 +4,7 @@
|---|---|
| 状态 | **Active** |
| 日期 | 2026-10-15 |
| 相关 | [view.md](view.md) · [view-debug.md](view-debug.md) · [view-markup-surface.md](view-markup-surface.md) · [view-execution-trace.md](view-execution-trace.md) · [view-frame-selection.md](view-frame-selection.md) · [code-as-docs-gaps.md](code-as-docs-gaps.md) |
| 相关 | [view.md](view.md) · [view-debug.md](view-debug.md) · [view-markup-surface.md](view-markup-surface.md) · [view-execution-trace.md](view-execution-trace.md) · [view-frame-selection.md](view-frame-selection.md) · [view-prose-annot-gaps.md](view-prose-annot-gaps.md) · [code-as-docs-gaps.md](code-as-docs-gaps.md) |
---
+145
View File
@@ -0,0 +1,145 @@
# View:叙述面标注缺口(prose annot gaps)
| | |
|---|---|
| 状态 | **Active** |
| 日期 | 2026-10-15 |
| 相关 | [view-document-annotations.md](view-document-annotations.md) · [view-static-interactive.md](view-static-interactive.md) · [view-frame-selection.md](view-frame-selection.md) · [view-execution-trace.md](view-execution-trace.md) |
> **定位**:对既有文档面设计的**缺口专章**——写清用户可见的四类问题、根因(文件:行)与验收口径。不改写 ADR;不发明作者侧标注 DSL。
---
## 1. 为何「改了却几乎没变化」
| 原因 | 说明 |
|------|------|
| 错二进制 | PATH 上的 `marqdo` 可能仍是旧版(如 1.3.0);改动在源码 / `./target/debug/marqdo`。验收必须用 `cargo build && ./target/debug/marqdo view …` |
| Chip 被溢出藏掉 | 同 band 堆多帧 return/print/who;`MAX_VISIBLE_ANNOTS=4` 把当前帧的 `who` 打成 `annot-chip-overflow`(`display:none`);帧切换不重算溢出 |
| L-leader 几何错位 | `initAnnotLeaders` 里 `y0=0` 只在 `.annot-band` 顶画短折,**连不到**上方 underline;窄屏或 chip>4 时整段关掉 leaders |
| 旧静态快照 | gh-pages / 旧 `view output` 不会自动刷新 |
---
## 2. 目标视觉语法(既有,不新造)
锚定 [view-document-annotations.md](view-document-annotations.md) §3:
```
文本段…… `变量` …… **语句** …… *返回*
━━━━━
│
└──→ 结果标注 chip
```
- underline:`mark.mq-anchor.mq-{decl|code|return}`
- band:段落下方 `.annot-band` → leaders + chips
- L-leader:CSS `border-left` + `border-bottom`(禁止 SVG)
- 公式/代码:[view-static-interactive.md](view-static-interactive.md) G3 **一层主视觉**;卡下不叠 Bind dump
- 帧:[view-frame-selection.md](view-frame-selection.md) **源码常显、只滤 chip**;默认末帧
---
## 3. 管线
```mermaid
flowchart LR
CLI["marqdo view / view output"] --> build["build_file_view"]
build --> cap["run_file_capture"]
cap --> annot["build_view_annots"]
annot --> doc["document::render_module_document"]
doc --> page["html.rs page_file"]
page --> js["initAnnotLeaders + initCallPickers"]
```
| 步骤 | 位置 |
|------|------|
| CLI / 导出 | `src/main.rs` · `src/view/mod.rs` · `src/view/output.rs` |
| Capture + Call 事件 | `src/interp.rs` |
| Annot 组装 | `src/view/annot.rs` `build_view_annots` |
| 文档面 HTML | `src/view/document.rs` |
| 卡片 chrome | `src/view/render.rs` `render_*_assign_card` |
| CSS / leader JS | `src/view/html.rs` |
---
## 4. 四条用户问题 → 根因 → 验收
复现源:`public/00-welcome.mq.md` · `public/00-欢迎.mq.md`
对照金样:`tests/view/document-annot-demo.mq.md` · `tests/markup-v03/inc-prose.mq.md`
### P1 — 叙述与代码同段时信息不显示
| | |
|---|---|
| 现象 | underline 在,chip / 连线弱或无 |
| 根因 | `html.rs` `initAnnotLeaders`:`y0=0` 相对 band;`document.rs` `MAX_VISIBLE_ANNOTS` 把当前帧 chip 溢出;PATH 旧 CLI |
| 目标 | 同 `inc-prose`:underline + L-leader(从锚点底到 chip)+ 可见 chip |
| 验收 | `./target/debug/marqdo view public/00-welcome.mq.md`:主文 `**greet…**` 与 `## greet` 体内均有可见 chip 与接到 underline 的 leader(宽屏) |
### P2 — 函数内变量不显示
| | |
|---|---|
| 现象 | `` `who` `` 有浅蓝 underline,无值 chip |
| 根因 | bind chip 存在但带 `annot-chip-overflow`;`applyCallFrame` 只 `hidden` 非选中帧,不解除当前帧 overflow |
| 目标 | 选中调用帧时常显 `who` · `print` · `return` 各一枚 |
| 验收 | 切换「调用」下拉,三枚 chip 随帧变且默认可见(无需点 +N) |
### P3 — ` ```text ` 渲染不一致
| | |
|---|---|
| 现象 | 变成内联 `text marqdo ext add…`,无卡片 |
| 根因 | Comment 围栏若进 `flush_para` → `join(" ")` → pulldown 当 inline code;应走 `render_narrative_fence` |
| 目标 | 与 bound python 同 `.code-card` chrome(badge + `pre.code-body`),**无** Run/foot |
| 验收 | welcome install 段:`data-lang="text"` + 两行 `marqdo ext add`;无 `<code>text marqdo` |
### P4 — 函数引用 / 代码卡未渲染
| | |
|---|---|
| 现象 | `**greet who=…**` 无结果;或 `hi` python 只见头/Run、不见源码 |
| 根因 | 同行多 Call 未与 BoldCode zip;或 Run 挤在 head;静态假 Run;错看 Debug 路径 |
| 目标 | 主文两 greet 各有 call chip(stdout 摘要);`hi` = `render_code_assign_card` head→body→foot;静态藏 Run |
| 验收 | 两枚 greet mark 均有非空 `data-annot-id`;`code-body` 含 `hello-from-python`;`MARQDO_VIEW_LIVE=false` 无 Run 按钮 |
### call_id 语义(写死)
| 位置 | call_id |
|------|---------|
| 主文 `**greet…**` 的 call chip | 入口帧(`# main` 的 call);结果为 stdout/返回摘要 |
| `## greet` 体内 print / who / return | callee 的 CallFrame id;跟函数标题旁 picker |
---
## 5. 修复约定(实现红线)
1. **帧感知密度**:band 内按选中 `call_id` 计可见 chip;默认末帧;同帧最多 4 枚常显;溢出只藏非当前帧或多余 kind。切换后客户端重算 overflow + `layoutAll()`。
2. **Leader 容器**:提升到 `.doc-block`(prose + band);`y0` = 锚点底相对 block;按锚点 x 排序再画线。
3. **卡片**:只复用 Browse `render_*_assign_card` / `render_narrative_fence` + `html.rs` `.code-*` / `.formula-card`;本波不改 Debug AST 路径(除非回归红)。
4. **静态 Run**:`MARQDO_VIEW_LIVE=false` 时隐藏 Run,不留禁用假按钮(对齐 view-static-interactive D5)。
5. **非目标**:不做 debug 断点;不发明作者标注 DSL;不用 SVG leader;不单独发版/推 gh-pages(除非另嘱)。
---
## 6. 验收命令
```bash
cargo build -q
./target/debug/marqdo version # 须为当前工作树版本
./target/debug/marqdo view public/00-welcome.mq.md --no-open
./target/debug/marqdo view tests/view/document-annot-demo.mq.md --no-open
cargo test -q --lib view::
```
---
## 7. 修订历史
| 日期 | 说明 |
|------|------|
| 2026-10-15 | 初稿:四问题、错二进制/溢出/leader 根因、帧感知密度与 leader 几何约定 |
| 2026-10-15 | L-leader 拥挤/消失:同跨度多帧 chip **共用 annot id**;可见 chip 按锚点 x 重排;leader 挂 doc-block |
| 2026-10-15 | L-leader 黑线碎段:JS `border*: "1.5px solid"` 无色覆盖 class → 改为 kind 色写入 shorthand;chip SSR/DOM 按锚点 L→R |
| 2026-10-15 | L-leader 改为 SVG:描述底边中点 → 轨道 → 短竖线 → 值 chip 顶边中点;chip 间距加宽 |
@@ -0,0 +1,58 @@
# 缺陷记录:LLM 传输失败硬中断 Agent Runtime
| | |
|---|---|
| 状态 | **Fixed(2026-10-16)** — 见 §3;金样 `tests/ext/agent-llm-transport-soft.mq.md` |
| 日期 | 2026-10-16 |
| 发现方 | qdagent(接通 `agent.run` live 路径时) |
| 影响面 | `host/net` · `plugins/llm` · `ext/ai/llm` · `ext/ai/llm/openai` · `ext/ai/agent` |
| 相关 | [agent-run-efi-resolve-bugs-2026-10.md](agent-run-efi-resolve-bugs-2026-10.md) · [ext-agent.md](../design/ext-agent.md) |
---
## 0. 一句话
上游 LLM **连接拒绝 / 超时 / 非 200** 时,Runtime 曾以 **硬错误** 终止整个 `marqdo run`;产品(qdagent ASK/BUILD)无法落到结构化 `status=error`。
---
## 1. 复现
```bash
OPENAI_API_KEY=offline OPENAI_BASE_URL=http://127.0.0.1:9 \
marqdo run tests/ext/agent-llm-transport-soft.mq.md
# 修复前:exit≠0,error: http POST …
# 修复后:exit 0,ask/step/run 均带 finish|status=error + error 文本
```
---
## 2. 根因
| 层 | 行为 |
|----|------|
| `host_http_post` / SSE | 传输失败 `Err(...)` → VM abort |
| `plugins/llm` `llm_chat_completions` | 同上 `Err` |
| `ext/ai/llm/openai` | HTTP≠200 → `sys.exit 1` |
| `agent.model_turn` / `step` / `plan` | 无软失败臂 |
---
## 3. 修复清单
| # | 改动 | 文件 |
|---|------|------|
| 1 | 传输失败 → `{status:0, body}` / SSE error event | `src/host/net.rs` |
| 2 | 插件返回 `{finish:error, error, …}` 而非 `Err` | `plugins/llm/src/lib.rs` |
| 3 | openai 非 200 → 结构化 pack,禁止 `sys.exit` | `ext/ai/llm/openai.mq.md` |
| 4 | `llm.ask` 透传 `error` / `finish` | `ext/ai/llm.mq.md` |
| 5 | `model_turn` / `step` / `plan` 收成 `status=error` | `ext/ai/agent.mq.md` |
| 6 | 金样 | `tests/ext/agent-llm-transport-soft.mq.md` |
**契约**:Agent Runtime 对 LLM 传输失败必须可捕获;`status=error` + `error` 字符串;进程退出码 0(程序正常返回错误图)。
---
## 4. 对 qdagent
`ensure_llm` 有 key 但上游不可达时,`ask`/`build` 可走既有 `mode=error` 臂写 workbook,不再整 CLI 崩。
@@ -0,0 +1,117 @@
# 缺陷记录:`agent.run` EFI / resolve 在消费方崩溃(qdagent 复现)
| | |
|---|---|
| 状态 | **Fixed(2026-10-16)** — 见 §3;金样 `tests/ext/agent-run-cold-efi.mq.md` |
| 日期 | 2026-10-16 |
| 发现方 | qdagent(`~/work/qdagent`)接通 Marqdo ≥1.4.2 Agent EFI 全路径时 |
| 影响面 | `ext/ai/agent` · `## run` · `plugins/agent` `agent_resolve` |
| 相关 | [ext-agent.md](../design/ext-agent.md) · [engineering-knowledge.md](../design/engineering-knowledge.md) · [ai-agent-llm-optimize-2026-10.md](ai-agent-llm-optimize-2026-10.md) |
---
## 0. 一句话
`agent.run` 在 **空 OKF / 无 skill 命中** 或 **`efi=True`** 时,会以硬错误退出,而不是降级到 `plan` 探索。
这是 **Agent Runtime 契约不完整**,不是 qdagent 业务写错。
---
## 1. 复现(qdagent)
环境:工作区 Marqdo `1.4.2` + `MARQDO_EXT=~/work/marqdo/ext`;qdagent 调用:
```text
助手 = > agent.agent …
`助手`.run task=… efi=True # 或 efi=False 后仍进 resolve
```
本仓几乎没有 `.marqdo/agent-kb` 编译 skill。
### Bug A — `efi=True` → `unknown function preflight`
**位置**:[`ext/ai/agent.mq.md`](../../ext/ai/agent.mq.md) `## run`
```markdown
1. `efi`
**efi_pf = preflight task=`task`**
```
同文件其它同对象方法一律写 `` `self`.route `` / `` `self`.resolve ``。
裸标识 `preflight` 在导入消费方(如 `cli/求道-构建.mq.md`)里被解析成**全局函数**,找不到 → 硬错。
**期望**:`` **efi_pf = > `self`.preflight task=`task`** ``(或等价 method call)。
### Bug B — miss resolve → `missing map key llm_free`
**位置**:
1. Plugin:[`plugins/agent/src/memory.rs`](../../plugins/agent/src/memory.rs) `resolve_task`
当 `kb_lookup` 返回 null:
```rust
Ok(json!({
"mode": "unknown",
"skill": null,
"resource": null,
"confidence": 0.0,
// 缺 llm_free
}))
```
2. Facade:`## run` 在 resolve 之后:
```markdown
**llm_free = [llm_free](resolved)**
```
脚注式取键在缺键时 **硬失败**,无法落到后面的 `plan` 探索臂。
**期望**:
- miss 结果始终带 `"llm_free": false`(与命中路径字段集对齐)
- facade 侧对可选键用 `json.get`(缺省 `False`),防御其它调用方
---
## 2. 为何金样没拦住
| 金样 | 为何漏网 |
|------|----------|
| `tests/ext/agent-zero-llm.mq.md` | 先编译 skill / policy,再 `run`;resolve **命中** |
| `tests/ext/eng-preflight.mq.md` | 只测 `agent_eng_preflight`,**不**走 `agent.run efi=` |
| 多数 `agent-*-live` | 自带 kb 或 `efi=False` / 不经过 cold miss |
qdagent 是第一个「冷库 + 产品 EFI + `run`」消费方,因此打穿。
---
## 3. 修复清单(本轮)
| # | 改动 | 文件 |
|---|------|------|
| 1 | miss resolve 补 `llm_free: false` | `plugins/agent/src/memory.rs` |
| 2 | `` `self`.preflight ``;`llm_free`/`mode` 用 `json.get` | `ext/ai/agent.mq.md` |
| 3 | 回归金样:冷 kb + `run efi=True` 不崩溃,落到 explore/plan | `tests/ext/agent-run-cold-efi.mq.md` |
| 4 | 重建 `marqdo_plugin_agent` 并跑金样 | `cargo build -p marqdo_plugin_agent --release` |
---
## 4. 对 qdagent 的含义
修复后 qdagent 可恢复:
```text
product preflight → latch → agent.run (efi=True|False) / plan
```
而不必长期绕开 `agent.run`。
在插件未装入新 so 前,qdagent 侧 EFI latch + `step`/`plan` 仍是合法降级。
---
## 5. 非目标
- 不改 EKC / `agent_eng_preflight` 语义
- 不引入第二套 Agent loop
- 不把向量检索升为权威真相
@@ -0,0 +1,136 @@
# 调研:代码即文档 × OKF × 沙盒 下的智能体 / 大模型优化(2026-10)
| | |
|---|---|
| 状态 | **调研 + 问题分析 · 供方案 [ext-ai-optimize.md](../design/ext-ai-optimize.md) 参考** |
| 日期 | 2026-10-16 |
| 前置 | [agent-frameworks-and-marqdo.md](agent-frameworks-and-marqdo.md) · [agent-framework-gaps-after-a4.md](agent-framework-gaps-after-a4.md) · [agent-framework-2026-09.md](agent-framework-2026-09.md) · [ai-agent-sandbox.md](ai-agent-sandbox.md) · [okf-and-marqdo.md](okf-and-marqdo.md) |
| 本仓能力 | [ext-agent.md](../design/ext-agent.md) · [okf.md](../design/okf.md) · [engineering-knowledge.md](../design/engineering-knowledge.md) · [ext-ai-sandbox.md](../design/ext-ai-sandbox.md) · [ai-stack.md](../roadmap/ai-stack.md) · [ext-agent-optimize.md](../roadmap/ext-agent-optimize.md) |
| 外部抽样 | Anthropic Agent Skills · SoK Agentic Skills (arXiv 2602.20867) · ForgeAgent · ACES / Skill Lift · Switchcraft 路由 · DSPy/GEPA ·「工作流编译进权重」· MalSkillBench |
---
## 0. 一句话
业界正把 Agent 优化从「更好的 chat loop」转向 **可验证、可复用、可隔离执行的技能层**;Marqdo 已具备 **可执行 Markdown(代码即文档)+ OKF/EKC 复用 + CoW 沙盒** 三件套,优化目标应是把这三件套焊成闭环:**试跑 → 证据 → 晋升 → 零/少 LLM 复用**,而不是追平 LangChain 功能清单。
---
## 1. 调研方法
| 动作 | 说明 |
|------|------|
| 扫业界技能层 | Anthropic Skills(渐进披露 `SKILL.md`);SoK Agentic Skills 生命周期;ForgeAgent「能力铸造」;ACES Skill Lift |
| 扫成本/路由 | Switchcraft 工具调用路由;DSPy+GEPA 提示优化;工作流编译进小模型权重 |
| 扫安全 | MalSkillBench 运行时验证恶意技能;沙盒粒度讨论(per-skill / per-session) |
| 对照本仓 | A0–A4 / Wave B 已落地;沙盒 S0–S4 已落地;OKF + EKC + adaptive routing 现状 |
**不修改** `doc/next/*`(手写真相源);本文件只进 `doc/research/`。
---
## 2. 业界态势(与 Marqdo 相关的收敛点)
### 2.1 技能层取代「纯工具调用叙事」
| 来源 | 要点 | 对 Marqdo |
|------|------|-----------|
| Anthropic Agent Skills | 目录 + `SKILL.md`;元数据预载、正文按需;脚本可执行 | 形态近 OKF 任务包,但 Marqdo 技能 **本身可跑** |
| SoK Agentic Skills | 发现→练习→蒸馏→存储→组合→评测→更新;代码技能需沙盒 | 生命周期与 OKF promote / sandbox.run 对齐 |
| ForgeAgent | 生成能力 → 沙盒证明 → 信任复用;失败进回归隔离 | **验证后晋升** 正是 solidify / `llm_free` 缺的硬闸 |
| ACES | 有/无技能配对试验 → Skill Lift | 可作 Marqdo 评测 harness 指标 |
### 2.2 成本:路由 + 编译 + 小模型
| 来源 | 要点 | 对 Marqdo |
|------|------|-----------|
| Switchcraft | 工具调用场景路由到最便宜仍正确的模型;报道大幅降本 | 已有 `marqdo \| small-llm \| llm \| auto`;可加 **工具步/试跑步** 细分 |
| DSPy / GEPA | 用反馈改写指令,小模型追上大模型 | 优化对象应是 **`.mq.md` 提示文档 / standing / 路由策略文件**,不是黑盒字符串 |
| 工作流编译进权重 | 程序化流程进小模型权重,上下文常数级 | 长期对应「高频 OKF 技能 → 专用小模型」;短期优先 **llm_free 源码路径** |
### 2.3 安全:技能不可信 + 沙盒必验
恶意技能基准与 SoK 均强调:**技能说明与代码默认不可信**,须运行时隔离。Marqdo 新沙盒(默认拒 plugin/net/exec)正好是技能晋升前的证明场。
---
## 3. Marqdo 已有资产(优化起点)
| 资产 | 现状 | 优化杠杆 |
|------|------|----------|
| 代码即文档 | `.mq.md` = 程序 = 提示工作面;GFM 表 / 链接取元 | 减少 json 胶水;让模型「看见」结构 |
| 写回 | `marqdo-out` 槽;step 默认写回 | 晋升证据、失败避坑 |
| OKF / agent-kb | 命中 / near / promote / solidify / `llm_free` | 复用率、降本主轴 |
| EKC / EFI | L0–L4 catalog;preflight REUSE/ADAPT/CREATE | 生成前先查工程知识 |
| Adaptive routing | marqdo / small-llm / llm / auto + policy 文件 | 再细分 sandbox 试跑 vs 开放推理 |
| Sandbox | CoW + capped `run`;`ext/ai/sandbox` | **试跑闸门**;未 commit 不污染仓库 |
| MCP / corpus | 证据工具,`authority=workbook` | 保持不升权威 |
---
## 4. 问题分析(差距,不是功能羡慕)
### P1 — 试跑与晋升未焊死
沙盒已能 CoW 试跑,但 **`agent.step` / `plan` 默认仍可直接改真实树**;OKF `llm_free` 晋升不强制「沙盒金样绿」。
→ 风险:未验证技能进入复用池(业界 ForgeAgent / MalSkillBench 已证明此问题)。
### P2 — 技能形态双轨
Anthropic Skills = 文件夹 + 渐进披露;Marqdo = 可执行 `.mq.md` + OKF 类型。
→ 机会:用 **OKF frontmatter + 可执行体** 统一,不必引入第二套 SKILL.md 权威;缺的是 **披露层级与验证元数据**(适用条件、fixture、sandbox profile)。
### P3 — 路由粒度偏粗
现有路由偏「整任务后端」;未区分:查 OKF / 写补丁 / 沙盒 `run` / 对外 HTTP。
→ Switchcraft 启示:按 **步骤类型** 选模型,且「便宜模型推理 token 爆炸」要计入总成本。
### P4 — 评测未锁差异化
Wave B1 仍 Active:缺稳定 **Skill Lift / 沙盒证明率 / llm_free 命中率 / 文档可读性** 指标。
→ 没有指标,优化会滑回「多调几次模型」。
### P5 — 代码即文档在 Agent 产出侧仍易腐化
模型仍可能吐 `json.set` 链、混用 EN/ZH API、跳过表结构。
→ 沙盒可跑通语法,但 **风格门禁**(lint / preflight)未默认卡在 commit 前。
### P6 — 上下文与 EKC 未默认进每步
EFI / catalog 能力在,但 step 默认上下文仍可能偏源码切片,缺少「先 REUSE」硬门槛。
### P7 — LLM 表面与 Agent 成本账未统一
`llm.ask` / `stream` / usage 已有;Agent 侧缺 **按会话/技能的成本仪表盘** 写回文档(可 git 审)。
---
## 5. 优化目标函数(建议)
在固定任务集上最大化:
\[
\begin{align*}
& \text{ReuseRate}(\text{OKF hit} + \text{llm\_free}) \\
&+ \text{SandboxProofRate}(\text{试跑绿再 commit/promote}) \\
&- \lambda_1 \cdot \text{TokenCost} - \lambda_2 \cdot \text{HostMutationsWithoutCommit}
\end{align*}
\]
约束:权威真相仍是 `.mq.md`;不引入隐藏 TOOL 主循环;不改 `doc/next/*`。
---
## 6. 外部资料索引(抽样)
| 类型 | 链接 |
|------|------|
| Skills | [Anthropic engineering: Agent Skills](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills) · [Claude Code skills](https://code.claude.com/docs/en/skills) · [agentskills / anthropics/skills](https://github.com/anthropics/skills) |
| 论文 | [SoK: Agentic Skills](https://arxiv.org/html/2602.20867) · [Skill Issue (repo SKILL 优化)](https://arxiv.org/html/2609.12742v1) · [ACES Skill Lift](https://arxiv.org/html/2608.20614) · [MalSkillBench](https://arxiv.org/html/2606.07131v3) · [Compiling workflows into weights](https://arxiv.org/html/2605.22502) |
| 开源/产品 | [ForgeAgent](https://github.com/Yashasm18/ForgeAgent) · [DSPy GEPA](https://dspy.ai/current/getting-started/gepa-optimization/) · [Switchcraft](https://www.microsoft.com/en-us/research/publication/switchcraft-ai-model-router-for-agentic-tool-calling/) |
---
## 7. 结语
Marqdo 不缺「再一个 Agent SDK」;缺的是把 **文档真相、知识晋升、沙盒证明、模型路由** 收成可度量闭环。下一篇方案文档按问题编号给出可选方向与推荐波次,**不预设唯一银弹**。
+392
View File
@@ -0,0 +1,392 @@
# 调研:AI 扩展库沙盒设计(执行隔离 × 工作区暂存 × 写回)
| | |
|---|---|
| 状态 | **调研笔记 · 供 `ext/ai/*` 沙盒设计锁定参考**(**2026-10-16 修订**:纳入「AI 只生成 Marqdo」产品约束) |
| 日期 | 2026-10-16 |
| 范围 | 浏览器 WASM / WASI、OS 进程沙盒、容器 / microVM、CoW 叠加工作区、符号链接暂存方案;对照 Marqdo AI 栈 |
| 相关设计 | **[ext-ai-sandbox.md](../design/ext-ai-sandbox.md)** · **[roadmap/ext-ai-sandbox.md](../roadmap/ext-ai-sandbox.md)** · [ext-agent.md](../design/ext-agent.md) · [ext-agent-plan.md](../design/ext-agent-plan.md) · [stdlib-writeback.md](../design/stdlib-writeback.md) · [stdlib-subtask.md](../design/stdlib-subtask.md) · [browser-marqdo-wasm.md](../design/browser-marqdo-wasm.md) · [adr/0002-browser-marqdo-wasm.md](../adr/0002-browser-marqdo-wasm.md) · [ai-stack.md](../roadmap/ai-stack.md) · [ext-abi.md](../design/ext-abi.md) · [stdlib-modules.md](../design/stdlib-modules.md) |
| 姊妹调研 | [agent-frameworks-and-marqdo.md](agent-frameworks-and-marqdo.md) · [agent-framework-gaps-after-a4.md](agent-framework-gaps-after-a4.md) · [ai-agent-llm-optimize-2026-10.md](ai-agent-llm-optimize-2026-10.md) · [ext-ai-optimize.md](../design/ext-ai-optimize.md) |
---
## 0. 一句话结论
**在「Agent 只生成 / 只执行 Marqdo(`.mq.md`)」这一产品约束下,Marqdo AI 沙盒应以 Wasm 为主沙盒——且必须是「内核 + `lib/*` + `ext/*`」可运行的完整运行时沙盒,而不是只跑光秃解释器。**
锁定方向:
1. **执行隔离 = Marqdo Wasm 运行时(与 ADR 0002 同构)** — deny-by-default 能力导入;**默认不授予** shell / 外联 / `dlopen` 原生插件。AI 产出面收窄到 Markdown 程序后,不必为通用 coding Agent 的「任意二进制」付沙盒税。
2. **库完备性是验收线,不是加分项** — 沙盒内须能 `import lib/…` 与受支持的 `ext/…`;缺能力时 **显式失败**,禁止静默空成功(对齐 [browser-marqdo-wasm.md](../design/browser-marqdo-wasm.md) §5)。
3. **工作区突变 = CoW / whiteout / 显式 commit**(应用层 VFS,必要时 overlay)— 「修改短暂不生效 → 确认写回」保留;**禁止以符号链接森林作为隔离边界**。
4. **OS 沙盒 / microVM** — 降为 **逃逸舱 / 托管 Profile**:仅当显式启用 foreign、`subtask` 真子进程、或未来多租户托管时使用;**不是** Marqdo-only Agent 的默认主路径。
初稿曾按「通用 coding Agent 还需 shell」把 Wasm 与 bwrap 并列;在 Marqdo-only 约束下,主结论改为 **Wasm-first + 库完备 + CoW 写回**。
---
## 1. 调研方法与边界
### 1.1 做了什么
| 动作 | 说明 |
|------|------|
| 厂商/产品文档 | Anthropic Claude Code sandbox / `@anthropic-ai/sandbox-runtime`;VS Code Copilot agent sandboxing;Docker AI Sandboxes;Turso AgentFS CoW overlay;OpenSandbox isolation sessions;WASI / jco preview2-shim;VS Code for the Web WASM webshell |
| 工程对比文 | Northflank / Cosmonic 2026 agent sandbox 指南;gVisor vs Firecracker vs Docker 对比;ContainAI security comparison |
| 学术与标准 | USENIX SEC'22 *Provably-Safe Multilingual Software Sandboxing using WebAssembly*;KTH 论文 *Intra-process Fault Isolation Using WebAssembly*(SFI 通信开销);ACM TOSEM *Research on WebAssembly Runtimes: A Survey*(2025);WASI Design Principles(capability / 无 ambient authority) |
| 安全案例 | brig 项目挂载 symlink 逃逸(CVSS 8.2);Anthropic sandbox-runtime 对 bwrap bind 跟随 symlink 的修复;Docker Sandbox「direct mount vs clone」威胁说明 |
| 对照本仓 | `ext/ai/{llm,agent}`、`lib/writeback`、`lib/subtask`、ADR 0002 浏览器 WASM、`RunOptions.allow_fs_write` 默认关 |
### 1.2 不做的
- 不把「发版用的容器镜像」与「Agent 工具沙盒」混为一谈。
- 不宣称某一云厂商托管沙盒(E2B / Modal / Daytona)为 Marqdo 默认依赖。
- 不修改 `doc/next/*`(手写真相源);本文件只进 `doc/research/`。
### 1.3 产品约束(修订要点):AI 只生成 Marqdo
这是相对通用 coding Agent(Claude Code / Codex)的 **结构性好消息**:
| 通用 Agent | Marqdo AI Agent(本调研采纳) |
|------------|------------------------------|
| 模型可产出 shell / Python / 任意仓库补丁 | 模型产出面收束为 **`.mq.md`(文档即程序)** |
| 沙盒必须兼容「整台开发机能力」 | 沙盒只需跑 **同构 Marqdo 运行时** |
| Wasm 往往只能罩住工具子集 | Wasm **可以**成为主执行边界 |
| 标准库 = 语言运行时 + OS 工具链 | 标准库 / 扩展 = **`lib/*.mq.md` + `ext/**`(+ 受控宿主能力)** |
硬规则建议(后续 design 应写死):
1. Agent 工具表 **不提供** 通用 `bash` / 任意语言 `exec`(除非显式 opt-in 逃逸舱,且另套 OS 沙盒)。
2. Agent 改仓库的权威动作是 **写 / 修订 `.mq.md`(含写回块)**,经 CoW 会话再 `commit`。
3. 运行 Agent 生成的程序 = 在沙盒里 `run` 该 `.mq.md`,且其 `import` 的 **lib 与 ext 必须可用或显式报不可用**。
### 1.4 威胁模型(在 Marqdo-only 下收窄)
| 威胁 | 典型路径 | Marqdo-only 下的权重 |
|------|----------|----------------------|
| T1 提示注入 | 诱导读密钥、外泄、破坏 `.mq.md` | 仍高 — 靠能力拒绝 + CoW |
| T2 工具侧信道 | `subtask`/`foreign` 子进程 | **默认关闭**;opt-in 才升权 |
| T3 工作区即时破坏 | 直接写穿源树 | 靠 CoW;commit 闸门 |
| T4 挂载逃逸 | symlink / bind 打穿 | 仍相关 — **勿用 symlink 作 W 层** |
| T5 网络外泄 | `lib/net` / llm 出站 | 能力开关 + 域名策略 |
| T6 Wasm 宿主桥过宽 | `host_fs_*` / 插件桥过宽 | **主威胁** — 桥 = 权限表 |
| T7 库残缺假安全 | 沙盒能跑 `# main` 但 `import lib/fs` 怪崩或静默空 | **产品验收威胁** |
**成功标准:** 失败域止于会话 upper + 已授予能力;真实工作区默认只读直至 commit;**代表性 `lib/*` 与 `ext/*` 金样在沙盒内可跑或得到明确 `unavailable in sandbox` 诊断**。
---
## 2. 先拆三个「沙盒放哪里」(Marqdo-only 版)
业界「Harness / Workspace / Execution」三分法仍成立,但 **Execution 的内容变了**:
```text
┌──────────────────────────────────────────────────────────┐
│ Pattern H — Harness(编排环) │
│ ext/ai/agent · parent Plan-and-Move ·(可选)MCP client │
├──────────────────────────────────────────────────────────┤
│ Pattern W — Workspace mutation(.mq.md 真相如何变) │
│ CoW upper / whiteout / diff / commit → writeback │
├──────────────────────────────────────────────────────────┤
│ Pattern X — 执行 Agent 产出的 Marqdo 程序 │
│ Wasm Marqdo runtime + embedded/packaged lib + sandboxed │
│ ext 能力面(非「任意 shell」) │
└──────────────────────────────────────────────────────────┘
```
| Pattern | Marqdo 落点 | 沙盒要解决什么 |
|---------|-------------|----------------|
| H | `ext/ai/agent` | 编排是否持有过宽宿主权限;模型 I/O 与执行隔离 |
| W | `lib/writeback` · 工作簿 | 编辑是否立刻污染仓库;人审/父裁决后落盘 |
| X | **Wasm 内 `run` `.mq.md`** | 计算隔离;**lib/ext 能力是否完备且受控** |
用户提案在 Marqdo-only 下:**Wasm 作 X 主底层是正确方向**;仍须改掉 symlink,并补上 **库完备性** 这一环。
---
## 3. 方案谱系(横向对比)
> **读法(修订后):** 下表仍保留通用 Agent 对照。对 **Marqdo-only** 产品,请以 §0 / §5 / §6 的 Wasm-first 结论为准;本节 A/B 升权,F/G/H 降为逃逸舱或托管。
评分:● 强 / ◐ 中 / ○ 弱 / — 不适用。维度按「隔离强度」等客观属性;**产品适配**见 §5。
| 方案 | 隔离强度 | 启动成本 | 任意二进制 | 工作区暂存语义 | 与 Marqdo 契合 | 主要风险 |
|------|----------|----------|------------|----------------|----------------|----------|
| **A. 浏览器 WASM + 虚拟 FS** | ●(计算)◐(经桥的 FS) | ● | ○ | ◐(需自建 overlay) | ●(已有 ADR 0002) | 宿主桥过宽;无原生 toolchain |
| **B. WASI / Component Model(非浏览器亦可)** | ● | ● | ○(仅 WASI 目标) | ●(preopens + 自建 VFS) | ● | 工具链不全则需降级 |
| **C. 符号链接映射工作区 + 延迟写回** | ○ | ● | —(本身不是执行沙盒) | ◐(似是而非) | ○ | **symlink 逃逸是已知高危类** |
| **D. 应用层 CoW VFS(upper/lower/whiteout)** | ◐(逻辑) | ● | — | ● | ●● | 必须配合 X 层 OS 隔离 |
| **E. OS overlayfs / FUSE / AgentFS 式 CoW** | ●(配合命名空间) | ◐ | ●(在沙盒内) | ●● | ● | Linux/macOS 实现分叉;运维 |
| **F. bubblewrap / Seatbelt(Anthropic SRT 类)** | ● | ● | ● | ○(默认可写 cwd) | ● | 需另加 CoW 才有「暂不生效」 |
| **G. Docker / 硬化容器 / gVisor** | ◐→● | ◐ | ● | ◐(volume 策略) | ◐ | 共享内核或兼容性;重 |
| **H. microVM(Firecracker / Kata)** | ●● | ◐ | ● | ◐ | ○(本地 CLI 过重) | KVM/运维;适合多租户托管 |
| **I. 纯语言沙盒(RestrictedPython 等)** | ○ | ● | ○ | — | ○ | 对 Agent 工具链几乎无用 |
### 3.1 A — 浏览器 WASM + 虚拟 FS
**机制:** Wasm 线性内存 + 校验 + 仅经 import 碰宿主;FS 多为内存树 / OPFS / 经 JS bridge 的映射。Marqdo 已锁定「解释器进 WASM」(ADR 0002),默认 `allow_fs_write=false`。
**文献支撑:** Wasm 作为 SFI 窄腰(Bosamiya et al., USENIX SEC'22);进程内沙盒跨边界通信可比 IPC 快数量级(KTH Wasm-SFI 论文);WASI 明确 **capability / 无 ambient authority**。
**优点:** 与「代码即文档」浏览器运行同构;deny-by-default;适合跑 `.mq.md`、公式库、受控插件。
**缺点:** 不能原生跑 `cargo test` / `npm` / 任意 shell;安全上限取决于 **bridge 授予了什么**;浏览器场景下「符号链接用户工作区」通常要落到 File System Access API 或服务端中继,**不是** POSIX symlink 语义。
**判定:** 在 **Marqdo-only** 约束下,作为 **Pattern X 主沙盒** 与浏览器内 Marqdo 的正解;完备性取决于 **lib 嵌入 + ext 能力桥**(§4.1),不取决于 shell。通用「任意二进制 Agent」场景下仍不够——但那不是本产品默认。
### 3.2 B — WASI / Component Model(CLI 宿主)
**机制:** preopens 显式挂载;`WASIShim({ sandbox: { preopens, enableNetwork: false } })`(Bytecode Alliance jco);Component imports = 权限表。
**优点:** 与 Wasm 同族,可在 Node/Wasmtime 跑,不绑浏览器;能力模型清晰。
**缺点:** 同 A,任意 Linux 工具链需「宿主机再包一层」或改编译目标。
**判定:** Marqdo **官方扩展 / 插件中可 Wasm 化的部分**应走这条;与 ADR 0002 一条能力谱。
### 3.3 C — 符号链接映射工作区(用户设想的核心)
**设想:** 沙盒根下用 symlink 指向用户文件;写改暂存别处;确认后写回;删除 = 删 symlink。
**表面吸引力:** 零拷贝读、实现「像」有暂存。
**为何行业不把它当隔离原语:**
1. **Bind/mount 跟随 symlink 是经典逃逸。** Anthropic `sandbox-runtime` 明确修复:`allowWrite` 若为指向界外的 symlink,bwrap `--bind` 会把真实目标变成可写(PR #138 类修复)。
2. **项目挂载未解析 symlink 可致高危。** Endor Labs 披露 brig「project mount symlink traversal」(CVSS 8.2):Agent 在可写项目内把子目录换成指向 `/etc` 的 symlink,下一轮宿主机按字符串挂载即打穿。
3. **「删除 = 删 symlink」语义错误。** 删链接只去掉别名,**不等于**对 lower 层文件的 whiteout;若链接指向真实文件且曾被打开写穿,删除链接也救不回。
4. **写穿风险。** 若某路径被解析成真实 inode 且可写,暂存承诺立刻破产。
5. **跨平台。** Windows junction/reparse、浏览器虚拟 FS、容器 bind,行为不一致。
**判定:** **否决作为 Marqdo 默认工作区隔离方案。** 允许在 **只读、已解析、拒绝可写路径上的 symlink 组件** 的前提下作实现细节,但不得当作安全边界叙事。
### 3.4 D — 应用层 CoW VFS(推荐作 W 层默认)
**机制:**
```text
lower = 真实工作区(只读打开 / 只读能力)
upper = 会话 delta(内存 map / 目录树 / SQLite blob)
whiteout = 逻辑删除集合
view(path) = upper ?? (!whiteout && lower)
commit() = 把 upper/whiteout 应用到 lower(可经人审)
discard() = 丢弃会话
diff() = 可审变更集
```
与 Turso AgentFS、OpenSandbox `workspace.mode: overlay` 的 **语义同构**,但不强制 FUSE。
**与 Marqdo 对齐点:**
| VFS 操作 | Marqdo 真相 |
|----------|-------------|
| `diff()` | 父 Agent / 人审材料;可进 view |
| `commit()` 文本结果 | `writeback.record` / 工作簿补丁 |
| `discard()` | 失败回合不污染仓库 |
| 只读 lower | 防止 T3 即时破坏 |
**优点:** 纯用户态、可测、可嵌入 `plugins/agent`;不依赖 root/FUSE;金样可锁。
**缺点:** 对 **沙盒内原生工具**(它们走真实 POSIX)不可见,除非再挂 E 或把工具 I/O 也改走 VFS API。
**判定:** **Pattern W 的默认实现。**
### 3.5 E — OS CoW overlay(AgentFS / overlayfs / OpenSandbox)
**机制(AgentFS 公开设计):** lower=宿主只读;upper=SQLite delta;删除=whiteout;Linux 用 FUSE+user/mount ns;macOS 用 NFS+`sandbox-exec`。OpenSandbox 默认 `overlay`,会话结束可丢 upper。
**优点:** 对 **任意 shell 工具**透明;「改了但宿主机 `git status` 仍干净」体验业界已验证。
**缺点:** 平台依赖;FUSE 权限;与应用层 D 双实现成本。
**判定:** **当 Pattern X 需要真实 shell 且要求暂存对工具透明时**启用(Linux 优先);作为 D 的「native 加速后端」,不是替代 Wasm。
### 3.6 F — OS 进程沙盒(bwrap / Seatbelt / VS Code Copilot)
**机制:** Anthropic:文件系统隔离 + **网络隔离(缺一不可)**;Linux bubblewrap + 代理;macOS Seatbelt。VS Code Copilot:同类依赖;路径分 readwrite / readonly / denied;明确「不是 VM,削弱授权即削弱边界」。
**优点:** 本地编码 Agent 的事实标准;覆盖子进程;启动轻。
**缺点:** 默认可写 cwd → **没有**「暂不生效」;必须与 D/E 组合。Symlink 配置路径需解析与拒绝。
**判定:** **Pattern X(shell/外联/MCP)本地默认。**
### 3.7 G / H — 容器与 microVM
**共识:** 裸 Docker 共享内核,不适合作唯一边界应对不可信代码;gVisor 加强 syscall 拦截;Firecracker/Kata 适合多租户与高威胁(~125ms 级启动)。Docker AI Sandboxes 区分 **direct mount(即时写穿)** vs **clone(只读源 + 私有副本)**——后者才接近「确认后写回」。
**判定:** Marqdo **本地 CLI 默认不强制 microVM**;若未来 `ext` 提供托管执行 / 多租户,再把 H 定为服务端 Profile。G 可作为打包/CI Profile。
### 3.8 I — 语言级沙盒
对「Agent 会跑 shell」场景基本无效;仅作极窄解释器内嵌参考。
---
## 4. 专门评估:用户提案(修订)
> 浏览器 WASM 作沙盒底层 + 符号链接用户工作区 + 修改短暂不生效 + 确认后写回 + 删除变删 symlink。
> **补充约束:** AI 只生成 Marqdo;沙盒须能用标准库与扩展库。
| 子命题 | 评价 | 说明 |
|--------|------|------|
| Wasm 作主执行隔离 | **在 Marqdo-only 下正确** | 与 ADR 0002 同构;不必为 shell 全集付税 |
| 沙盒须含 lib + ext | **正确且关键** | 「能跑 `# main`」≠ 好沙盒;见 §4.1 |
| 「短暂不生效」 | **正确** | CoW / staging(方案 D,必要时 E) |
| 「确认后写回」 | **正确且极契合 Marqdo** | 与 writeback、父裁决、workbook 同一哲学 |
| 「symlink 映射工作区」 | **仍错误** | 与是否只用 Marqdo 无关;换 CoW/whiteout |
| 「删除 = 删 symlink」 | **仍错误** | 应为 whiteout;commit 时再对 lower 删除 |
| 「性能最好且安全性最高」 | **Wasm 主路径下接近成立** | 前提是桥最小权限 + CoW;symlink 仍拉低安全 |
**修订后的正确表述:**
> **以 Wasm Marqdo 运行时为主沙盒,打包/嵌入 `lib/*`,经能力桥提供受支持的 `ext/*`;工作区用 lower 只读 + upper/whiteout;人审/父裁决后 commit。符号链接不作隔离原语。shell/外联/原生 dlopen 默认关闭,仅作显式逃逸舱。**
### 4.1 库完备性:什么叫「标准库和扩展库也能用」
沙盒若只能跑内置算术、却不能 `import lib/json` 或跑 `ext/ai/llm` 的离线金样,对 Agent 开发几乎无用——模型生成的 `.mq.md` **本来就会依赖 lib/ext**。
三类库进沙盒的路径不同:
| 类别 | 今日形态 | 沙盒策略 |
|------|----------|----------|
| **纯 `.mq.md` 库**(多数 `lib/*`、部分 `ext` 门面) | 源码 import | **嵌入或会话导入图**(已有 `embedded_lib` / import map)— 必须默认可用 |
| **需宿主能力的 lib**(`fs` / `net` / `writeback` / `subtask` / `datetime.now`…) | `host_*` | **经能力表桥接**到会话 VFS / 受控 clock / 受控 fetch;默认最小集 |
| **原生 ABI 扩展**(`plugins/web`、`plugins/agent`、quantum 等 `.so`) | `plugin.load` + dlopen | 浏览器/Wasm **无 dlopen**(ADR 0002 N2)。沙盒内应:**纯 mq 门面 + Wasm 插件**、或 **宿主代执行且权限计入会话**、或 **显式 `unavailable`** — 禁止假装成功 |
与 [browser-marqdo-wasm.md](../design/browser-marqdo-wasm.md) §5 对齐并扩展为 **Agent 沙盒能力矩阵(草案)**:
| 能力 | 沙盒默认 | 备注 |
|------|----------|------|
| builtins / 表 / 控制流 / 对象 | ✓ | 内核 |
| `lib/math` `json` `table` `text` `collection`… | ✓ | embedded |
| `lib/fs` | ✓(仅会话 VFS) | 写进 upper;禁越出 root |
| `lib/writeback` | ✓(写会话内 `.mq.md`) | commit 后才进真实树 |
| `lib/net` / llm HTTP | 默认关 / 域名白名单 | 与 T5 |
| `lib/subtask` `fn=` | ✓(同 Wasm 会话) | |
| `lib/subtask` `path=`/`code=`/`lang=` | **默认拒** | 逃逸舱 + OS 沙盒 |
| `lib/plugin` + 原生 `.so` | **默认拒** | 改 Wasm 插件或宿主代执行 |
| `ext/ai/llm` · `ext/ai/agent`(mq 门面) | ✓(能力允许时) | 网络/密钥另授 |
| `ext/web` 服务端 listen/DB | **默认拒或只读 stub** | 完整 axum 不进沙盒;可测「装配/渲染」子集 |
| `ext/quantum` / `ext/linalg` | 目标 ✓ | 优先纯逻辑 / Wasm 插件,避免强依赖宿主机 .so |
**验收金样建议(S 阶段):**
一套「沙盒烟囱」:`import` 常用 lib → 写会话文件 → writeback → `diff` → `commit`;再各挑一个 **纯逻辑 ext** 与一个 **需网络的 ext**(后者验证拒绝/白名单)。缺能力必须诊断含 `unavailable in sandbox`(或同等措辞)。
---
## 5. 综合评分(面向 Marqdo-only 产品)
加权(修订):执行隔离契合 Marqdo-only 30%、**lib/ext 完备** 25%、与 writeback 真相 20%、实现成本 15%、跨端(CLI↔浏览器)10%。
(「任意 shell 工具完备」不再作为主权重。)
| 候选架构 | 分(相对) | 备注 |
|----------|------------|------|
| **A/B Wasm-first + D CoW + 库完备能力表** | **100(推荐)** | 主路径 |
| 同上 + 逃逸舱 F(opt-in shell) | 95 | 兼容少数 foreign 需求 |
| D + E + F(通用 coding 向) | 75 | 过重;非 Marqdo-only 默认 |
| 仅 F(可写 cwd) | 55 | 缺 CoW;且鼓励非 mq 工具 |
| 仅 A 但无 CoW、无 lib 嵌入 | 40 | 「半沙盒」 |
| 仅 C(symlink) | 20 | 否决 |
| 仅 H(人人 microVM) | 50 | 托管可选;本地默认过重 |
**绝对「最硬」单点**仍可以是 microVM(托管 Profile)。
**Marqdo「最合适」:** **Wasm 完整运行时沙盒(内核+lib+受控 ext)+ CoW 工作区 + commit 闸门**。
---
## 6. Marqdo 应采用的沙盒设计(推荐锁定草案 · 修订)
### 6.1 产品形态
| Profile | 场景 | 执行(X) | 工作区(W) |
|---------|------|-----------|------------|
| **`mq-sandbox`(默认)** | Agent 生成并试跑 `.mq.md` | **Wasm Marqdo** + embedded `lib/*` + 能力桥 `ext` | 应用层 CoW;commit→真实树 / writeback |
| **`browser`** | 页内同构运行 | 同左(ADR 0002) | 内存/OPFS;无 POSIX symlink |
| **`escape`(显式)** | foreign / 真子进程 | OS 沙盒包一层 | 仍建议 CoW 可见 |
| **`hosted`(可选)** | 多租户 | Wasm 或 microVM | 会话隔离;凭证外置 |
### 6.2 推荐架构图
```text
┌──────────────────────────────────────┐
│ ext/ai/agent (Harness) │
│ 只产出 / 编排 .mq.md · 人审闸门 │
└──────────────────┬───────────────────┘
│
┌─────────────────────┼─────────────────────┐
▼ ▼
┌────────────────────┐ ┌────────────────────┐
│ Workspace CoW (W) │ │ Marqdo Wasm (X) │
│ lower = RO 工作区 │◄── fs/writeback ───│ 同构解释器/VM │
│ upper / whiteout │ 能力导入 │ + embedded lib/* │
│ diff·commit·discard│ │ + ext 能力面 │
└─────────┬──────────┘ │ 默认: 无 shell/so │
│ └────────────────────┘
▼
真实工作区 / <!-- marqdo-out -->
(代码即文档真相源)
(可选)escape: subtask path|code|lang → OS sandbox
```
### 6.3 API 面建议(设计级,非立即实现)
落点倾向:宿主能力进 `plugins/agent` / Wasm HostContext;作者面经 `ext/ai` 或 `lib`(**禁止** `ext/**` 直调 `host_*`)。
| 概念 | 语义 |
|------|------|
| `sandbox.open(root, mode=cow)` | 打开会话;lower 只读 |
| `sandbox.read/write/list/remove` | 经 view;remove → whiteout |
| `sandbox.diff` / `stat` | 供父 Agent / view |
| `sandbox.commit` / `discard` | 闸门;可要求 HITL |
| `sandbox.run(source\|path, caps=…)` | **Wasm 跑 Marqdo**;caps 声明 fs/net/llm/… |
| `sandbox.caps` / `inspect` | 列出已授能力与 lib/ext 可用性(对齐 `web.inspect` 哲学) |
| `sandbox.exec(...)` | **仅 escape Profile**;默认不存在于作者工具表 |
与现有衔接:
- `lib/*`:优先 embedded;FS/时间/网络走会话 caps。
- `ext/*`:mq 门面可加载;原生插件默认不可用并诊断;长期可 Wasm 化的(linalg/quantum 逻辑)优先进 Wasm。
- `lib/writeback`:写入会话内文件;`commit` 时进入真实树。
- `lib/subtask`:默认仅 `fn=`;`path`/`code`/`lang` 要 escape。
### 6.4 硬禁(写入后续 design 时应保留)
1. **不以 symlink 森林作为隔离或暂存原语。**
2. **不以「能跑空 `# main`」宣称沙盒完成** — 必须过 lib/ext 完备金样或显式 unavailable。
3. **默认工具表不提供通用 shell / 任意语言 exec。**
4. **不默认可写 bind 真实工作区。**
5. **宿主桥 = 权限表**:未授能力不得 ambient 成功。
6. **不静默降级**:缺 Wasm/能力时不得假装已隔离。
7. **`ext/**` 不调 `host_*`**(既有宪法)。
### 6.5 分阶段落地(建议,非 `doc/next`)
| 阶段 | 交付 | 验收线索 |
|------|------|----------|
| S0 | 威胁模型 + 本调研(含 Marqdo-only) | 本文 |
| S1 | Wasm `run` + embedded 核心 `lib/*` + 会话 VFS 读 | 常用 lib 金样在沙盒绿 |
| S2 | CoW upper/whiteout + diff/commit/discard | 未 commit 时宿主机 `git status` 干净 |
| S3 | writeback / 工作簿在会话内可写;commit 对齐 | agent step 金样 |
| S4 | caps:net/llm 白名单;`subtask` 非 fn 默认拒 | 诊断字符串稳定 |
| S5 | 代表性 ext(纯逻辑 + 需网)沙盒金样 | 完备性验收 |
| S6 | (可选)escape OS 沙盒;hosted microVM | 文档标明非默认 |
---
## 7. 外部资料索引(抽样)
| 类型 | 引用 |
|------|------|
| 产品 | [Anthropic: Claude Code sandboxing](https://www.anthropic.com/engineering/claude-code-sandboxing) · [sandbox-runtime](https://github.com/anthropic-experimental/sandbox-runtime) · [VS Code agent sandboxing](https://code.visualstudio.com/docs/agents/run/agent-sandboxing) · [Docker AI Sandboxes security](https://docs.docker.com/ai/sandboxes/security/) · [Turso AgentFS overlay](https://turso.tech/blog/agentfs-overlay) · [OpenSandbox isolation sessions](https://open-sandbox.ai/guides/isolation-sessions) |
| 标准 | [WASI Design Principles](https://github.com/WebAssembly/WASI/blob/main/docs/DesignPrinciples.md) · [jco preview2-shim sandbox](https://github.com/bytecodealliance/jco/blob/main/packages/preview2-shim/README.md) |
| 论文 | Bosamiya et al., *Provably-Safe Multilingual Software Sandboxing using WebAssembly*, USENIX Security 2022 · *Intra-process Fault Isolation Using WebAssembly* (KTH) · *Research on WebAssembly Runtimes: A Survey*, ACM TOSEM 2025 |
| 综述 | [Cosmonic: Choosing an AI agent sandbox (2026)](https://cosmonic.com/ai-sandbox/choosing-an-agent-sandbox/) · [Northflank: MicroVMs, gVisor & isolation](https://northflank.com/blog/how-to-sandbox-ai-agents) |
| 安全案例 | [brig symlink traversal / Endor Labs](https://www.endorlabs.com/learn/project-mount-symlink-traversal-in-brig-how-a-sandbox-reaches-host-directories-it-was-never-given) · Anthropic sandbox-runtime symlink bind 修复 |
---
## 8. 结语
- **产品好消息:** Agent 只生成 Marqdo → **Wasm 主沙盒设计成立**,不必默认扛通用 shell Agent 的复杂度。
- **好沙盒的定义:** 不只跑内核,还要 **`lib/*` 可用、`ext/*` 在能力表内可用或显式不可用**;再加 **CoW + commit** 保护仓库真相。
- **仍要改掉的:** 符号链接当暂存/隔离;默认开放 foreign/dlopen;「空 main 能跑」就算完工。
- **降级为非默认的:** bwrap/microVM — 留给 escape / hosted,而不是 Marqdo-only 主路径。
不急着实现。下一步若继续收敛,再开 `doc/design/ext-ai-sandbox.md` 把 **能力矩阵 + 库完备金样清单** 锁死即可;**仍不要改 `doc/next/*`**。
+1
View File
@@ -18,6 +18,7 @@
|----|------|------|
| **Intelligence primitive** | `ask` / `stream` / `collect`;结果对象 | `ext/ai/llm` |
| **Agent runtime** | `run` / `step` / `plan` / `route` | `ext/ai/agent` |
| **Sandbox (trial)** | CoW workspace + capped `run` | `ext/ai/sandbox` · [ext-ai-sandbox.md](../design/ext-ai-sandbox.md) |
| **Memory / compile** | episode → skill → policy | `plugins/agent`(ABI;**不进** `src/host`) |
| **Transport** | HTTP / SSE / provider | `ext/ai/llm/openai` · `ollama`;用户不可见 |
+1
View File
@@ -7,6 +7,7 @@
| 调研 | [agent-frameworks-and-marqdo.md](../research/agent-frameworks-and-marqdo.md) · [agent-framework-gaps-after-a4.md](../research/agent-framework-gaps-after-a4.md) · **[agent-framework-2026-09.md](../research/agent-framework-2026-09.md)** |
| 设计 | [ext-agent.md](../design/ext-agent.md) · [ext-agent-plan.md](../design/ext-agent-plan.md) · [ext-agent-parent.md](../design/ext-agent-parent.md) · [okf.md](../design/okf.md) |
| 邻近 | [agent-streaming.md](agent-streaming.md) · [okf-near-match.md](okf-near-match.md) |
| 后续 | 优化波次 O1–O5 见 [ext-ai-optimize.md](ext-ai-optimize.md)(沙盒证明 · Skill Lift · 路由/成本 · lint · 隔离) |
---
+29
View File
@@ -0,0 +1,29 @@
# Roadmap: Agent/LLM 优化波次 O1–O5
| | |
|---|---|
| Status | **Active** |
| Date | 2026-10-16 |
| Design | [ext-ai-optimize.md](../design/ext-ai-optimize.md) |
| Research | [ai-agent-llm-optimize-2026-10.md](../research/ai-agent-llm-optimize-2026-10.md) |
| Prior | [ext-agent-optimize.md](ext-agent-optimize.md) (A0–A4 / Wave B) · [ext-ai-sandbox.md](../design/ext-ai-sandbox.md) |
## Principles
1. Executable `.mq.md` remains authority; OKF reuses; sandbox proves; LLM explores.
2. No hidden TOOL: loop; no vector-store memory authority; no default shell Agent.
3. Do not edit `doc/next/*`.
## Waves
| Wave | Scope | Gold / deliverable | Status |
|------|-------|--------------------|--------|
| **O1** | A1 opt-in `sandbox=` · A3 profile table · D1 skill · D4 example | `agent-sandbox-optin` · `examples/agent-sandbox-okf` | **done** |
| **O2** | A2 promote proof · B1 OKF meta · B3/E1 harness Skill Lift | `agent-sandbox-promote` · harness subset | **done** |
| **O3** | C1 step route · C2 cost · B6 EFI default | `agent-step-route` · run usage/efi | **done** |
| **O4** | A4/D2 lint · A5 diff review report | `scripts/mq-style-lint.sh` · `sandbox.report` | **done** |
| **O5** | B4 quarantine · C3 offline policy rewrite | `agent-sandbox-quarantine` · `optimize-standing-policy.sh` | **done** |
## Non-goals
F1–F5 · wasmtime CLI embed · default `sandbox=True` without ADR · edit `doc/next/*`
+28
View File
@@ -0,0 +1,28 @@
# Roadmap: `ext/ai/sandbox` (AI CoW trial runtime)
| | |
|---|---|
| Status | **Active** |
| Date | 2026-10-16 |
| Design | [ext-ai-sandbox.md](../design/ext-ai-sandbox.md) |
| Research | [ai-agent-sandbox.md](../research/ai-agent-sandbox.md) |
## Milestones
| ID | Deliverable | Done when |
|----|-------------|-----------|
| S0 | Design + this roadmap + research cross-links | **done** |
| S1 | CoW open/write/read/remove/diff/commit/discard | **done** (`sandbox-cow`) |
| S2 | `run` + default caps + core lib inside run | **done** (`sandbox-run-lib`) |
| S3 | `inspect` + deny plugin.load | **done** (`sandbox-deny`) |
| S4 | Polish: ai-stack + CHANGELOG | **done** |
## Gold samples (target)
- `tests/ext/sandbox-cow.mq.md`
- `tests/ext/sandbox-run-lib.mq.md`
- `tests/ext/sandbox-deny.mq.md`
## Out of scope this wave
wasmtime CLI embed · Wasm ABI plugins · bwrap/microVM · full web in sandbox · auto-wire agent.step · edit `doc/next/*`
+13
View File
@@ -0,0 +1,13 @@
# agent-sandbox-okf(O1)
演示 **opt-in 沙盒试跑 + 表风格**(少 json 胶水):`agent.trial` 写入先落 upper,`commit=False` 时宿主不变,成功后再 `commit=True`。
```bash
cargo build --release -p marqdo -p marqdo_plugin_agent
export MARQDO_EXT="$PWD/ext"
./target/release/marqdo run examples/agent-sandbox-okf/index.mq.md
```
期望:`dry-ok` · `host-clean` · `wet-ok` · `proof-visible` · `profile-default`。
见 [ext-ai-optimize.md](../../doc/roadmap/ext-ai-optimize.md) · 金样 [agent-sandbox-optin.mq.md](../../tests/ext/agent-sandbox-optin.mq.md)。
+59
View File
@@ -0,0 +1,59 @@
---
title: agent-sandbox-okf — trial then commit
description: >
O1 example: opt-in sandbox trial with GFM-table returns and link-index reads.
import agent:ext/ai/agent.mq.md
import fs:lib/fs.mq.md
---
# main
Short demo: write via `agent.trial`, prove the host is clean until commit.
**had = fs.exists path="_example_sbx"**
1. `had`
> fs.remove path="_example_sbx"
2. *
**_ = 0**
> fs.make_dir path="_example_sbx"
**dry = agent.trial root="_example_sbx" write_path="note.txt" text=sandbox-okf commit=False**
**dry_ok = [ok](dry)**
**dry_committed = [committed](dry)**
1. `dry_ok`
1. not `dry_committed`
> print text=dry-ok
2. *
> print text=dry-bad-commit
2. *
> print text=dry-fail
**leaked = fs.exists path="_example_sbx/note.txt"**
1. `leaked`
> print text=host-leaked
2. *
> print text=host-clean
**wet = agent.trial root="_example_sbx" write_path="note.txt" text=sandbox-okf commit=True**
**wet_ok = [ok](wet)**
**wet_committed = [committed](wet)**
1. `wet_ok`
1. `wet_committed`
> print text=wet-ok
2. *
> print text=wet-no-commit
2. *
> print text=wet-fail
**body = fs.read_text path="_example_sbx/note.txt"**
1. `body` == "sandbox-okf"
> print text=proof-visible
2. *
> print text=proof-missing
**pname = [name]([profile](wet))**
1. `pname` == "default"
> print text=profile-default
2. *
> print text=profile-other
+1 -1
View File
@@ -1 +1 @@
1.4.0
1.4.5
+641 -336
View File
File diff suppressed because it is too large. Load diff
+42 -15
View File
@@ -5,6 +5,7 @@ description: >-
OpenAI-compatible and Ollama HTTP stay behind the semantic surface.
import sys:lib/sys.mq.md
import json:lib/json.mq.md
import table:lib/table.mq.md
import fs:lib/fs.mq.md
import text:lib/text.mq.md
import openai:ext/ai/llm/openai.mq.md
@@ -25,14 +26,14 @@ Collapse a stream event list to final answer text (`done.result` or `error.messa
**answer = ""**
- [ev](events)
**t = [type](ev)**
**t = > json.get value=`ev` key="type"**
1. `t` == "done"
**answer = [result](ev)**
**answer = > json.get value=`ev` key="result"**
2. `t` == "error"
**answer = [message](ev)**
**answer = > json.get value=`ev` key="message"**
3. `t` == "delta"
4. `t` == "text"
**chunk = [text](ev)**
**chunk = > json.get value=`ev` key="text"**
1. `chunk`
**answer = answer + chunk**
2. *
@@ -167,7 +168,8 @@ Module convenience: create a default handle and return **answer text**. Pass `pa
**m = > create model=`model` base_url=`base_url` api_key=`api_key`**
**r = > m.ask prompt=`p`**
*[text](r)*
**_ret = > json.get value=`r` key="text"**
*_ret*
# llm
+ `model`=None
@@ -283,9 +285,21 @@ Semantic ask → **LLMResult** (`text`, `model`, `finish`, `backend`, `usage`, `
**text = > json.get value=`pack` key="text"**
**usage = > json.get value=`pack` key="usage"**
**finish = > json.get value=`pack` key="finish"**
**err = > json.get value=`pack` key="error"**
**tc = > json.get value=`pack` key="tool_calls"**
1. not `tc`
**tc = > json.parse text=[]**
**tc = > table.empty_list**
2. *
**_ = 1**
1. not `text`
**text = ""**
2. *
**_ = 1**
1. not `finish`
1. `err`
**finish = "error"**
2. *
**finish = "stop"**
2. *
**_ = 1**
@@ -293,16 +307,27 @@ Hoist receiver fields before the result table — table cells can shadow `self`
column names (`model`, `usage`), so `[model](self)` / `[total_tokens](usage)`
inside the table is unsafe.
**mdl = [model](self)**
**b = [backend](self)**
**nm = [name](self)**
**tt = [total_tokens](usage)**
**mdl = > json.get value=`self` key="model"**
**b = > json.get value=`self` key="backend"**
**nm = > json.get value=`self` key="name"**
1. `usage`
**tt = > json.get value=`usage` key="total_tokens"**
2. *
**tt = 0**
1. not `tt`
**tt = 0**
2. *
**_ = 1**
1. `err`
**calls = 0**
2. *
**calls = 1**
`result` =
| text | model | finish | backend | usage | name | llm_calls | tokens | tool_calls |
|------|-------|--------|---------|-------|------|-----------|--------|------------|
| `text` | `mdl` | `finish` | `b` | `usage` | `nm` | 1 | `tt` | `tc` |
| text | model | finish | backend | usage | name | llm_calls | tokens | tool_calls | error |
|------|-------|--------|---------|-------|------|-----------|--------|------------|-------|
| `text` | `mdl` | `finish` | `b` | `usage` | `nm` | `calls` | `tt` | `tc` | `err` |
*result*
@@ -345,10 +370,12 @@ Internal / compatibility primitive. Prefer `ask` / `stream` in new code.
1. `stream`
**pack = > self.complete_result prompt=`prompt` stream=True echo=`echo`**
*[events](pack)*
**_ret = > json.get value=`pack` key="events"**
*_ret*
2. *
**pack = > self.complete_result prompt=`prompt` stream=False echo=`echo`**
*[text](pack)*
**_ret = > json.get value=`pack` key="text"**
*_ret*
## chat
+ `prompt`
+65 -26
View File
@@ -4,6 +4,7 @@ description: >-
OpenAI-compatible chat completions transport (HTTP + SSE).
Prefers optional plugins/llm ABI when present; falls back to lib/net.
Used by ext/ai/llm; authors should prefer llm.ask / llm.stream.
Result packs are GFM tables / table.put (code-as-documentation).
import net:lib/net.mq.md
import json:lib/json.mq.md
import table:lib/table.mq.md
@@ -53,21 +54,18 @@ Extract OpenAI-shaped `choices[0].message.tool_calls` as a list (or empty list).
**choices = > json.get value=`data` key="choices"**
1. not `choices`
**empty = > json.parse text=[]**
*empty*
*> table.empty_list*
2. *
**c0 = [1](choices)**
**msg = > json.get value=`c0` key="message"**
1. not `msg`
**empty = > json.parse text=[]**
*empty*
*> table.empty_list*
2. *
**tc = > json.get value=`msg` key="tool_calls"**
1. `tc`
*tc*
2. *
**empty = > json.parse text=[]**
*empty*
*> table.empty_list*
## chat_completions
+ `base_url`
@@ -79,7 +77,7 @@ Extract OpenAI-shaped `choices[0].message.tool_calls` as a list (or empty list).
+ `suffix`=/chat/completions
+ `bearer`="Bearer "
POST `/chat/completions` (or stream SSE). Prefers `plugins/llm` ABI (`llm_chat_completions`) when the native plugin is resolvable; otherwise `lib/net` HTTP/SSE. Non-stream → `{text,usage,finish,tool_calls}`; stream → `{events,finish}`.
POST `/chat/completions` (or stream SSE). Prefers `plugins/llm` ABI (`llm_chat_completions`) when the native plugin is resolvable; otherwise `lib/net` HTTP/SSE. Non-stream → `{text,usage,finish,tool_calls}`; stream → `{events,finish}`. Soft transport / HTTP failures return `finish=error` packs (never abort the VM).
**p = > plugin.native_path name="llm"**
1. `p`
@@ -106,40 +104,81 @@ POST `/chat/completions` (or stream SSE). Prefers `plugins/llm` ABI (`llm_chat_c
**req = > table.put in=`req` at="stream" value=True**
**body = > json.stringify value=`req`**
**resp = > net.http_post_sse url=`url` body=`body` headers=`headers` echo=`echo`**
1. [status](resp) == 200
**pack = > json.parse text={"finish":"stop"}**
**pack = > json.set map=`pack` key="events" value=[events](resp)**
**st = > json.get value=`resp` key="status"**
1. `st` == 200
**evs = > json.get value=`resp` key="events"**
`pack` =
| finish | events |
|--------|--------|
| stop | `evs` |
*pack*
2. *
> print text=ext/ai/llm/openai: HTTP error (stream)
> print text=[status](resp)
> sys.exit code=1
Soft transport / HTTP failure — structured pack, never abort the VM.
**msg = "HTTP " + `st`**
**evs = > json.get value=`resp` key="events"**
1. `evs`
**e0 = > at value=`evs` index=0**
**m2 = > json.get value=`e0` key="message"**
1. `m2`
**msg = `m2`**
`pack` =
| finish | text | error | events |
|--------|------|-------|--------|
| error | "" | `msg` | `evs` |
*pack*
2. *
**body = > json.stringify value=`req`**
**resp = > net.http_post url=`url` body=`body` headers=`headers`**
1. [status](resp) == 200
**data = > json.parse text=[body](resp)**
**msg = [message]([1]([choices](data)))**
**st = > json.get value=`resp` key="status"**
1. `st` == 200
**raw = > json.get value=`resp` key="body"**
**data = > json.parse text=`raw`**
**choices = > json.get value=`data` key="choices"**
**choice0 = > at value=`choices` index=0**
**msg = > json.get value=`choice0` key="message"**
**text = > json.get value=`msg` key="content"**
1. not `text`
**text = ""**
2. *
**_ = 1**
**usage = > usage_from data=`data`**
**fr = > json.get value=[1]([choices](data)) key="finish_reason"**
**fr = > json.get value=`choice0` key="finish_reason"**
1. not `fr`
**fr = "stop"**
2. *
**_ = 1**
**tc = > tool_calls_from data=`data`**
**pack = > json.parse text={"finish":"stop"}**
**pack = > json.set map=`pack` key="text" value=`text`**
**pack = > json.set map=`pack` key="usage" value=`usage`**
**pack = > json.set map=`pack` key="finish" value=`fr`**
**pack = > json.set map=`pack` key="tool_calls" value=`tc`**
`pack` =
| finish | text | usage | tool_calls |
|--------|------|-------|------------|
| `fr` | `text` | `usage` | `tc` |
*pack*
2. *
> print text=ext/ai/llm/openai: HTTP error
> print text=[status](resp)
> print text=[body](resp)
> sys.exit code=1
Soft transport / HTTP failure — structured pack, never abort the VM.
**raw = > json.get value=`resp` key="body"**
**msg = "HTTP " + `st`**
1. `raw`
**msg = `msg` + ": " + `raw`**
`usage0` =
| prompt_tokens | completion_tokens | total_tokens |
|---------------|-------------------|--------------|
| 0 | 0 | 0 |
**empty_tc = > table.empty_list**
`pack` =
| finish | text | error | usage | tool_calls |
|--------|------|-------|-------|------------|
| error | "" | `msg` | `usage0` | `empty_tc` |
*pack*
+25
View File
@@ -0,0 +1,25 @@
---
title: step-route policy (O3)
description: >
Table-driven map from step class → Adaptive Routing backend.
Readable authority for agent.step_route; do not invent a second dialect.
---
# Step class → backend
Authors and `agent.step_route` read this table. Backends match Adaptive Routing:
`marqdo` | `small-llm` | `llm` | `auto`.
**routes = **
| kind | backend | note |
|------|---------|------|
| lookup | marqdo | OKF / kb skill hit — no parent LLM |
| patch | small-llm | Local patch / soft match |
| sandbox_run | marqdo | CoW trial of `.mq.md` |
| net | llm | Needs network / toolful turn |
| default | auto | Cascade policy → kb → small-llm |
# main
*routes*
+191
View File
@@ -0,0 +1,191 @@
---
title: ext/ai/sandbox
description: >-
CoW trial workspace for AI-generated Marqdo: open / run / diff / commit.
Native work lives in plugins/agent (`sandbox_*` → host_query).
import plugin:lib/plugin.mq.md
import sys:lib/sys.mq.md
import table:lib/table.mq.md
import fs:lib/fs.mq.md
---
Load the agent plugin once per call (same pattern as other `ext/ai` helpers).
Capability profiles are GFM tables under `## profile` (readable authority; `open` / `inspect` echo them). Do not invent a second table dialect.
## ensure_plugin
**p = plugin.native_path name="agent"**
1. `p`
> plugin.load path=`p`
2. *
> print text=ext/ai/sandbox: native agent plugin not found (build marqdo_plugin_agent or marqdo ext add agent)
> sys.exit code=1
## profile
+ `name`=default
Named caps row as a map table (`name`, `fs_write`, `exec`, `net`, `plugin`). Unknown names → `default`.
Catalog (documentation + authority):
| name | fs_write | exec | net | plugin |
|------|----------|------|-----|--------|
| default | true | false | false | false |
| readonly | false | false | false | false |
1. `name` == "readonly"
**row = **
| name | fs_write | exec | net | plugin |
|------|----------|------|-----|--------|
| readonly | false | false | false | false |
2. *
**row = **
| name | fs_write | exec | net | plugin |
|------|----------|------|-----|--------|
| default | true | false | false | false |
*row*
## open_native
+ `root`=None
Native CoW open (plugin ABI italic return). Prefer `## open` for the profile-aware facade.
> ensure_plugin
1. `root`
*> sandbox_open root=`root`*
2. *
*> sandbox_open*
## open
+ `root`=None
+ `profile`=default
Open a CoW session. Optional `root=` (directory). Default: entry file directory.
`profile=` selects the caps row from `## profile` (echoed by `inspect`).
**sess = open_native root=`root`**
**prof = profile name=`profile`**
**sess = table.put in=`sess` at="profile" value=`prof`**
*sess*
## write
+ `id`
+ `path`
+ `text`
Write text into the session upper layer (host tree unchanged until `commit`).
> ensure_plugin
*> sandbox_write id=`id` path=`path` text=`text`*
## read
+ `id`
+ `path`
Read through CoW (upper, else lower; whiteout misses).
> ensure_plugin
*> sandbox_read id=`id` path=`path`*
## remove
+ `id`
+ `path`
Logical delete (whiteout).
> ensure_plugin
*> sandbox_remove id=`id` path=`path`*
## diff
+ `id`
List of `{ path, kind }` pending changes (`write` | `remove`).
> ensure_plugin
*> sandbox_diff id=`id`*
## commit
+ `id`
Apply upper / whiteout to the real workspace root.
> ensure_plugin
*> sandbox_commit id=`id`*
## discard
+ `id`
Drop the session without writing the host tree.
> ensure_plugin
*> sandbox_discard id=`id`*
## run
+ `id`
+ `source`=None
+ `path`=None
Run Marqdo under sandbox caps (`exec` / `net` / `plugin` off; fs via CoW).
Pass `source=` or `path=` (relative to session root).
> ensure_plugin
1. `source`
*> sandbox_run id=`id` source=`source`*
2. `path`
*> sandbox_run id=`id` path=`path`*
3. *
> print text=ext/ai/sandbox.run: need source= or path=
> sys.exit code=1
## inspect_native
+ `id`
> ensure_plugin
*> sandbox_inspect id=`id`*
## inspect
+ `id`
+ `profile`=default
Native caps / change counts, plus the GFM `## profile` row under key `profile`.
**raw = inspect_native id=`id`**
**prof = profile name=`profile`**
**out = table.put in=`raw` at="profile" value=`prof`**
*out*
## report
+ `id`
+ `path`=sandbox-diff.md
O4: write a markdown review of `## diff` for `marqdo view` / HITL (static report file, not a UI rewrite).
**pending = diff id=`id`**
**n = len value=`pending`**
**body = "# Sandbox diff report\n\nPending changes: "**
**body = body + n**
**body = body + "\n\n| path | kind |\n|------|------|\n"**
- [row](pending)
**p = [path](row)**
**k = [kind](row)**
**body = body + "| "**
**body = body + p**
**body = body + " | "**
**body = body + k**
**body = body + " |\n"**
> fs.write_text path=`path` text=`body`
**out = **
| path | count |
|------|-------|
| `path` | `n` |
*out*
+57 -21
View File
@@ -3,6 +3,8 @@ title: ext/ai/智能体
description: 文档驱动智能体 v2 — 单步 / 多步 / 运行;技能编译经 ABI v2 agent 插件(情景 → 学习 → llm_free)。
导入 大模型:ext/ai/大模型.mq.md
import agent:ext/ai/agent.mq.md
import sandbox:ext/ai/sandbox.mq.md
import table:lib/table.mq.md
import json:lib/json.mq.md
导入 系统:lib/系统.mq.md
导入 自写回:lib/自写回.mq.md
@@ -86,13 +88,14 @@ import json:lib/json.mq.md
> 系统.退出 码=1
**id = > agent_alloc**
**h = > json.parse text={"_type":"智能体"}**
**h = > json.set map=`h` key="id" value=`id`**
**h = > json.set map=`h` key="model" value=`大模型`**
**h = > json.set map=`h` key="tools" value=`工具`**
`h` =
| _type | id | model | tools |
|-------|-----|-------|-------|
| 智能体 | `id` | `大模型` | `工具` |
1. `站立提示`
**h = > json.set map=`h` key="standing" value=`站立提示`**
**h = table.put in=`h` at="standing" value=`站立提示`**
2. *
**_ = 1**
@@ -106,23 +109,42 @@ import json:lib/json.mq.md
> agent_history_clear id=`id`
****
## 试跑
+ `路径`=None
+ `源码`=None
+ `根目录`=None
+ `写入路径`=None
+ `文本`=None
+ `提交`=真
+ `配置`=default
O1 可选沙盒试跑。转发 `agent.trial`(表风格返回;宿主树在成功 `提交` 前不变)。
*> agent.trial path=`路径` source=`源码` root=`根目录` write_path=`写入路径` text=`文本` commit=`提交` profile=`配置`*
## 单步
+ `任务`
+ `写回`=真
+ `流式`=假
+ `打印增量`=假
+ `沙盒`=假
+ `沙盒配置`=default
原子一轮:上下文 → 大模型 → 可选子任务调工具。返回 *map*(`status`、`task`、`decision`;成功含 `result`,可选 `tool` / `tool_result`;失败含 `error`)。默认(`写回=真`)将结果写入调用行下的命名槽 `ok` / `error`;传 `写回=假` 可关闭。
流式=真 时走 SSE;打印增量=真 边读边打印。返回 map 仍是最终 result 文本。
流式=真 时走 SSE;打印增量=真 边读边打印。返回 map 仍是最终 result 文本。`沙盒=真` 时结果附带沙盒配置行。
**ctx = > 构建单步上下文 智能体=`自` 任务=`任务`**
**id = > json.get value=`自` key="id"**
**model = > json.get value=`自` key="model"**
**tools = > json.get value=`自` key="tools"**
**user_turn = > json.parse text={"role":"user"}**
**user_turn = > json.set map=`user_turn` key="content" value=`任务`**
`user_turn` =
| role | content |
|------|---------|
| user | `任务` |
> agent_history_append id=`id` item=`user_turn`
1. `流式`
@@ -134,9 +156,11 @@ import json:lib/json.mq.md
**reply = > trim value=`reply`**
**decision = reply**
**tool_name = > 提取工具名 回复=`reply`**
**out = > json.parse text={"status":"ok"}**
**out = > json.set map=`out` key="task" value=`任务`**
**out = > json.set map=`out` key="decision" value=`decision`**
`out` =
| status | task | decision |
|--------|------|----------|
| ok | `任务` | `decision` |
1. `tool_name`
**tool_out = > 运行工具 工具=`tools` 名=`tool_name`**
@@ -144,8 +168,8 @@ import json:lib/json.mq.md
**deny = > split value=`tool_s` sep="Tool not allowed"**
**deny_n = > len value=`deny`**
1. `deny_n` > 1
**out = > json.set map=`out` key="status" value="error"**
**out = > json.set map=`out` key="error" value=`tool_s`**
**out = table.put in=`out` at="status" value="error"**
**out = table.put in=`out` at="error" value=`tool_s`**
**reply = tool_s**
2. *
**task_s = > json.stringify value=`任务`**
@@ -161,16 +185,25 @@ import json:lib/json.mq.md
**reply = > 大模型.流式结果 事件=`evs2`**
2. *
**reply = > model.运行 提示=`fp`**
**out = > json.set map=`out` key="tool" value=`tool_name`**
**out = > json.set map=`out` key="tool_result" value=`tool_s`**
**out = > json.set map=`out` key="result" value=`reply`**
**out = table.put in=`out` at="tool" value=`tool_name`**
**out = table.put in=`out` at="tool_result" value=`tool_s`**
**out = table.put in=`out` at="result" value=`reply`**
2. *
**out = > json.set map=`out` key="result" value=`reply`**
**out = table.put in=`out` at="result" value=`reply`**
**as_turn = > json.parse text={"role":"assistant"}**
**as_turn = > json.set map=`as_turn` key="content" value=`reply`**
`as_turn` =
| role | content |
|------|---------|
| assistant | `reply` |
> agent_history_append id=`id` item=`as_turn`
1. `沙盒`
**配置行 = sandbox.profile name=`沙盒配置`**
**out = table.put in=`out` at="sandbox_profile" value=`配置行`**
2. *
**_ = 1**
1. `写回`
**body = > json.stringify value=`out`**
**st = > json.get value=`out` key="status"**
@@ -296,14 +329,17 @@ Engineering First Information 门闩。转发 `agent.preflight`。
+ `强制`=False
+ `路由`=auto
+ `路由模型`=None
+ `沙盒`=False
+ `沙盒配置`=default
+ `工程预检`=True
v2+P4 主入口:自适应路由 → 执行 → 记录情景 → 尝试学习(技能 + 策略)。`路由`=`auto`|`marqdo`|`small-llm`|`llm`(`jev` 为 `small-llm` 兼容别名);决策器用 `路由模型`(任意 llm 句柄)。
v2+P4 主入口:自适应路由 → 执行 → 记录情景 → 尝试学习(技能 + 策略)。`路由`=`auto`|`marqdo`|`small-llm`|`llm`(`jev` 为 `small-llm` 兼容别名);决策器用 `路由模型`(任意 llm 句柄)。`沙盒=真` 走 CoW 试跑;`工程预检` 对应 `efi`。
**model = > json.get value=`self` key="model"**
**tools = > json.get value=`self` key="tools"**
**standing = > json.get value=`self` key="standing"**
**en = > agent.agent model=`model` tools=`tools` standing=`standing`**
*> en.run task=`任务` kb_dir=`知识库目录` memory_dir=`记忆目录` learn=`学习` verify=`校验` writeback=`写回` reuse=`复用` near_match=`近义命中` near_threshold=`近义阈值` promote=`晋升` improve_every=`改进周期` explore_n=`探索次数` max_rounds=`最大轮次` stream=`流式` echo=`打印增量` trace=`轨迹` force=`强制` router=`路由` router_model=`路由模型`*
*> en.run task=`任务` kb_dir=`知识库目录` memory_dir=`记忆目录` learn=`学习` verify=`校验` writeback=`写回` reuse=`复用` near_match=`近义命中` near_threshold=`近义阈值` promote=`晋升` improve_every=`改进周期` explore_n=`探索次数` max_rounds=`最大轮次` stream=`流式` echo=`打印增量` trace=`轨迹` force=`强制` router=`路由` router_model=`路由模型` sandbox=`沙盒` sandbox_profile=`沙盒配置` efi=`工程预检`*
## 路由
+ `任务`
+63
View File
@@ -0,0 +1,63 @@
---
title: ext/ai/沙盒
description: >-
AI 试跑 CoW 工作区:打开 / 运行 / 差异 / 提交(中文面,委托英文 sandbox 模块)。
import sandbox:ext/ai/sandbox.mq.md
---
## 打开
+ `根`=None
*> sandbox.open root=`根`*
## 写入
+ `id`
+ `路径`
+ `文本`
*> sandbox.write id=`id` path=`路径` text=`文本`*
## 读取
+ `id`
+ `路径`
*> sandbox.read id=`id` path=`路径`*
## 删除
+ `id`
+ `路径`
*> sandbox.remove id=`id` path=`路径`*
## 差异
+ `id`
*> sandbox.diff id=`id`*
## 提交
+ `id`
*> sandbox.commit id=`id`*
## 丢弃
+ `id`
*> sandbox.discard id=`id`*
## 运行
+ `id`
+ `源码`=None
+ `路径`=None
1. `源码`
*> sandbox.run id=`id` source=`源码`*
2. `路径`
*> sandbox.run id=`id` path=`路径`*
3. *
> print text=沙盒.运行: 需要 源码= 或 路径=
> sys.exit code=1
## 检查
+ `id`
*> sandbox.inspect id=`id`*
+170 -1
View File
@@ -511,11 +511,25 @@ fn write_skill_concept(
quality: i64,
hits: i64,
now: &str,
meta: &SkillMeta,
) -> Result<(), String> {
if let Some(p) = path.parent() {
fs::create_dir_all(p).map_err(|e| format!("mkdir: {e}"))?;
}
let title: String = goal.chars().take(48).collect();
let mut extra = String::new();
if let Some(ref s) = meta.applies_when {
extra.push_str(&format!("applies_when: {}\n", yaml_escape(s)));
}
if let Some(ref s) = meta.fixtures {
extra.push_str(&format!("fixtures: {}\n", yaml_escape(s)));
}
if let Some(ref s) = meta.sandbox_profile {
extra.push_str(&format!("sandbox_profile: {}\n", yaml_escape(s)));
}
if let Some(ref s) = meta.proof {
extra.push_str(&format!("proof: {}\n", yaml_escape(s)));
}
let body = format!(
r#"---
type: Marqdo Agent Skill
@@ -526,7 +540,7 @@ status: {status}
llm_free: {llm_free}
quality: {quality}
hits: {hits}
generated:
{extra}generated:
by: marqdo-agent/plan
at: {now}
verified:
@@ -545,11 +559,29 @@ Prefer spawning `resource` over re-planning.
llm_free = if llm_free { "true" } else { "false" },
quality = quality,
hits = hits,
extra = extra,
now = now,
);
fs::write(path, body).map_err(|e| format!("write skill: {e}"))
}
#[derive(Default)]
struct SkillMeta {
applies_when: Option<String>,
fixtures: Option<String>,
sandbox_profile: Option<String>,
proof: Option<String>,
}
fn skill_meta_from_src(src: &str) -> SkillMeta {
SkillMeta {
applies_when: extract_fm_field(src, "applies_when"),
fixtures: extract_fm_field(src, "fixtures"),
sandbox_profile: extract_fm_field(src, "sandbox_profile"),
proof: extract_fm_field(src, "proof"),
}
}
fn refresh_index(root: &Path, slug: &str, goal: &str) -> Result<(), String> {
let index = root.join("index.md");
let line = format!(
@@ -585,6 +617,10 @@ fn arg_text<'a>(args: &'a Value, key: &str) -> Result<&'a str, String> {
.ok_or_else(|| format!("missing text `{key}`"))
}
fn opt_text<'a>(args: &'a Value, key: &str) -> Option<&'a str> {
args.get(key).and_then(|v| v.as_str()).filter(|s| !s.is_empty())
}
fn opt_tools(args: &Value) -> Option<&Value> {
args.get("tools").filter(|t| !t.is_null())
}
@@ -1107,6 +1143,54 @@ pub fn kb_record_hit(args: &Value) -> Result<Value, String> {
}))
}
/// Run workbook (or FM `proof:` / `fixtures:` path) inside CoW sandbox; discard after.
/// O2: llm_free promote requires this proof to succeed (unless `require_proof=false`).
fn sandbox_proof(workbook: &Path, proof_rel: Option<&str>) -> Result<(bool, String), String> {
let root = workbook
.parent()
.ok_or_else(|| "sandbox_proof: workbook has no parent".to_string())?;
// Always pass an absolute root so host `sandbox_open` does not re-join cwd.
let root_abs = if root.is_absolute() {
root.to_path_buf()
} else {
fs::canonicalize(root).unwrap_or_else(|_| {
std::env::current_dir()
.unwrap_or_else(|_| PathBuf::from("."))
.join(root)
})
};
let root_abs = fs::canonicalize(&root_abs).unwrap_or(root_abs);
let default_name = workbook
.file_name()
.and_then(|s| s.to_str())
.ok_or_else(|| "sandbox_proof: bad workbook name".to_string())?;
let path = proof_rel.unwrap_or(default_name);
let open = crate::host_query_json_args(
"sandbox_open",
&json!({ "root": root_abs.to_string_lossy() }),
)?;
let id = open
.get("id")
.and_then(|v| v.as_str())
.ok_or_else(|| "sandbox_proof: open missing id".to_string())?
.to_string();
let run = crate::host_query_json_args("sandbox_run", &json!({ "id": id, "path": path }));
let _ = crate::host_query_json_args("sandbox_discard", &json!({ "id": id }));
match run {
Ok(r) => {
let ok = r.get("ok").and_then(|v| v.as_bool()).unwrap_or(false);
let detail = r
.get("stderr")
.and_then(|v| v.as_str())
.or_else(|| r.get("stdout").and_then(|v| v.as_str()))
.unwrap_or("")
.to_string();
Ok((ok, detail))
}
Err(e) => Ok((false, e)),
}
}
pub fn kb_promote(args: &Value) -> Result<Value, String> {
let kb = arg_text(args, "kb_dir")?;
let goal = arg_text(args, "goal")?;
@@ -1119,6 +1203,26 @@ pub fn kb_promote(args: &Value) -> Result<Value, String> {
let src_path = resolve_path(workbook);
let src = fs::read_to_string(&src_path)
.map_err(|e| format!("kb_promote read workbook: {e}"))?;
let require_proof = args
.get("require_proof")
.and_then(|v| v.as_bool())
.unwrap_or(true);
let proof_rel = extract_fm_field(&src, "proof")
.or_else(|| extract_fm_field(&src, "fixtures"));
if require_proof {
let (ok, detail) = sandbox_proof(&src_path, proof_rel.as_deref())?;
if !ok {
return Ok(json!({
"promoted": false,
"reason": "sandbox_proof_failed",
"detail": detail,
"sig": sig,
"slug": slug,
"workbook": workbook,
"proof": proof_rel.unwrap_or_else(|| src_path.file_name().and_then(|s| s.to_str()).unwrap_or("").to_string()),
}));
}
}
let (llm_free, quality) = llm_free_score(&src);
let status = if llm_free { "stable" } else { "candidate" };
@@ -1168,6 +1272,7 @@ pub fn kb_promote(args: &Value) -> Result<Value, String> {
&now,
&extra_aliases,
)?;
let meta = skill_meta_from_src(&src);
write_skill_concept(
&skill_path,
&sig,
@@ -1178,6 +1283,7 @@ pub fn kb_promote(args: &Value) -> Result<Value, String> {
quality,
prev_hits,
&now,
&meta,
)?;
refresh_index(&root, &slug, &goal_n)?;
@@ -1190,6 +1296,69 @@ pub fn kb_promote(args: &Value) -> Result<Value, String> {
"quality": quality,
"status": status,
"hits": prev_hits,
"proof_ok": true,
"applies_when": meta.applies_when,
"fixtures": meta.fixtures,
"sandbox_profile": meta.sandbox_profile,
}))
}
/// O5: mark a skill concept as `status: quarantine` and keep fixtures path in FM.
pub fn kb_quarantine(args: &Value) -> Result<Value, String> {
let kb = arg_text(args, "kb_dir")?;
let slug = arg_text(args, "slug")?;
let reason = opt_text(args, "reason").unwrap_or("sandbox_or_harness_failed");
let root = resolve_path(kb);
let skill_path = root.join("concepts/skills").join(format!("{slug}.md"));
if !skill_path.is_file() {
return Ok(json!({
"ok": false,
"reason": "skill_not_found",
"slug": slug,
"path": skill_path.to_string_lossy(),
}));
}
let src = fs::read_to_string(&skill_path)
.map_err(|e| format!("kb_quarantine read: {e}"))?;
let mut lines: Vec<String> = Vec::new();
let mut saw_status = false;
let mut in_fm = false;
for line in src.lines() {
let t = line.trim();
if t == "---" {
if !in_fm {
in_fm = true;
lines.push(line.to_string());
continue;
} else {
if !saw_status {
lines.push("status: quarantine".to_string());
}
lines.push(format!("quarantine_reason: {}", yaml_escape(reason)));
in_fm = false;
lines.push(line.to_string());
continue;
}
}
if in_fm && t.starts_with("status:") {
lines.push("status: quarantine".to_string());
saw_status = true;
continue;
}
if in_fm && t.starts_with("llm_free:") {
lines.push("llm_free: false".to_string());
continue;
}
lines.push(line.to_string());
}
let body = lines.join("\n") + "\n";
fs::write(&skill_path, body).map_err(|e| format!("kb_quarantine write: {e}"))?;
Ok(json!({
"ok": true,
"slug": slug,
"status": "quarantine",
"reason": reason,
"path": skill_path.to_string_lossy(),
}))
}
+39
View File
@@ -8,6 +8,7 @@ mod mcp_server;
mod memory;
mod resume;
mod route;
mod sandbox;
use std::collections::HashMap;
use std::ffi::{CStr, CString};
@@ -900,6 +901,7 @@ kb_ffi!(agent_corpus_search, corpus::corpus_search, "agent_corpus_search");
kb_ffi!(agent_mcp_fixture, corpus::mcp_fixture, "agent_mcp_fixture");
kb_ffi!(agent_kb_add_alias, kb::kb_add_alias, "agent_kb_add_alias");
kb_ffi!(agent_kb_promote, kb::kb_promote, "agent_kb_promote");
kb_ffi!(agent_kb_quarantine, kb::kb_quarantine, "agent_kb_quarantine");
kb_ffi!(agent_kb_record_hit, kb::kb_record_hit, "agent_kb_record_hit");
kb_ffi!(
agent_workbook_solidify,
@@ -1066,6 +1068,15 @@ pub unsafe extern "C" fn marqdo_plugin_init(host: *const MarqdoHostApi) -> c_int
{
return 1;
}
if register(
host,
"agent_kb_quarantine",
"kb_dir,slug",
agent_kb_quarantine,
) != 0
{
return 1;
}
if register(host, "agent_kb_record_hit", "kb_dir,goal", agent_kb_record_hit) != 0 {
return 1;
}
@@ -1180,6 +1191,34 @@ pub unsafe extern "C" fn marqdo_plugin_init(host: *const MarqdoHostApi) -> c_int
{
return 1;
}
// Optional JSON keys only — empty params CSV so omitted args are not required.
if register(host, "sandbox_open", "", sandbox::sandbox_open) != 0 {
return 1;
}
if register(host, "sandbox_write", "", sandbox::sandbox_write) != 0 {
return 1;
}
if register(host, "sandbox_read", "", sandbox::sandbox_read) != 0 {
return 1;
}
if register(host, "sandbox_remove", "", sandbox::sandbox_remove) != 0 {
return 1;
}
if register(host, "sandbox_diff", "", sandbox::sandbox_diff) != 0 {
return 1;
}
if register(host, "sandbox_commit", "", sandbox::sandbox_commit) != 0 {
return 1;
}
if register(host, "sandbox_discard", "", sandbox::sandbox_discard) != 0 {
return 1;
}
if register(host, "sandbox_inspect", "", sandbox::sandbox_inspect) != 0 {
return 1;
}
if register(host, "sandbox_run", "", sandbox::sandbox_run) != 0 {
return 1;
}
0
}
+336 -41
View File
@@ -1,9 +1,14 @@
//! Minimal MCP Server (stdio JSON-RPC) — tools map to `lib.member` via host `call_lib_path`.
//! Minimal MCP Server — stdio JSON-RPC + Streamable HTTP (spec 2025-03-26).
use serde_json::{json, Map, Value};
use std::ffi::CStr;
use std::io::{BufRead, BufReader, Write};
use std::io::{BufRead, BufReader, Read, Write};
use std::net::{TcpListener, TcpStream};
use std::os::raw::{c_char, c_int};
use std::sync::atomic::{AtomicU64, Ordering};
use std::sync::Arc;
use std::thread;
use std::time::{SystemTime, UNIX_EPOCH};
use crate::{host_query_json_args, parse_args, set_err, set_out};
@@ -98,6 +103,40 @@ fn call_tool(tools: &[Value], name: &str, arguments: &Value) -> Result<Value, St
}))
}
/// Handle one JSON-RPC message. Returns `None` for notifications (no response body).
fn dispatch_rpc(name: &str, tools: &[Value], msg: &Value) -> Option<Value> {
let id = msg.get("id").cloned().unwrap_or(Value::Null);
let method = msg.get("method").and_then(|v| v.as_str()).unwrap_or("");
let params = msg.get("params").cloned().unwrap_or(json!({}));
if matches!(id, Value::Null) && method.starts_with("notifications/") {
return None;
}
Some(match method {
"initialize" => rpc_result(
&id,
json!({
"protocolVersion": "2025-03-26",
"capabilities": { "tools": {} },
"serverInfo": { "name": name, "version": "0.1.0" }
}),
),
"tools/list" | "list_tools" => rpc_result(&id, tools_list(tools)),
"tools/call" | "call_tool" => {
let tname = params.get("name").and_then(|v| v.as_str()).unwrap_or("");
let args = params.get("arguments").cloned().unwrap_or(json!({}));
match call_tool(tools, tname, &args) {
Ok(r) => rpc_result(&id, r),
Err(e) => rpc_error(&id, -32000, &e),
}
}
"ping" => rpc_result(&id, json!({})),
"" => rpc_error(&id, -32600, "missing method"),
other => rpc_error(&id, -32601, &format!("method not found: {other}")),
})
}
/// Blocking stdio MCP loop. `tools` is a JSON array of `{name, fn, description}`.
pub fn serve_stdio(name: &str, tools: Vec<Value>) -> Result<Value, String> {
let stdin = std::io::stdin();
@@ -106,41 +145,274 @@ pub fn serve_stdio(name: &str, tools: Vec<Value>) -> Result<Value, String> {
let Some(msg) = read_line_message(&mut reader)? else {
break;
};
let id = msg.get("id").cloned().unwrap_or(Value::Null);
let method = msg.get("method").and_then(|v| v.as_str()).unwrap_or("");
let params = msg.get("params").cloned().unwrap_or(json!({}));
if matches!(id, Value::Null) && method.starts_with("notifications/") {
continue;
if let Some(reply) = dispatch_rpc(name, &tools, &msg) {
write_message(&reply)?;
}
let reply_msg = match method {
"initialize" => rpc_result(
&id,
json!({
"protocolVersion": "2024-11-05",
"capabilities": { "tools": {} },
"serverInfo": { "name": name, "version": "0.1.0" }
}),
),
"tools/list" | "list_tools" => rpc_result(&id, tools_list(&tools)),
"tools/call" | "call_tool" => {
let tname = params.get("name").and_then(|v| v.as_str()).unwrap_or("");
let args = params.get("arguments").cloned().unwrap_or(json!({}));
match call_tool(&tools, tname, &args) {
Ok(r) => rpc_result(&id, r),
Err(e) => rpc_error(&id, -32000, &e),
}
}
"ping" => rpc_result(&id, json!({})),
"" => rpc_error(&id, -32600, "missing method"),
other => rpc_error(&id, -32601, &format!("method not found: {other}")),
};
write_message(&reply_msg)?;
}
Ok(json!({ "ok": true, "transport": "stdio" }))
}
fn new_session_id() -> String {
static SEQ: AtomicU64 = AtomicU64::new(1);
let n = SEQ.fetch_add(1, Ordering::Relaxed);
let ms = SystemTime::now()
.duration_since(UNIX_EPOCH)
.map(|d| d.as_millis())
.unwrap_or(0);
format!("mcp-{ms}-{n}")
}
fn read_http_request(
reader: &mut BufReader<TcpStream>,
) -> Result<(String, String, Vec<(String, String)>, Vec<u8>), String> {
let mut request_line = String::new();
reader
.read_line(&mut request_line)
.map_err(|e| format!("read request: {e}"))?;
let parts: Vec<&str> = request_line.trim().split_whitespace().collect();
if parts.len() < 2 {
return Err("bad request line".into());
}
let method = parts[0].to_string();
let path = parts[1].to_string();
let mut headers = Vec::new();
let mut content_length = 0usize;
loop {
let mut line = String::new();
reader
.read_line(&mut line)
.map_err(|e| format!("read header: {e}"))?;
let line = line.trim_end_matches(['\r', '\n']);
if line.is_empty() {
break;
}
if let Some((k, v)) = line.split_once(':') {
let key = k.trim().to_string();
let val = v.trim().to_string();
if key.eq_ignore_ascii_case("content-length") {
content_length = val.parse().unwrap_or(0);
}
headers.push((key, val));
}
}
let mut body = vec![0u8; content_length];
if content_length > 0 {
reader
.read_exact(&mut body)
.map_err(|e| format!("read body: {e}"))?;
}
Ok((method, path, headers, body))
}
fn header_get<'a>(headers: &'a [(String, String)], name: &str) -> Option<&'a str> {
headers
.iter()
.find(|(k, _)| k.eq_ignore_ascii_case(name))
.map(|(_, v)| v.as_str())
}
fn write_http(
stream: &mut TcpStream,
status: u16,
reason: &str,
content_type: &str,
body: &[u8],
extra: &[(&str, &str)],
) -> Result<(), String> {
let mut head = format!(
"HTTP/1.1 {status} {reason}\r\nContent-Type: {content_type}\r\nContent-Length: {}\r\nConnection: close\r\n",
body.len()
);
for (k, v) in extra {
head.push_str(k);
head.push_str(": ");
head.push_str(v);
head.push_str("\r\n");
}
head.push_str("\r\n");
stream
.write_all(head.as_bytes())
.map_err(|e| e.to_string())?;
stream.write_all(body).map_err(|e| e.to_string())?;
stream.flush().map_err(|e| e.to_string())?;
Ok(())
}
fn path_matches(req_path: &str, endpoint: &str) -> bool {
let base = req_path.split('?').next().unwrap_or(req_path);
base == endpoint || base.trim_end_matches('/') == endpoint.trim_end_matches('/')
}
fn handle_http_conn(
stream: TcpStream,
name: &str,
tools: &[Value],
endpoint: &str,
) -> Result<(), String> {
let mut reader = BufReader::new(stream);
let (method, path, headers, body) = read_http_request(&mut reader)?;
let mut stream = reader.into_inner();
if !path_matches(&path, endpoint) {
write_http(
&mut stream,
404,
"Not Found",
"application/json",
br#"{"error":"not_found"}"#,
&[],
)?;
return Ok(());
}
if method.eq_ignore_ascii_case("GET") {
// Optional SSE stream — empty keepalive then close (clients may 405).
let accept = header_get(&headers, "Accept").unwrap_or("");
if accept.contains("text/event-stream") {
let body = b": mcp streamable-http ready\n\n";
write_http(
&mut stream,
200,
"OK",
"text/event-stream",
body,
&[("Cache-Control", "no-cache")],
)?;
} else {
write_http(
&mut stream,
405,
"Method Not Allowed",
"application/json",
br#"{"error":"use_post_for_jsonrpc"}"#,
&[("Allow", "POST, GET")],
)?;
}
return Ok(());
}
if method.eq_ignore_ascii_case("DELETE") {
write_http(&mut stream, 200, "OK", "application/json", b"{}", &[])?;
return Ok(());
}
if !method.eq_ignore_ascii_case("POST") {
write_http(
&mut stream,
405,
"Method Not Allowed",
"application/json",
br#"{"error":"method_not_allowed"}"#,
&[("Allow", "POST, GET, DELETE")],
)?;
return Ok(());
}
let text = String::from_utf8_lossy(&body);
let msg: Value = serde_json::from_str(text.trim()).map_err(|e| format!("json-rpc: {e}"))?;
let is_init = msg.get("method").and_then(|v| v.as_str()) == Some("initialize");
let existing_sid = header_get(&headers, "Mcp-Session-Id")
.or_else(|| header_get(&headers, "MCP-Session-Id"))
.map(|s| s.to_string());
// Notifications → 202 Accepted, empty body.
if msg.get("id").is_none() || msg.get("id") == Some(&Value::Null) {
if let Some(m) = msg.get("method").and_then(|v| v.as_str()) {
if m.starts_with("notifications/") {
write_http(
&mut stream,
202,
"Accepted",
"application/json",
b"",
&[],
)?;
return Ok(());
}
}
}
let reply = match dispatch_rpc(name, tools, &msg) {
Some(r) => r,
None => {
write_http(
&mut stream,
202,
"Accepted",
"application/json",
b"",
&[],
)?;
return Ok(());
}
};
let body = serde_json::to_vec(&reply).map_err(|e| e.to_string())?;
let mut extras: Vec<(&str, String)> = Vec::new();
if is_init {
let sid = existing_sid.unwrap_or_else(new_session_id);
extras.push(("Mcp-Session-Id", sid));
} else if let Some(sid) = existing_sid {
extras.push(("Mcp-Session-Id", sid));
}
let extra_refs: Vec<(&str, &str)> = extras.iter().map(|(k, v)| (*k, v.as_str())).collect();
write_http(
&mut stream,
200,
"OK",
"application/json",
&body,
&extra_refs,
)?;
Ok(())
}
/// Streamable HTTP MCP (POST JSON-RPC + optional GET SSE). Spec 2025-03-26.
pub fn serve_http(
name: &str,
tools: Vec<Value>,
host: &str,
port: u16,
path: &str,
) -> Result<Value, String> {
let addr = format!("{host}:{port}");
let listener = TcpListener::bind(&addr).map_err(|e| format!("bind {addr}: {e}"))?;
let name = Arc::new(name.to_string());
let tools = Arc::new(tools);
let endpoint = Arc::new(if path.is_empty() {
"/mcp".to_string()
} else if path.starts_with('/') {
path.to_string()
} else {
format!("/{path}")
});
eprintln!(
"mcp streamable-http listening on http://{addr}{} (POST JSON-RPC)",
&*endpoint
);
for conn in listener.incoming() {
match conn {
Ok(stream) => {
let name = Arc::clone(&name);
let tools = Arc::clone(&tools);
let endpoint = Arc::clone(&endpoint);
thread::spawn(move || {
if let Err(e) = handle_http_conn(stream, &name, &tools, &endpoint) {
eprintln!("mcp http conn: {e}");
}
});
}
Err(e) => eprintln!("mcp http accept: {e}"),
}
}
Ok(json!({
"ok": true,
"transport": "http",
"host": host,
"port": port,
"path": *endpoint
}))
}
pub fn tools_from_value(v: &Value) -> Vec<Value> {
if let Some(arr) = v.as_array() {
return arr.clone();
@@ -180,13 +452,6 @@ pub unsafe extern "C" fn agent_mcp_serve(
.get("transport")
.and_then(|v| v.as_str())
.unwrap_or("stdio");
if transport != "stdio" {
set_err(
err_msg,
&format!("mcp serve: transport `{transport}` not implemented yet (use stdio)"),
);
return 1;
}
let name = args
.get("name")
.and_then(|v| v.as_str())
@@ -195,7 +460,38 @@ pub unsafe extern "C" fn agent_mcp_serve(
.get("tools")
.map(tools_from_value)
.unwrap_or_default();
match serve_stdio(name, tools) {
let result = match transport {
"stdio" => serve_stdio(name, tools),
"http" | "streamable-http" | "sse" => {
let host = args
.get("host")
.and_then(|v| v.as_str())
.unwrap_or("127.0.0.1");
let port = args
.get("port")
.and_then(|v| {
v.as_u64()
.or_else(|| v.as_i64().map(|n| n as u64))
.or_else(|| v.as_str().and_then(|s| s.parse().ok()))
})
.unwrap_or(7434) as u16;
let path = args
.get("path")
.and_then(|v| v.as_str())
.unwrap_or("/mcp");
serve_http(name, tools, host, port, path)
}
other => {
set_err(
err_msg,
&format!("mcp serve: transport `{other}` not supported (use stdio|http)"),
);
return 1;
}
};
match result {
Ok(v) => {
set_out(out_json, &v.to_string());
0
@@ -207,6 +503,5 @@ pub unsafe extern "C" fn agent_mcp_serve(
}
}
// Silence unused import if CStr unused in this module.
#[allow(dead_code)]
fn _unused(_: &CStr) {}
+26
View File
@@ -1038,10 +1038,17 @@ pub fn resolve_task(args: &Value) -> Result<Value, String> {
"tools": args.get("tools").cloned().unwrap_or(Value::Null),
}))?;
if hit.is_null() {
// Keep field set aligned with hit path so facades can safely read llm_free
// (qdagent / agent.run crashed on missing key when OKF is cold).
return Ok(json!({
"mode": "unknown",
"match": null,
"skill": null,
"resource": null,
"slug": null,
"status": null,
"llm_free": false,
"score": null,
"confidence": 0.0,
}));
}
@@ -1078,6 +1085,25 @@ pub fn resolve_task(args: &Value) -> Result<Value, String> {
mod tests {
use super::*;
#[test]
fn resolve_miss_includes_llm_free() {
let dir = std::env::temp_dir().join(format!(
"marqdo-resolve-miss-{}",
std::process::id()
));
let _ = fs::remove_dir_all(&dir);
fs::create_dir_all(&dir).unwrap();
let out = resolve_task(&json!({
"task": "no-such-skill-cold-miss",
"kb_dir": dir.to_string_lossy(),
"near_match": false,
}))
.unwrap();
assert_eq!(out.get("mode").and_then(|v| v.as_str()), Some("unknown"));
assert_eq!(out.get("llm_free").and_then(|v| v.as_bool()), Some(false));
let _ = fs::remove_dir_all(&dir);
}
#[test]
fn record_and_cluster_and_compile() {
let dir = std::env::temp_dir().join(format!(
+53
View File
@@ -0,0 +1,53 @@
//! Forward `sandbox_*` plugin calls to host_query (CoW lives in the host).
use super::{
host_query_json_args, parse_args, set_err, set_out,
};
use std::os::raw::c_char;
fn forward(
query_name: &str,
args_json: *const c_char,
out_json: *mut *mut c_char,
err_msg: *mut *mut c_char,
) -> i32 {
let v = match parse_args(args_json) {
Ok(v) => v,
Err(e) => {
set_err(err_msg, &format!("{query_name}: {e}"));
return 1;
}
};
match host_query_json_args(query_name, &v) {
Ok(j) => {
set_out(out_json, &j.to_string());
0
}
Err(e) => {
set_err(err_msg, &e);
1
}
}
}
macro_rules! sandbox_ffi {
($fn_name:ident, $query:expr) => {
pub unsafe extern "C" fn $fn_name(
args_json: *const c_char,
out_json: *mut *mut c_char,
err_msg: *mut *mut c_char,
) -> i32 {
forward($query, args_json, out_json, err_msg)
}
};
}
sandbox_ffi!(sandbox_open, "sandbox_open");
sandbox_ffi!(sandbox_write, "sandbox_write");
sandbox_ffi!(sandbox_read, "sandbox_read");
sandbox_ffi!(sandbox_remove, "sandbox_remove");
sandbox_ffi!(sandbox_diff, "sandbox_diff");
sandbox_ffi!(sandbox_commit, "sandbox_commit");
sandbox_ffi!(sandbox_discard, "sandbox_discard");
sandbox_ffi!(sandbox_inspect, "sandbox_inspect");
sandbox_ffi!(sandbox_run, "sandbox_run");
+42 -6
View File
@@ -245,7 +245,12 @@ fn chat_completions(args: &Value) -> Result<Value, String> {
let status = r.status();
if status != 200 {
let text = read_limited(r.into_reader())?;
return Err(format!("HTTP {status}: {text}"));
return Ok(json!({
"events": [{"type": "error", "message": format!("HTTP {status}: {text}")}],
"finish": "error",
"error": format!("HTTP {status}: {text}"),
"transport": "plugin",
}));
}
let events = parse_sse_events(BufReader::new(r.into_reader()), echo)?;
Ok(json!({
@@ -256,9 +261,19 @@ fn chat_completions(args: &Value) -> Result<Value, String> {
}
Err(ureq::Error::Status(code, r)) => {
let text = read_limited(r.into_reader()).unwrap_or_default();
Err(format!("HTTP {code}: {text}"))
Ok(json!({
"events": [{"type": "error", "message": format!("HTTP {code}: {text}")}],
"finish": "error",
"error": format!("HTTP {code}: {text}"),
"transport": "plugin",
}))
}
Err(e) => Err(format!("http POST sse {url}: {e}")),
Err(e) => Ok(json!({
"events": [{"type": "error", "message": format!("http POST sse {url}: {e}")}],
"finish": "error",
"error": format!("http POST sse {url}: {e}"),
"transport": "plugin",
})),
}
} else {
match req.send_string(&body) {
@@ -266,7 +281,14 @@ fn chat_completions(args: &Value) -> Result<Value, String> {
let status = r.status();
let text = read_limited(r.into_reader())?;
if status != 200 {
return Err(format!("HTTP {status}: {text}"));
return Ok(json!({
"text": "",
"usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0},
"finish": "error",
"error": format!("HTTP {status}: {text}"),
"tool_calls": [],
"transport": "plugin",
}));
}
let data: Value = serde_json::from_str(&text).map_err(|e| e.to_string())?;
let msg = data
@@ -299,9 +321,23 @@ fn chat_completions(args: &Value) -> Result<Value, String> {
}
Err(ureq::Error::Status(code, r)) => {
let text = read_limited(r.into_reader()).unwrap_or_default();
Err(format!("HTTP {code}: {text}"))
Ok(json!({
"text": "",
"usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0},
"finish": "error",
"error": format!("HTTP {code}: {text}"),
"tool_calls": [],
"transport": "plugin",
}))
}
Err(e) => Err(format!("http POST {url}: {e}")),
Err(e) => Ok(json!({
"text": "",
"usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0},
"finish": "error",
"error": format!("http POST {url}: {e}"),
"tool_calls": [],
"transport": "plugin",
})),
}
}
}
+64 -1
View File
@@ -3,6 +3,7 @@ package httpx
import (
"encoding/json"
"fmt"
"io"
"net/http"
"net/url"
"strings"
@@ -118,12 +119,74 @@ func jsonWrapInvokeValue(value any) (int, map[string]any, error) {
}
switch t := value.(type) {
case map[string]any:
return http.StatusOK, t, nil
code := http.StatusOK
if ok, has := t["ok"]; has {
if b, isBool := ok.(bool); isBool && !b {
if errMsg, _ := t["error"].(string); strings.EqualFold(errMsg, "unauthorized") ||
strings.Contains(strings.ToLower(errMsg), "unauthorized") {
code = http.StatusUnauthorized
}
}
}
return code, t, nil
default:
return http.StatusOK, map[string]any{"ok": true, "result": value}, nil
}
}
// writeSSE writes OpenAI-compatible (or generic) Server-Sent Events.
// value may be a string body, or a map with key "sse" / "body" / "chunks".
func writeSSE(w http.ResponseWriter, value any) {
w.Header().Set("Content-Type", "text/event-stream; charset=utf-8")
w.Header().Set("Cache-Control", "no-cache")
w.Header().Set("Connection", "keep-alive")
w.WriteHeader(http.StatusOK)
flusher, _ := w.(http.Flusher)
writeChunk := func(s string) {
if s == "" {
return
}
if !strings.HasSuffix(s, "\n") {
s += "\n"
}
_, _ = io.WriteString(w, s)
if flusher != nil {
flusher.Flush()
}
}
switch t := value.(type) {
case string:
writeChunk(t)
case map[string]any:
if s, ok := t["sse"].(string); ok {
writeChunk(s)
return
}
if s, ok := t["body"].(string); ok {
writeChunk(s)
return
}
if chunks, ok := t["chunks"].([]any); ok {
for _, c := range chunks {
writeChunk(fmt.Sprint(c))
}
return
}
b, err := json.Marshal(t)
if err != nil {
writeChunk("data: {\"error\":\"sse_encode\"}\n\n")
return
}
writeChunk("data: " + string(b) + "\n\n")
writeChunk("data: [DONE]\n\n")
default:
writeChunk("data: " + fmt.Sprint(value) + "\n\n")
writeChunk("data: [DONE]\n\n")
}
}
func invokeArgsFromPayload(payload map[string]any) map[string]any {
args := map[string]any{}
for k, v := range payload {
+29 -12
View File
@@ -17,8 +17,8 @@ type artifactRoute struct {
Path string
File string
Request string // json | form | none | query
Response string // json | text | html
Auth string // role name or "" / "none"
Response string // json | text | html | sse
Auth string // role name or "" / "none" | "bearer"
}
func parseArtifactRoute(v any) artifactRoute {
@@ -57,16 +57,21 @@ func (st *state) mountArtifactRoutes(mux *http.ServeMux) {
func (st *state) handleArtifact(w http.ResponseWriter, r *http.Request, route artifactRoute) {
if route.Auth != "" && !strings.EqualFold(route.Auth, "none") {
cookie := r.Header.Get("Cookie")
user := session.UsernameFromCookie(cookie)
if user == "" {
jsonErr(w, http.StatusUnauthorized, "auth required")
return
}
role := session.RoleFromCookie(cookie)
if role != "" && role != route.Auth && role != "admin" && role != "superadmin" {
jsonErr(w, http.StatusForbidden, "forbidden")
return
if strings.EqualFold(route.Auth, "bearer") || strings.EqualFold(route.Auth, "api_key") {
// Bearer / API-key auth is enforced inside the Endpoint `# main`
// (payload carries authorization / api_key). Do not require cookies.
} else {
cookie := r.Header.Get("Cookie")
user := session.UsernameFromCookie(cookie)
if user == "" {
jsonErr(w, http.StatusUnauthorized, "auth required")
return
}
role := session.RoleFromCookie(cookie)
if role != "" && role != route.Auth && role != "admin" && role != "superadmin" {
jsonErr(w, http.StatusForbidden, "forbidden")
return
}
}
}
@@ -115,6 +120,16 @@ func (st *state) handleArtifact(w http.ResponseWriter, r *http.Request, route ar
payload[name] = v
}
}
if auth := strings.TrimSpace(r.Header.Get("Authorization")); auth != "" {
if _, exists := payload["authorization"]; !exists {
payload["authorization"] = auth
}
}
if key := strings.TrimSpace(r.Header.Get("X-Api-Key")); key != "" {
if _, exists := payload["api_key"]; !exists {
payload["api_key"] = key
}
}
value, err := plugin.RunArtifact(route.File, payload)
if err != nil {
@@ -131,6 +146,8 @@ func (st *state) handleArtifact(w http.ResponseWriter, r *http.Request, route ar
w.Header().Set("Content-Type", "text/html; charset=utf-8")
w.WriteHeader(http.StatusOK)
_, _ = io.WriteString(w, fmt.Sprint(value))
case "sse", "event-stream", "text/event-stream":
writeSSE(w, value)
default:
code, body, err := jsonWrapInvokeValue(value)
if err != nil {
@@ -102,6 +102,18 @@ func (st *state) handleInvoke(w http.ResponseWriter, r *http.Request, route invo
}
}
// Bind common auth / client headers for OpenAI-compatible and API-key surfaces.
if auth := strings.TrimSpace(r.Header.Get("Authorization")); auth != "" {
if _, exists := payload["authorization"]; !exists {
payload["authorization"] = auth
}
}
if key := strings.TrimSpace(r.Header.Get("X-Api-Key")); key != "" {
if _, exists := payload["api_key"]; !exists {
payload["api_key"] = key
}
}
args := invokeArgsFromPayload(payload)
value, err := plugin.CallLibPath(route.FnPath, args)
if err != nil {
@@ -114,10 +126,22 @@ func (st *state) handleInvoke(w http.ResponseWriter, r *http.Request, route invo
strings.Contains(msg, "no active host") {
code = http.StatusBadRequest
}
if strings.Contains(msg, "unauthorized") || strings.Contains(msg, "Unauthorized") {
code = http.StatusUnauthorized
}
jsonErr(w, code, msg)
return
}
// OpenAI-compatible handlers may return `{sse: "..."}` to force event-stream
// even when the route's declared return mode is json.
if m, ok := value.(map[string]any); ok {
if s, ok := m["sse"].(string); ok && strings.TrimSpace(s) != "" {
writeSSE(w, m)
return
}
}
switch route.ReturnMode {
case "text", "html", "text/html":
if value == nil {
@@ -135,6 +159,8 @@ func (st *state) handleInvoke(w http.ResponseWriter, r *http.Request, route invo
w.Header().Set("Content-Type", ct)
w.WriteHeader(http.StatusOK)
_, _ = io.WriteString(w, text)
case "sse", "event-stream", "text/event-stream":
writeSSE(w, value)
case "status":
if value == nil {
jsonErr(w, http.StatusInternalServerError, "null result")
+1 -1
View File
@@ -5,7 +5,7 @@ description: Local browse and static export from the interpreter
# main
Milestone: `marqdo view` and `view output` (v1.4.0 document surface + inferred params).
Milestone: `marqdo view` and `view output` (v1.4.1 mid→mid annot leaders + document surface).
Local: `marqdo view public`
+1 -1
View File
@@ -5,7 +5,7 @@ description: 本地浏览与静态导出由解释器生成
# main
里程碑:`marqdo view` 与 `view output`(v1.4.0 文档面 + 推断形参)。
里程碑:`marqdo view` 与 `view output`(v1.4.1 mid→mid 标注连线 + 文档面)。
本地浏览:`marqdo view public`
+26
View File
@@ -39,6 +39,32 @@ run tests/ext/agent-zero-llm.mq.md
run tests/ext/agent-plan-preserve.mq.md
run examples/agent-okf-flywheel/index.mq.md
run examples/agent-pong/index.mq.md
run examples/agent-sandbox-okf/index.mq.md
run tests/ext/agent-sandbox-optin.mq.md
run tests/ext/agent-sandbox-promote.mq.md
# Skill Lift subset (O2 B3/E1): with-skill vs without — constitution / writeback / call_site markers.
skill_lift_pass=0
skill_lift_fail=0
assert_out_has() {
local f="$1"
local needle="$2"
local label="$3"
local out
out="$("$MQ" run "$f" 2>/dev/null || true)"
if printf '%s' "$out" | grep -q -- "$needle"; then
echo "SkillLift OK $label ($needle)"
skill_lift_pass=$((skill_lift_pass + 1))
else
echo "SkillLift MISS $label (need $needle)" >&2
skill_lift_fail=$((skill_lift_fail + 1))
fail=1
fi
}
assert_out_has tests/ext/agent-constitution-b0.mq.md "constitution-ok" "constitution"
assert_out_has tests/ext/agent-writeback-shadow.mq.md "writeback-shadow-ok" "writeback"
assert_out_has tests/ext/agent-dump-context.mq.md "dump-context-ok" "call_site"
echo "SkillLift summary: pass=$skill_lift_pass fail=$skill_lift_fail"
if [[ "${AGENT_HARNESS_LIVE:-}" == "1" ]]; then
export AGENT_LIVE=1
+45
View File
@@ -0,0 +1,45 @@
#!/usr/bin/env bash
# O4 — style gate: flag json.set dict chains and mixed EN/ZH API hints in .mq.md.
# Exit 0 when clean; 1 when findings (unless MQ_STYLE_LINT_SOFT=1).
set -euo pipefail
ROOT="$(cd "$(dirname "$0")/.." && pwd)"
cd "$ROOT"
TARGETS=("$@")
if [[ ${#TARGETS[@]} -eq 0 ]]; then
TARGETS=(ext/ai/sandbox.mq.md ext/ai/policies examples/agent-sandbox-okf)
fi
fail=0
findings=0
scan_file() {
local f="$1"
[[ -f "$f" ]] || return 0
# Long json.set pipelines (3+ consecutive) — style bug per marqdo-dev.
if awk '
/json\.set/ { c++; if (c >= 3) { found=1 } next }
{ c=0 }
END { exit found ? 0 : 1 }
' "$f"; then
echo "STYLE json.set-chain: $f" >&2
findings=$((findings + 1))
fail=1
fi
# Same file imports both EN agent and ZH 智能体 (or web/网页) — mixed surface.
if grep -qE 'import[[:space:]]+agent:|import[[:space:]]+web:' "$f" \
&& grep -qE '导入[[:space:]]+(智能体|网页):' "$f"; then
echo "STYLE mixed-EN/ZH-import: $f" >&2
findings=$((findings + 1))
fail=1
fi
}
while IFS= read -r -d '' f; do
scan_file "$f"
done < <(find "${TARGETS[@]}" -type f -name '*.mq.md' -print0 2>/dev/null)
echo "mq-style-lint: findings=$findings"
if [[ "${MQ_STYLE_LINT_SOFT:-}" == "1" ]]; then
exit 0
fi
exit "$fail"
+58
View File
@@ -0,0 +1,58 @@
#!/usr/bin/env bash
# O5 — offline standing / policy rewrite hint (GEPA-like, products stay repo files).
# Does NOT change the runtime agent path. Reads harness failure log + current files,
# writes a suggested patch under .marqdo/optimize/ for human review.
set -euo pipefail
ROOT="$(cd "$(dirname "$0")/.." && pwd)"
cd "$ROOT"
LOG="${1:-}"
STANDING="${2:-ext/ai/agent.mq.md}"
POLICY="${3:-ext/ai/policies/step-route.mq.md}"
OUT_DIR="${OUT_DIR:-.marqdo/optimize}"
mkdir -p "$OUT_DIR"
STAMP="$(date -u +%Y%m%dT%H%M%SZ)"
REPORT="$OUT_DIR/standing-policy-$STAMP.md"
{
echo "# Offline standing / policy optimize suggestion"
echo
echo "Generated: $STAMP"
echo
echo "This is a **review artifact**, not an automatic runtime rewrite."
echo "Strong models may rewrite \`standing\` / \`policies/*.mq.md\` offline;"
echo "commit only after harness green."
echo
echo "## Inputs"
echo
echo "- standing source: \`$STANDING\`"
echo "- policy source: \`$POLICY\`"
if [[ -n "$LOG" && -f "$LOG" ]]; then
echo "- harness log: \`$LOG\`"
echo
echo "## Harness failure excerpt"
echo
echo '```'
tail -n 80 "$LOG" || true
echo '```'
else
echo "- harness log: _(none — pass path as \$1)_"
fi
echo
echo "## Current policy table (copy)"
echo
if [[ -f "$POLICY" ]]; then
echo '```markdown'
cat "$POLICY"
echo '```'
else
echo "_missing $POLICY_"
fi
echo
echo "## Suggested next edits"
echo
echo "1. Keep step-class → backend as a **GFM table** (no json glue)."
echo "2. Quarantine failing skills (\`status: quarantine\`) before promoting again."
echo "3. Re-run \`scripts/agent-harness.sh\` and confirm Skill Lift summary."
} > "$REPORT"
echo "wrote $REPORT"
+6 -2
View File
@@ -32,7 +32,9 @@ Canonical design (repo): `doc/design/markdown-mapping.md`, `doc/design/keywords.
8. Prefer ending side-effect-only function bodies with `*None*` / `*无*` / whole-line `**` / `****` when later lines must belong to an outer function. Lone `---` / `***` are Markdown thematic breaks (skipped), not function end.
9. **`ext/**` never calls `host_*`.** Agent/OKF helpers are plugin names (`agent_kb_*`, …) after `plugin.load`. Do **not** add agent/OKF domain code to `src/host/` (core bloat). See `doc/design/ext-agent.md` §4.
10. **Browser (route C/D/E/F):** client logic is Marqdo on WASM; official bridge is host glue and **may** implement lists, routing, storage, WebSocket, canvas, file read, observers — authors **must not hand-write business JS**. Use `web.client_embed` / `data-mq-source-url` auto-mount. Client effects: prefer **GFM tables + `lib/browser`** ([marqdo-dev](../marqdo-dev/SKILL.md)).
11. **Code-as-documentation / no bag glue:** Prefer **GFM tables** for maps, lists, wire, commands. Prefer **`table.put` / named helpers** over `json.set` / `json.append` chains. `lib/json` is for parse/stringify/quote only — not a dict builder. Unreadable json pipelines are a style bug.
11. **Code-as-documentation / no bag glue:** Prefer **GFM tables** for maps, lists, wire, commands. Prefer **`table.put` / named helpers** over `json.set` / `json.append` chains. `lib/json` is for parse/stringify/quote only — not a dict builder. Unreadable json pipelines are a style bug. Read collections with **`[key](map)`**, not footnote indexes for new code.
12. **One language per file:** Do not mix EN and ZH API names in the same `.mq.md` (imports, calls, params). Pick English *or* Chinese mirrors for the whole program.
13. **Sandbox trial habit (Agent / generated `.mq.md`):** Prefer `agent.trial` / `run sandbox=True` / `sandbox.open`→`run`→`diff`→`commit` before promoting to the host tree. Host files must stay unchanged until a successful commit. Caps live in `sandbox.profile` GFM tables — do not invent a second dialect.
## Markup → meaning (v0.2)
@@ -162,7 +164,9 @@ Install: `marqdo ext add agent` (ZH: `智能体`). Native: `cargo build --releas
APIs: `# agent` / `# 智能体` → `## step` / `## plan`. Context: standing + task + **call site** + tools + budgeted source/skill + act protocol (`CALL:` / `READ:`).
Examples: [agent-pong](../../examples/agent-pong/) · [agent-okf-flywheel](../../examples/agent-okf-flywheel/). Harness: `scripts/agent-harness.sh`. Design: [ext-agent.md](../../doc/design/ext-agent.md) · Wave B: [agent-framework-2026-09.md](../../doc/research/agent-framework-2026-09.md). Dev LLM: copy [.env.example](../../.env.example) → `.env` (gitignored).
Examples: [agent-pong](../../examples/agent-pong/) · [agent-okf-flywheel](../../examples/agent-okf-flywheel/) · [agent-sandbox-okf](../../examples/agent-sandbox-okf/). Harness: `scripts/agent-harness.sh`. Design: [ext-agent.md](../../doc/design/ext-agent.md) · optimize O1–O5: [ext-ai-optimize.md](../../doc/roadmap/ext-ai-optimize.md). Dev LLM: copy [.env.example](../../.env.example) → `.env` (gitignored).
**Sandbox (O1):** `import sandbox:ext/ai/sandbox.mq.md` · `agent.trial` / `智能体.试跑` · `run`/`step` take `sandbox=` / `沙盒=` (default `False`). Profile caps are tables via `sandbox.profile`. Gold: `tests/ext/agent-sandbox-optin.mq.md`.
## Official extension: `ext/web` (Document / Endpoint / Resource)
+14 -1
View File
@@ -34,6 +34,9 @@ fn text_body(v: &Value) -> Result<&str, String> {
pub fn read_text(ctx: &HostContext, path: &Value) -> Result<Value, String> {
let p = ctx.resolve_path(text_path(path)?)?;
if let Some(hit) = crate::host::sandbox_fs::cow_read_text(ctx, &p)? {
return Ok(Value::Text(hit));
}
let s = fs::read_to_string(&p).map_err(|e| format!("read_text {}: {e}", p.display()))?;
Ok(Value::Text(s))
}
@@ -43,10 +46,14 @@ pub fn write_text(ctx: &HostContext, path: &Value, text: &Value) -> Result<Value
return Err("write_text denied by host policy".into());
}
let p = ctx.resolve_path(text_path(path)?)?;
let body = text_body(text)?;
if crate::host::sandbox_fs::cow_write_text(ctx, &p, body)? {
return Ok(Value::None);
}
if let Some(parent) = p.parent() {
fs::create_dir_all(parent).map_err(|e| format!("write_text mkdir: {e}"))?;
}
fs::write(&p, text_body(text)?).map_err(|e| format!("write_text {}: {e}", p.display()))?;
fs::write(&p, body).map_err(|e| format!("write_text {}: {e}", p.display()))?;
Ok(Value::None)
}
@@ -68,6 +75,9 @@ pub fn append_text(ctx: &HostContext, path: &Value, text: &Value) -> Result<Valu
pub fn exists(ctx: &HostContext, path: &Value) -> Result<Value, String> {
let p = ctx.resolve_path(text_path(path)?)?;
if let Some(hit) = crate::host::sandbox_fs::cow_exists(ctx, &p)? {
return Ok(Value::Bool(hit));
}
Ok(Value::Bool(p.exists()))
}
@@ -109,6 +119,9 @@ pub fn remove(ctx: &HostContext, path: &Value) -> Result<Value, String> {
return Err("remove denied by host policy".into());
}
let p = ctx.resolve_path(text_path(path)?)?;
if crate::host::sandbox_fs::cow_remove(ctx, &p)? {
return Ok(Value::None);
}
if p.is_dir() {
fs::remove_dir_all(&p).map_err(|e| format!("remove {}: {e}", p.display()))?;
} else {
+4
View File
@@ -40,6 +40,7 @@ pub mod subtask;
#[cfg(not(feature = "exec-host"))]
#[path = "subtask_stub.rs"]
pub mod subtask;
pub mod sandbox_fs;
mod sys;
pub mod writeback;
@@ -111,6 +112,8 @@ pub struct HostContext {
pub subtask_seq: u64,
/// Minimum log level (`lib/log`); default Info (20). See `host::log::Level`.
pub log_min_level: u8,
/// Active `ext/ai/sandbox` CoW session id (nested `sandbox_run`).
pub sandbox_session: Option<String>,
}
impl Default for HostContext {
@@ -135,6 +138,7 @@ impl Default for HostContext {
subtasks: HashMap::new(),
subtask_seq: 0,
log_min_level: 20, // Info
sandbox_session: None,
}
}
}
+26 -3
View File
@@ -120,6 +120,9 @@ pub fn http_get(
headers: Option<&Value>,
) -> Result<Value, String> {
if !ctx.allow_net() {
if ctx.sandbox_session.is_some() {
return Err("net unavailable in sandbox".into());
}
return Err("http_get denied by host policy".into());
}
let url = as_text(url, "url")?;
@@ -137,6 +140,9 @@ pub fn http_post(
headers: Option<&Value>,
) -> Result<Value, String> {
if !ctx.allow_net() {
if ctx.sandbox_session.is_some() {
return Err("net unavailable in sandbox".into());
}
return Err("http_post denied by host policy".into());
}
let url = as_text(url, "url")?;
@@ -146,8 +152,13 @@ pub fn http_post(
Some(v) => as_text(v, "content_type")?,
};
let hdrs = headers_from_value(headers)?;
let (status, resp) = http_exchange("POST", url, Some(body), Some(ct), &hdrs)?;
Ok(response_map(status, resp))
// Transport failures → status 0 + body message (soft). Callers that need
// hard errors can still treat status!=200 as failure; Agent/LLM surfaces
// map this into structured `{finish:error}` instead of aborting the VM.
match http_exchange("POST", url, Some(body), Some(ct), &hdrs) {
Ok((status, resp)) => Ok(response_map(status, resp)),
Err(e) => Ok(response_map(0, e)),
}
}
/// Generic request: method + url + optional body / content_type / headers.
@@ -161,6 +172,9 @@ pub fn http_request(
headers: Option<&Value>,
) -> Result<Value, String> {
if !ctx.allow_net() {
if ctx.sandbox_session.is_some() {
return Err("net unavailable in sandbox".into());
}
return Err("http_request denied by host policy".into());
}
let method = as_text(method, "method")?.to_uppercase();
@@ -838,7 +852,16 @@ pub fn http_post_sse(
),
]))
}
Err(e) => Err(format!("http POST sse {url}: {e}")),
Err(e) => Ok(Value::Map(vec![
("status".into(), Value::Int(0)),
(
"events".into(),
Value::List(vec![event_map(vec![
("type", Value::Text("error".into())),
("message", Value::Text(format!("http POST sse {url}: {e}"))),
])]),
),
])),
}
}
+13
View File
@@ -417,6 +417,16 @@ unsafe extern "C" fn host_query(
});
Ok(serde_json::json!({ "ok": true }))
})(),
name if name.starts_with("sandbox_") => (|| {
let raw = args_owned.as_deref().unwrap_or("{}");
let args: serde_json::Value = serde_json::from_str(if raw.trim().is_empty() {
"{}"
} else {
raw
})
.map_err(|e| format!("{name} args: {e}"))?;
crate::host::sandbox_fs::query(name, &args, ctx)
})(),
// ADR 0007: run a Document/Endpoint `.mq.md` for HTTP handlers.
"run_artifact" => (|| {
let raw = args_owned.as_deref().unwrap_or("{}");
@@ -472,6 +482,9 @@ unsafe extern "C" fn host_query(
pub fn load(ctx: &mut HostContext, path: &Value) -> Result<Value, String> {
if !ctx.allow_plugin() {
if ctx.sandbox_session.is_some() {
return Err("plugin.load unavailable in sandbox".into());
}
return Err("plugin_load denied by host policy".into());
}
let rel = as_text(path, "path")?;
+475
View File
@@ -0,0 +1,475 @@
//! Application-level CoW workspace for `ext/ai/sandbox` (via host_query).
//!
//! lower = real root (read-through); upper = session bytes; whiteout = logical delete.
//! Not a symlink forest. Symlink targets are not followed as isolation primitives.
use std::collections::{HashMap, HashSet};
use std::path::{Component, Path, PathBuf};
use std::sync::{Mutex, OnceLock};
use serde_json::json;
use crate::host::HostContext;
use crate::value::Value;
#[derive(Debug, Default)]
pub struct CowSession {
pub root: PathBuf,
/// Relative path (forward slashes) → file bytes.
pub upper: HashMap<String, Vec<u8>>,
pub whiteout: HashSet<String>,
}
static SESSIONS: OnceLock<Mutex<HashMap<String, CowSession>>> = OnceLock::new();
static SEQ: OnceLock<Mutex<u64>> = OnceLock::new();
fn sessions() -> &'static Mutex<HashMap<String, CowSession>> {
SESSIONS.get_or_init(|| Mutex::new(HashMap::new()))
}
fn next_id() -> String {
let mut g = SEQ
.get_or_init(|| Mutex::new(0))
.lock()
.unwrap_or_else(|e| e.into_inner());
*g = g.wrapping_add(1);
format!("sbx-{g}")
}
fn normalize_rel(rel: &str) -> Result<String, String> {
let rel = rel.trim().trim_start_matches('/');
if rel.is_empty() {
return Err("sandbox path must not be empty".into());
}
let p = Path::new(rel);
if p.is_absolute() {
return Err("sandbox path must be relative to session root".into());
}
let mut out = Vec::new();
for c in p.components() {
match c {
Component::Normal(s) => out.push(s.to_string_lossy().into_owned()),
Component::CurDir => {}
Component::ParentDir => {
return Err("sandbox path must not contain '..'".into());
}
_ => return Err(format!("sandbox path invalid component in `{rel}`")),
}
}
if out.is_empty() {
return Err("sandbox path must not be empty".into());
}
Ok(out.join("/"))
}
fn abs_under_root(root: &Path, rel: &str) -> Result<PathBuf, String> {
let rel_n = normalize_rel(rel)?;
let abs = root.join(Path::new(&rel_n));
let root_canon = root.canonicalize().unwrap_or_else(|_| root.to_path_buf());
// Parent may not exist yet for writes — check by prefix components.
let mut check = abs.clone();
loop {
if check == *root || check == root_canon {
return Ok(root.join(Path::new(&rel_n)));
}
if !check.pop() {
break;
}
if let Ok(c) = check.canonicalize() {
if crate::host::path_under_root(&c, &root_canon) || c == root_canon {
return Ok(root.join(Path::new(&rel_n)));
}
}
}
if crate::host::path_under_root(&abs, root) {
return Ok(abs);
}
Err(format!("path escapes sandbox root: {rel_n}"))
}
pub fn open(root: Option<&str>, ctx: &HostContext) -> Result<serde_json::Value, String> {
let root_pb = if let Some(r) = root.filter(|s| !s.is_empty()) {
let p = PathBuf::from(r);
if p.is_absolute() {
p
} else {
ctx.cwd.join(p)
}
} else if let Some(ep) = &ctx.entry_path {
ep.parent()
.unwrap_or(ep.as_path())
.to_path_buf()
} else {
ctx.cwd.clone()
};
let root_pb = if root_pb.is_absolute() {
root_pb
} else {
std::env::current_dir()
.unwrap_or_else(|_| PathBuf::from("."))
.join(root_pb)
};
if !root_pb.is_dir() {
return Err(format!(
"sandbox_open: root is not a directory: {}",
root_pb.display()
));
}
let id = next_id();
let sess = CowSession {
root: root_pb.clone(),
upper: HashMap::new(),
whiteout: HashSet::new(),
};
sessions()
.lock()
.unwrap_or_else(|e| e.into_inner())
.insert(id.clone(), sess);
Ok(json!({
"id": id,
"root": root_pb.to_string_lossy(),
}))
}
fn with_session_mut<F, T>(id: &str, f: F) -> Result<T, String>
where
F: FnOnce(&mut CowSession) -> Result<T, String>,
{
let mut map = sessions().lock().unwrap_or_else(|e| e.into_inner());
let sess = map
.get_mut(id)
.ok_or_else(|| format!("sandbox: unknown session `{id}`"))?;
f(sess)
}
fn with_session<F, T>(id: &str, f: F) -> Result<T, String>
where
F: FnOnce(&CowSession) -> Result<T, String>,
{
let map = sessions().lock().unwrap_or_else(|e| e.into_inner());
let sess = map
.get(id)
.ok_or_else(|| format!("sandbox: unknown session `{id}`"))?;
f(sess)
}
pub fn write(id: &str, path: &str, text: &str) -> Result<serde_json::Value, String> {
with_session_mut(id, |sess| {
let rel = normalize_rel(path)?;
let _ = abs_under_root(&sess.root, &rel)?;
sess.whiteout.remove(&rel);
sess.upper.insert(rel, text.as_bytes().to_vec());
Ok(json!({ "ok": true }))
})
}
pub fn read(id: &str, path: &str) -> Result<serde_json::Value, String> {
with_session(id, |sess| {
let rel = normalize_rel(path)?;
if sess.whiteout.contains(&rel) {
return Err(format!("sandbox_read: not found (whiteout): {rel}"));
}
if let Some(bytes) = sess.upper.get(&rel) {
let s = String::from_utf8(bytes.clone())
.map_err(|e| format!("sandbox_read: not utf-8: {e}"))?;
return Ok(json!(s));
}
let abs = abs_under_root(&sess.root, &rel)?;
let s = std::fs::read_to_string(&abs)
.map_err(|e| format!("sandbox_read {}: {e}", abs.display()))?;
Ok(json!(s))
})
}
pub fn remove(id: &str, path: &str) -> Result<serde_json::Value, String> {
with_session_mut(id, |sess| {
let rel = normalize_rel(path)?;
let _ = abs_under_root(&sess.root, &rel)?;
sess.upper.remove(&rel);
sess.whiteout.insert(rel);
Ok(json!({ "ok": true }))
})
}
pub fn diff(id: &str) -> Result<serde_json::Value, String> {
with_session(id, |sess| {
let mut items: Vec<serde_json::Value> = Vec::new();
for p in sess.upper.keys() {
items.push(json!({ "path": p, "kind": "write" }));
}
for p in &sess.whiteout {
items.push(json!({ "path": p, "kind": "remove" }));
}
items.sort_by(|a, b| {
let pa = a.get("path").and_then(|v| v.as_str()).unwrap_or("");
let pb = b.get("path").and_then(|v| v.as_str()).unwrap_or("");
pa.cmp(pb)
});
Ok(json!(items))
})
}
pub fn commit(id: &str) -> Result<serde_json::Value, String> {
with_session_mut(id, |sess| {
let mut applied = 0i64;
let writes: Vec<(String, Vec<u8>)> = sess
.upper
.iter()
.map(|(k, v)| (k.clone(), v.clone()))
.collect();
let removes: Vec<String> = sess.whiteout.iter().cloned().collect();
for (rel, bytes) in writes {
let abs = abs_under_root(&sess.root, &rel)?;
if let Some(parent) = abs.parent() {
std::fs::create_dir_all(parent)
.map_err(|e| format!("sandbox_commit mkdir: {e}"))?;
}
std::fs::write(&abs, &bytes)
.map_err(|e| format!("sandbox_commit write {}: {e}", abs.display()))?;
applied += 1;
}
for rel in removes {
let abs = abs_under_root(&sess.root, &rel)?;
if abs.exists() {
if abs.is_dir() {
std::fs::remove_dir_all(&abs)
.map_err(|e| format!("sandbox_commit remove {}: {e}", abs.display()))?;
} else {
std::fs::remove_file(&abs)
.map_err(|e| format!("sandbox_commit remove {}: {e}", abs.display()))?;
}
applied += 1;
}
}
sess.upper.clear();
sess.whiteout.clear();
Ok(json!({ "ok": true, "applied": applied }))
})
}
pub fn discard(id: &str) -> Result<serde_json::Value, String> {
let mut map = sessions().lock().unwrap_or_else(|e| e.into_inner());
if map.remove(id).is_none() {
return Err(format!("sandbox: unknown session `{id}`"));
}
Ok(json!({ "ok": true }))
}
pub fn inspect(id: &str) -> Result<serde_json::Value, String> {
with_session(id, |sess| {
Ok(json!({
"id": id,
"root": sess.root.to_string_lossy(),
"upper_count": sess.upper.len(),
"whiteout_count": sess.whiteout.len(),
"caps_default": {
"fs_write": true,
"exec": false,
"net": false,
"plugin": false
},
"notes": [
"ext/web listen/DB/WS unavailable in sandbox",
"plugin.load denied during sandbox.run",
"subtask path/code/lang unavailable in sandbox"
]
}))
})
}
/// Read file bytes through CoW for an active host sandbox session (lib/fs).
pub fn cow_read_text(ctx: &HostContext, abs: &Path) -> Result<Option<String>, String> {
let Some(id) = ctx.sandbox_session.as_deref() else {
return Ok(None);
};
with_session(id, |sess| {
let rel = rel_from_root(&sess.root, abs)?;
if sess.whiteout.contains(&rel) {
return Err(format!("read_text: not found (sandbox whiteout): {rel}"));
}
if let Some(bytes) = sess.upper.get(&rel) {
let s = String::from_utf8(bytes.clone())
.map_err(|e| format!("read_text: not utf-8: {e}"))?;
return Ok(Some(s));
}
Ok(None)
})
}
pub fn cow_write_text(ctx: &HostContext, abs: &Path, text: &str) -> Result<bool, String> {
let Some(id) = ctx.sandbox_session.as_deref() else {
return Ok(false);
};
with_session_mut(id, |sess| {
let rel = rel_from_root(&sess.root, abs)?;
sess.whiteout.remove(&rel);
sess.upper.insert(rel, text.as_bytes().to_vec());
Ok(true)
})
}
pub fn cow_remove(ctx: &HostContext, abs: &Path) -> Result<bool, String> {
let Some(id) = ctx.sandbox_session.as_deref() else {
return Ok(false);
};
with_session_mut(id, |sess| {
let rel = rel_from_root(&sess.root, abs)?;
sess.upper.remove(&rel);
sess.whiteout.insert(rel);
Ok(true)
})
}
pub fn cow_exists(ctx: &HostContext, abs: &Path) -> Result<Option<bool>, String> {
let Some(id) = ctx.sandbox_session.as_deref() else {
return Ok(None);
};
with_session(id, |sess| {
let rel = rel_from_root(&sess.root, abs)?;
if sess.whiteout.contains(&rel) {
return Ok(Some(false));
}
if sess.upper.contains_key(&rel) {
return Ok(Some(true));
}
Ok(None)
})
}
fn rel_from_root(root: &Path, abs: &Path) -> Result<String, String> {
let root_canon = root.canonicalize().unwrap_or_else(|_| root.to_path_buf());
let abs_canon = abs.canonicalize().unwrap_or_else(|_| abs.to_path_buf());
let rel = abs_canon
.strip_prefix(&root_canon)
.or_else(|_| abs.strip_prefix(root))
.map_err(|_| format!("path not under sandbox root: {}", abs.display()))?;
normalize_rel(&rel.to_string_lossy().replace('\\', "/"))
}
pub fn run(
id: &str,
source: Option<&str>,
path: Option<&str>,
) -> Result<serde_json::Value, String> {
let (root, body) = with_session(id, |sess| {
let body = if let Some(src) = source.filter(|s| !s.trim().is_empty()) {
src.to_string()
} else if let Some(p) = path.filter(|s| !s.is_empty()) {
let rel = normalize_rel(p)?;
if sess.whiteout.contains(&rel) {
return Err(format!("sandbox_run: path whiteout: {rel}"));
}
if let Some(bytes) = sess.upper.get(&rel) {
String::from_utf8(bytes.clone())
.map_err(|e| format!("sandbox_run: path not utf-8: {e}"))?
} else {
let abs = abs_under_root(&sess.root, &rel)?;
std::fs::read_to_string(&abs)
.map_err(|e| format!("sandbox_run read {}: {e}", abs.display()))?
}
} else {
return Err("sandbox_run requires `source` or `path`".into());
};
Ok((sess.root.clone(), body))
})?;
let opts = crate::RunOptions {
allow_fs_write: true,
allow_exec: false,
allow_net: false,
allow_plugin: false,
fs_root: Some(root),
sandbox_session: Some(id.to_string()),
sleep_limit_ms: Some(0),
..crate::RunOptions::default()
};
match crate::run_source(&body, &opts) {
Ok(cap) => {
let value = crate::host::json::value_to_json(&cap.value)
.unwrap_or(serde_json::Value::Null);
Ok(json!({
"ok": true,
"value": value,
"stdout": cap.stdout,
}))
}
Err(e) => {
let msg = e.to_string();
Ok(json!({
"ok": false,
"error": msg,
"value": null,
"stdout": "",
}))
}
}
}
/// Parse host_query JSON args helper used from plugin.rs.
pub fn query(name: &str, args: &serde_json::Value, ctx: &HostContext) -> Result<serde_json::Value, String> {
let id = || {
args.get("id")
.and_then(|v| v.as_str())
.ok_or_else(|| format!("{name}: missing `id`"))
};
match name {
"sandbox_open" => {
let root = args.get("root").and_then(|v| v.as_str());
open(root, ctx)
}
"sandbox_write" => {
let path = args
.get("path")
.and_then(|v| v.as_str())
.ok_or_else(|| "sandbox_write: missing `path`".to_string())?;
let text = args
.get("text")
.and_then(|v| v.as_str())
.ok_or_else(|| "sandbox_write: missing `text`".to_string())?;
write(id()?, path, text)
}
"sandbox_read" => {
let path = args
.get("path")
.and_then(|v| v.as_str())
.ok_or_else(|| "sandbox_read: missing `path`".to_string())?;
read(id()?, path)
}
"sandbox_remove" => {
let path = args
.get("path")
.and_then(|v| v.as_str())
.ok_or_else(|| "sandbox_remove: missing `path`".to_string())?;
remove(id()?, path)
}
"sandbox_diff" => diff(id()?),
"sandbox_commit" => commit(id()?),
"sandbox_discard" => discard(id()?),
"sandbox_inspect" => inspect(id()?),
"sandbox_run" => {
let source = args.get("source").and_then(|v| v.as_str());
let path = args.get("path").and_then(|v| v.as_str());
run(id()?, source, path)
}
other => Err(format!("sandbox query unknown `{other}`")),
}
}
#[allow(dead_code)]
pub fn unavailable(feature: &str) -> String {
format!("{feature} unavailable in sandbox")
}
/// Used by fs layer when session is active — map Value path through resolve then CoW.
pub fn try_read_value(ctx: &HostContext, path: &Value) -> Result<Option<Value>, String> {
let rel = match path {
Value::Text(s) => s.as_str(),
_ => return Ok(None),
};
let abs = ctx.resolve_path(rel)?;
match cow_read_text(ctx, &abs)? {
Some(s) => Ok(Some(Value::Text(s))),
None => Ok(None),
}
}
+3
View File
@@ -150,6 +150,9 @@ pub fn exec(
{
use std::process::Stdio;
if !ctx.allow_exec() {
if ctx.sandbox_session.is_some() {
return Err("exec unavailable in sandbox".into());
}
return Err("exec denied by host policy".into());
}
let cmd = as_text(cmd, "cmd")?;
+67 -2
View File
@@ -652,7 +652,26 @@ impl Interpreter {
&[("kind", "call"), ("callee", call.callee.as_str())],
);
}
let _ = self.eval_call(module, fun, env, call)?;
let stdout_before = self.captured_stdout.len();
let v = self.eval_call(module, fun, env, call)?;
if self.capture {
let (loop_id, iter) = self.loop_ctx();
self.record_stmt_event(StmtEvent {
line: span.line,
kind: StmtEventKind::Call,
name: Some(call.callee.clone()),
value_after: Some(call_event_brief(
&v,
&self.captured_stdout,
stdout_before,
&call.callee,
)),
arm_index: None,
iter,
call_id: self.current_call_id(),
loop_id,
});
}
Ok(None)
}
Stmt::Expr { value, span } => {
@@ -665,7 +684,30 @@ impl Interpreter {
&[("kind", "expr")],
);
}
let _ = self.eval_expr(module, fun, env, value)?;
let stdout_before = self.captured_stdout.len();
let v = self.eval_expr(module, fun, env, value)?;
if self.capture {
let (loop_id, iter) = self.loop_ctx();
let name = match value {
Expr::Call(c) => Some(c.callee.clone()),
_ => None,
};
self.record_stmt_event(StmtEvent {
line: span.line,
kind: StmtEventKind::Call,
name: name.clone(),
value_after: Some(call_event_brief(
&v,
&self.captured_stdout,
stdout_before,
name.as_deref().unwrap_or("expr"),
)),
arm_index: None,
iter,
call_id: self.current_call_id(),
loop_id,
});
}
Ok(None)
}
Stmt::Branch { arms, span } => {
@@ -1688,6 +1730,29 @@ fn brief_trace_val(v: &Value) -> String {
}
}
/// Prefer real result; else last stdout line emitted by the call; else callee name.
fn call_event_brief(v: &Value, stdout: &str, stdout_before: usize, callee: &str) -> String {
if !matches!(v, Value::None) {
return brief_trace_val(v);
}
if stdout.len() > stdout_before {
let delta = &stdout[stdout_before..];
let line = delta
.trim_end()
.lines()
.last()
.unwrap_or(delta.trim_end());
if !line.is_empty() {
return if line.chars().count() > 48 {
format!("{}…", line.chars().take(48).collect::<String>())
} else {
line.to_string()
};
}
}
callee.to_string()
}
fn stmt_span(stmt: &Stmt) -> Span {
match stmt {
Stmt::Assign { span, .. }
+11 -1
View File
@@ -92,6 +92,8 @@ pub struct RunOptions {
pub allow_fs_write: bool,
pub allow_exec: bool,
pub allow_net: bool,
/// Native `plugin.load` (default true for CLI; false inside `sandbox.run`).
pub allow_plugin: bool,
/// Exposed via `host_args` / `args`.
pub argv: Vec<String>,
/// Override filesystem sandbox root (default: source file's directory).
@@ -102,6 +104,8 @@ pub struct RunOptions {
pub emit_result: Option<std::path::PathBuf>,
/// CLI `--bind KEY=VALUE` pairs for Artifact Metadata Binding (`arg` + overlay).
pub binds: Vec<(String, String)>,
/// Attach CoW session for `ext/ai/sandbox` nested runs.
pub sandbox_session: Option<String>,
}
impl Default for RunOptions {
@@ -118,11 +122,13 @@ impl Default for RunOptions {
allow_fs_write: true,
allow_exec: true,
allow_net: true,
allow_plugin: true,
argv: Vec::new(),
fs_root: None,
sleep_limit_ms: None,
emit_result: None,
binds: Vec::new(),
sandbox_session: None,
}
}
}
@@ -154,7 +160,7 @@ impl RunOptions {
fs_write: self.allow_fs_write,
exec: self.allow_exec,
net: self.allow_net,
plugin: true,
plugin: self.allow_plugin,
}
}
}
@@ -289,6 +295,7 @@ pub fn run_file_capture(path: &Path, opts: &RunOptions) -> Result<RunCapture> {
host.set_entry_source(Some(path), &source);
host.argv = opts.argv.clone();
host.entry_metadata = module.metadata.clone();
host.sandbox_session = opts.sandbox_session.clone();
if let Some(root) = &opts.fs_root {
host.fs_root = Some(root.clone());
host.cwd = root.clone();
@@ -315,6 +322,7 @@ pub fn run_file_capture(path: &Path, opts: &RunOptions) -> Result<RunCapture> {
host.set_entry_source(Some(path), &source);
host.argv = opts.argv.clone();
host.entry_metadata = module.metadata.clone();
host.sandbox_session = opts.sandbox_session.clone();
if let Some(root) = &opts.fs_root {
host.fs_root = Some(root.clone());
host.cwd = root.clone();
@@ -351,6 +359,7 @@ pub fn run_source(source: &str, opts: &RunOptions) -> Result<RunCapture> {
host.set_entry_source(None, source);
host.argv = opts.argv.clone();
host.entry_metadata = module.metadata.clone();
host.sandbox_session = opts.sandbox_session.clone();
if let Some(root) = &opts.fs_root {
host.fs_root = Some(root.clone());
host.cwd = root.clone();
@@ -377,6 +386,7 @@ pub fn run_source(source: &str, opts: &RunOptions) -> Result<RunCapture> {
host.set_entry_source(None, source);
host.argv = opts.argv.clone();
host.entry_metadata = module.metadata.clone();
host.sandbox_session = opts.sandbox_session.clone();
if let Some(root) = &opts.fs_root {
host.fs_root = Some(root.clone());
host.cwd = root.clone();
+243 -7
View File
@@ -33,7 +33,7 @@ pub enum AnnotKind {
Return,
#[allow(dead_code)]
Call,
#[allow(dead_code)]
#[allow(dead_code)] // reserved for non-call BoldCode expr chips
Expr,
Branch,
Loop,
@@ -119,7 +119,9 @@ pub fn build_view_annots(module: &Module, source: &str, cap: &RunCapture) -> Vec
)
});
let key = (*line, name.clone(), ev.call_id, ev.loop_id, ev.iter);
let mut value = ev.value_after.clone().unwrap_or_else(|| "?".into());
let raw = ev.value_after.clone().unwrap_or_else(|| "?".into());
let (value, detail) = abbreviate_bind_value(&raw);
let mut value = value;
let count = assign_counts.get(&key).copied().unwrap_or(1);
if count > 1 {
value = format!("{value} · ×{count}");
@@ -132,7 +134,7 @@ pub fn build_view_annots(module: &Module, source: &str, cap: &RunCapture) -> Vec
anchor_byte_end: end,
label: name.clone(),
value: brief_str(&value),
detail: ev.value_after.clone().unwrap_or_default(),
detail,
arm_index: None,
call_id: Some(ev.call_id),
loop_id: ev.loop_id,
@@ -352,6 +354,62 @@ pub fn build_view_annots(module: &Module, source: &str, cap: &RunCapture) -> Vec
}
}
// Call / expr events → chips; zip with BoldCode spans when several share a line.
let mut calls_by_line: HashMap<u32, Vec<&crate::capture::StmtEvent>> = HashMap::new();
for ev in &cap.view_trace.events {
if ev.kind == StmtEventKind::Call {
calls_by_line.entry(ev.line).or_default().push(ev);
}
}
for (line, evs) in calls_by_line {
let text = lines
.get(line.saturating_sub(1) as usize)
.copied()
.unwrap_or("");
let bold_spans: Vec<(usize, usize)> = scan_prose_line(text)
.into_iter()
.filter_map(|bit| match bit {
ProseBit::BoldCode { .. } => {
let start = bit_at(&bit);
Some((start, prose_bit_end(text, &bit)))
}
_ => None,
})
.collect();
for (i, ev) in evs.iter().enumerate() {
// One BoldCode on the line (e.g. print across call frames): share it.
// Several BoldCodes (e.g. two greets in main): zip by index.
let (start, end) = if bold_spans.is_empty() {
(None, None)
} else if bold_spans.len() == 1 {
let (s, e) = bold_spans[0];
(Some(s), Some(e))
} else {
bold_spans
.get(i)
.copied()
.map(|(s, e)| (Some(s), Some(e)))
.unwrap_or((None, None))
};
let value = ev.value_after.clone().unwrap_or_default();
let label = ev.name.clone().unwrap_or_else(|| "call".into());
out.push(ViewAnnot {
id: next_id("c"),
kind: AnnotKind::Call,
anchor_line: line,
anchor_byte_start: start,
anchor_byte_end: end,
label,
value: brief_str(&value),
detail: value,
arm_index: None,
call_id: Some(ev.call_id),
loop_id: ev.loop_id,
iter: ev.iter,
});
}
}
// Call frames for picker (label = fn name).
for fr in &cap.view_trace.calls {
out.push(ViewAnnot {
@@ -370,9 +428,167 @@ pub fn build_view_annots(module: &Module, source: &str, cap: &RunCapture) -> Vec
});
}
// Param Decl sites in function body: bind chips from CallFrame args.
for fr in &cap.view_trace.calls {
let Some(fun) = find_fun_by_line(module, fr.fn_line) else {
continue;
};
let body_lines = fun_stmt_lines(fun);
for (pname, pval) in parse_args_brief_pairs(&fr.args_brief) {
for (i, text) in lines.iter().enumerate() {
let line_no = (i + 1) as u32;
if !body_lines.contains(&line_no) {
continue;
}
for bit in scan_prose_line(text) {
if let ProseBit::Decl { name, at, .. } = &bit {
if name != &pname {
continue;
}
let end = prose_bit_end(text, &bit);
out.push(ViewAnnot {
id: next_id("p"),
kind: AnnotKind::Bind,
anchor_line: line_no,
anchor_byte_start: Some(*at),
anchor_byte_end: Some(end),
label: pname.clone(),
value: brief_str(&pval),
detail: format!("{pname}={pval}"),
arm_index: None,
call_id: Some(fr.call_id),
loop_id: None,
iter: None,
});
}
}
}
}
}
// Same prose span across call frames shares one id so <mark data-annot-id>
// matches the visible chip after frame filter (else L-leaders find no anchor).
unify_annot_ids_by_span(&mut out);
out
}
fn unify_annot_ids_by_span(annots: &mut [ViewAnnot]) {
let mut first: HashMap<(u32, Option<usize>, Option<usize>, u8), String> = HashMap::new();
for a in annots.iter_mut() {
if matches!(a.kind, AnnotKind::CallFrame | AnnotKind::LoopFrame) {
continue;
}
if a.anchor_byte_start.is_none() {
continue;
}
let tag = match a.kind {
AnnotKind::Bind => 1u8,
AnnotKind::Return => 2,
AnnotKind::Call | AnnotKind::Expr => 3,
AnnotKind::Branch => 4,
AnnotKind::Loop => 5,
AnnotKind::Writeback => 6,
AnnotKind::CallFrame | AnnotKind::LoopFrame => 0,
};
let key = (
a.anchor_line,
a.anchor_byte_start,
a.anchor_byte_end,
tag,
);
if let Some(id) = first.get(&key) {
a.id = id.clone();
} else {
first.insert(key, a.id.clone());
}
}
}
fn abbreviate_bind_value(raw: &str) -> (String, String) {
if raw.contains("<svg") {
return ("svg".into(), "svg".into());
}
if raw.contains("```") {
return ("code".into(), "code".into());
}
if raw.chars().count() > 96 {
let short: String = raw.chars().take(48).collect();
return (format!("{short}…"), format!("{short}…"));
}
(raw.to_string(), raw.to_string())
}
fn parse_args_brief_pairs(brief: &str) -> Vec<(String, String)> {
let mut out = Vec::new();
for part in brief.split(", ") {
if let Some((k, v)) = part.split_once('=') {
if !k.is_empty() {
out.push((k.to_string(), v.to_string()));
}
}
}
out
}
fn find_fun_by_line(module: &Module, line: u32) -> Option<&Function> {
fn walk<'a>(fun: &'a Function, line: u32) -> Option<&'a Function> {
if fun.span.line == line {
return Some(fun);
}
for c in &fun.children {
if let Some(f) = walk(c, line) {
return Some(f);
}
}
None
}
for f in &module.functions {
if let Some(found) = walk(f, line) {
return Some(found);
}
}
None
}
fn fun_stmt_lines(fun: &Function) -> std::collections::HashSet<u32> {
let mut s = std::collections::HashSet::new();
fn walk_stmt(st: &Stmt, s: &mut std::collections::HashSet<u32>) {
let line = match st {
Stmt::Assign { span, .. }
| Stmt::Return { span, .. }
| Stmt::Call { span, .. }
| Stmt::Expr { span, .. }
| Stmt::Branch { span, .. }
| Stmt::While { span, .. }
| Stmt::ForEach { span, .. } => span.line,
};
s.insert(line);
match st {
Stmt::Branch { arms, .. } => {
for arm in arms {
for b in &arm.body {
walk_stmt(b, s);
}
}
}
Stmt::While { body, .. } | Stmt::ForEach { body, .. } => {
for b in body {
walk_stmt(b, s);
}
}
_ => {}
}
}
for st in &fun.body {
walk_stmt(st, &mut s);
}
for c in &fun.children {
s.extend(fun_stmt_lines(c));
}
s
}
/// Serialize ViewExecutionReport JSON for AI / page embed.
pub fn view_execution_report_json(path: &str, ok: bool, cap: &RunCapture) -> String {
use serde_json::{json, Map};
@@ -583,8 +799,10 @@ pub fn bit_kind_class(bit: &ProseBit) -> &'static str {
/// Render annotation band HTML for a set of annots (already filtered to one block).
///
/// Chips with `iter` default to the **last** iteration per `loop_id` (SSR); the loop
/// picker toggles `hidden` / `is-frame-hidden` so only the selected iter is shown.
/// SSR defaults to the **last** call_id and **last** iteration per loop_id (matches
/// pickers). Overflow (`max_visible`) is counted only among chips visible in that
/// default frame — so `who` on the last greet call is not buried behind +N.
/// Client `applyCallFrame` recomputes overflow after switching.
pub fn render_annot_band(annots: &[ViewAnnot], max_visible: usize) -> String {
let annots: Vec<&ViewAnnot> = annots
.iter()
@@ -603,17 +821,35 @@ pub fn render_annot_band(annots: &[ViewAnnot], max_visible: usize) -> String {
}
}
}
// Default call frame = max call_id among chips that carry one (last invocation).
let default_call_id: Option<u32> = annots.iter().filter_map(|a| a.call_id).max();
// Left-to-right by anchor so chips sit under their underlines (avoids
// crossed L-leaders that look like “missing” lines when crowded).
let mut ordered: Vec<&ViewAnnot> = annots;
ordered.sort_by(|a, b| {
let sa = a.anchor_byte_start.unwrap_or(usize::MAX);
let sb = b.anchor_byte_start.unwrap_or(usize::MAX);
sa.cmp(&sb).then_with(|| a.id.cmp(&b.id))
});
let mut s = String::from(r#"<div class="annot-band" data-annot-band>"#);
// Leaders live on the parent .doc-block (see initAnnotLeaders); keep a hook
// for keyed mode / a11y but geometry is computed against the block.
s.push_str(r#"<div class="annot-leaders" aria-hidden="true"></div>"#);
s.push_str(r#"<div class="annot-chips">"#);
let mut shown = 0usize;
let mut overflow_count = 0usize;
for a in &annots {
let frame_miss = a
for a in &ordered {
let iter_miss = a
.iter
.zip(default_iter.get(&a.loop_id))
.map(|(i, d)| i != *d)
.unwrap_or(false);
let call_miss = match (a.call_id, default_call_id) {
(Some(cid), Some(def)) if cid != def => true,
_ => false,
};
let frame_miss = iter_miss || call_miss;
let overflow = if frame_miss {
false
} else if shown >= max_visible {
+167 -5
View File
@@ -110,9 +110,33 @@ pub fn render_module_document(module: &Module, source: &str, annots: &[ViewAnnot
LineKind::Comment => {
let text = l.text.trim_end_matches(['\r', '\n']);
let trimmed = text.trim();
// Unbound ```lang fences stay Comment in lex — render as code cards
// (same chrome as bound foreign), never join into a prose <code> blob.
if crate::foreign::is_fence_opener(trimmed)
|| (trimmed.starts_with("```") && !crate::foreign::is_fence_closer(trimmed))
{
flush_para(&mut para, &mut out);
let (html, next) = render_narrative_fence(&lines, i);
out.push_str(&html);
i = next;
continue;
}
if crate::foreign::is_fence_closer(trimmed) {
// Orphan closer — skip.
flush_para(&mut para, &mut out);
i += 1;
continue;
}
let bits = scan_prose_line(text);
let is_stmt = stmt_lines.contains(&l.line_no);
let has_annots = by_line
.get(&l.line_no)
.map(|v| !v.is_empty())
.unwrap_or(false);
// Decl alone stays in Markdown prose (keep links). Open a marked
// line block when there is executable prose, a stmt, or chips.
let needs_annot = is_stmt
|| has_annots
|| bits.iter().any(|b| {
matches!(b, ProseBit::BoldCode { .. } | ProseBit::ItalicReturn { .. })
});
@@ -722,14 +746,24 @@ fn render_line_block(
let trimmed = text.trim_start();
if bits.is_empty() {
if let Some(rest) = trimmed.strip_prefix("> ") {
let aid = match_annot_for_line(annots)
.map(|a| a.id.as_str())
.unwrap_or("");
s.push_str(&format!(
r#"<div class="doc-line doc-call"><span class="mq-call-gt">&gt;</span> {}</div>"#,
escape(rest)
r#"<div class="doc-line doc-call"><span class="mq-call-gt">&gt;</span> <mark class="mq-anchor mq-code" data-line="{line}" data-annot-id="{aid}">{rest}</mark></div>"#,
line = line_no,
aid = escape(aid),
rest = escape(rest),
));
} else if let Some(rest) = trimmed.strip_prefix('>') {
let aid = match_annot_for_line(annots)
.map(|a| a.id.as_str())
.unwrap_or("");
s.push_str(&format!(
r#"<div class="doc-line doc-call"><span class="mq-call-gt">&gt;</span> {}</div>"#,
escape(rest.trim_start())
r#"<div class="doc-line doc-call"><span class="mq-call-gt">&gt;</span> <mark class="mq-anchor mq-code" data-line="{line}" data-annot-id="{aid}">{rest}</mark></div>"#,
line = line_no,
aid = escape(aid),
rest = escape(rest.trim_start()),
));
} else {
s.push_str(&format!(
@@ -768,7 +802,8 @@ fn render_line_block(
s
}
/// Huge SVG / fenced code dumps make annot bands unreadable — keep those out of chips.
/// Huge SVG / fenced code dumps — prefer abbreviated chips from annot builder;
/// only drop if detail is still an unreadable dump.
fn is_noisy_result_chip(a: &ViewAnnot) -> bool {
if a.kind != crate::view::annot::AnnotKind::Bind {
return false;
@@ -777,6 +812,49 @@ fn is_noisy_result_chip(a: &ViewAnnot) -> bool {
v.contains("<svg") || v.contains("```") || v.chars().count() > 96
}
fn match_annot_for_line<'a>(annots: &[&'a ViewAnnot]) -> Option<&'a ViewAnnot> {
annots.iter().copied().find(|a| {
matches!(
a.kind,
crate::view::annot::AnnotKind::Call
| crate::view::annot::AnnotKind::Bind
| crate::view::annot::AnnotKind::Return
)
})
}
/// Unbound Markdown fence → same visual card as bound foreign (no Run chrome).
fn render_narrative_fence(lines: &[ClassifiedLine], start: usize) -> (String, usize) {
let opener = lines[start].text.trim_end_matches(['\r', '\n']).trim();
let lang = crate::foreign::fence_lang(opener).unwrap_or_else(|| "text".into());
let line_no = lines[start].line_no;
let mut body: Vec<String> = Vec::new();
let mut j = start + 1;
while j < lines.len() {
let t = lines[j].text.trim_end_matches(['\r', '\n']);
if crate::foreign::is_fence_closer(t.trim()) {
j += 1;
break;
}
body.push(t.to_string());
j += 1;
}
let source = body.join("\n");
let html = format!(
r#"<div class="doc-block doc-exec" data-line="{line}"><div class="card code-card" data-lang="{lang}">
<div class="code-head">
<span class="badge">code</span>
<span class="code-lang">{lang}</span>
</div>
<pre class="code-body"><code class="language-{lang}">{src}</code></pre>
</div></div>"#,
line = line_no,
lang = escape(&lang),
src = escape(&source),
);
(html, j)
}
/// Render prose with markers stripped; show surface inner text only.
fn render_prose_with_marks(
line_no: u32,
@@ -1244,4 +1322,88 @@ mod tests {
"{html}"
);
}
#[test]
fn narrative_text_fence_renders_as_code_card() {
let src = r#"# main
```text
marqdo ext add web
marqdo ext add quantum
```
> print text=ok
"#;
let module = crate::parse::parse_source(src).unwrap();
let html = render_module_document(&module, src, &[]);
assert!(
html.contains(r#"data-lang="text""#) && html.contains("code-body"),
"text fence should be a code card: {html}"
);
assert!(
html.contains("marqdo ext add web\nmarqdo ext add quantum"),
"fence body newlines preserved: {html}"
);
assert!(
!html.contains("<code>text marqdo"),
"must not collapse fence into inline code: {html}"
);
}
#[test]
fn prose_calls_and_param_decl_get_chips() {
let src = r#"# main
Try **greet who="a"** and **greet who="b"**.
## greet
When we meet `who`, we **print text="Hi, `who`!"** and close with*None*.
"#;
let module = crate::parse::parse_source(src).unwrap();
let cap = crate::run_source(src, &crate::RunOptions::default()).unwrap();
let annots = build_view_annots(&module, src, &cap);
assert!(
annots
.iter()
.filter(|a| a.kind == crate::view::annot::AnnotKind::Call)
.count()
>= 3,
"expected greet×2 + print call events: {:?}",
cap.view_trace.events
);
assert!(
annots.iter().any(|a| {
a.kind == crate::view::annot::AnnotKind::Bind && a.label == "who"
}),
"param Decl bind missing: {annots:?}"
);
let html = render_module_document(&module, src, &annots);
assert!(html.contains("annot-call"), "{html}");
assert!(html.contains("annot-bind"), "{html}");
assert!(html.contains("Hi, a!") || html.contains("Hi, b!"), "{html}");
// Dual greet call sites each get a non-empty annot id.
let greet_ids: Vec<_> = html
.match_indices("mq-code")
.filter(|_| true)
.collect();
assert!(
html.matches(r#"data-annot-id="c"#).count() >= 2
|| html.contains(r#"data-annot-id="c"#),
"expected call annot ids on greet marks: {html}"
);
let _ = greet_ids;
// Last-frame who bind must not be buried in overflow (frame-aware density).
assert!(
!html.contains(r#"annot-bind annot-chip-overflow"#)
|| html.contains(r#"annot-label">who</span>"#),
"who bind should be visible for default (last) call frame: {html}"
);
// Prefer: no overflow on bind chips when last frame has ≤4 chips.
let bind_overflow = html.matches("annot-bind annot-chip-overflow").count();
assert_eq!(
bind_overflow, 0,
"default frame who/print/return must not overflow: {html}"
);
}
}
+209 -79
View File
@@ -382,6 +382,20 @@ li.doc-list-meta {
margin-left: -1.25rem;
}
.doc-block { position: relative; margin: 0.55rem 0 0.85rem; }
.doc-block > .annot-leaders {
position: absolute;
inset: 0;
pointer-events: none;
overflow: visible;
z-index: 1; /* below .annot-chips */
}
.doc-block > .annot-leaders svg {
position: absolute;
left: 0;
top: 0;
overflow: visible;
}
.doc-block.is-keyed > .annot-leaders { display: none; }
.doc-block.doc-exec[style*="--doc-indent"] {
margin-left: calc(var(--doc-indent, 1) * 1.5rem);
}
@@ -466,38 +480,26 @@ ul.doc-loop, ol.doc-branch {
.mq-anchor.is-focus { background: rgba(0,0,0,0.04); }
.annot-band {
position: relative;
margin: 0.35rem 0 0.15rem;
margin: 0.55rem 0 0.15rem;
min-height: 1.6rem;
padding-top: 0.55rem;
}
.annot-leaders {
position: absolute;
inset: 0;
pointer-events: none;
overflow: visible;
padding-top: 1.15rem; /* room for mid→mid leaders (rail + stub) */
z-index: 1;
}
/* Band-local leaders kept for keyed fallback; geometry uses .doc-block > .annot-leaders */
.annot-band > .annot-leaders { display: none; }
.annot-leader {
position: absolute;
box-sizing: border-box;
border-left: 1.5px solid var(--line-strong);
border-bottom: 1.5px solid var(--line-strong);
border-color: #8e8e93;
fill: none;
stroke-width: 1.5;
stroke-linecap: round;
stroke-linejoin: round;
pointer-events: none;
}
.annot-leader.annot-bind { border-color: #5ac8fa; }
.annot-leader.annot-return { border-color: #30d158; }
.annot-leader.annot-code,
.annot-leader.annot-call,
.annot-leader.annot-expr { border-color: #ff9f0a; }
.annot-leader.annot-branch { border-color: #bf5af2; }
.annot-leader.annot-loop { border-color: #64d2ff; }
.annot-leader.annot-writeback { border-color: #8e8e93; }
.annot-chips {
display: flex;
flex-wrap: wrap;
gap: 0.4rem 0.55rem;
gap: 0.45rem 0.7rem;
position: relative;
z-index: 1;
z-index: 3; /* chips above leaders */
}
.annot-chip {
display: inline-flex;
@@ -531,9 +533,9 @@ ul.doc-loop, ol.doc-branch {
.annot-chip.annot-branch { border-left-color: #bf5af2; }
.annot-chip.annot-loop { border-left-color: #64d2ff; }
.annot-chip.annot-writeback { border-left-color: #8e8e93; }
.annot-chip-overflow { display: none; }
.annot-band.is-expanded .annot-chip-overflow:not([hidden]):not(.is-frame-hidden) {
display: inline-flex;
.annot-chip.annot-chip-overflow { display: none !important; }
.annot-band.is-expanded .annot-chip.annot-chip-overflow:not([hidden]):not(.is-frame-hidden) {
display: inline-flex !important;
}
.annot-label { color: var(--muted); font-weight: 600; }
.annot-value { color: var(--ink); word-break: break-word; }
@@ -546,17 +548,17 @@ ul.doc-loop, ol.doc-branch {
padding: 0.22rem 0.45rem;
cursor: pointer;
}
.annot-band.is-keyed .annot-leaders { display: none; }
.annot-band.is-keyed .annot-chip::before {
.annot-band.is-keyed .annot-chip::before,
.doc-block.is-keyed .annot-chip::before {
content: attr(data-annot-key) " ";
color: var(--muted);
}
@media (max-width: 640px) {
.annot-band { padding-top: 0.25rem; }
.annot-leaders { display: none; }
.doc-block > .annot-leaders { display: none; }
}
@media (prefers-reduced-motion: reduce) {
.annot-leaders { display: none; }
.doc-block > .annot-leaders { display: none; }
}
.badge {
display: inline-block;
@@ -1340,6 +1342,16 @@ ul.doc-loop, ol.doc-branch {
align-items: center;
gap: 0.4rem;
}
.code-foot {
display: flex;
flex-wrap: wrap;
align-items: center;
gap: 0.45rem 0.65rem;
padding: 0.5rem 0.85rem;
border-top: 1px solid var(--line);
background: #fafafa;
}
.code-foot .code-actions { margin-left: auto; }
.code-cmd {
min-width: 9rem;
max-width: 16rem;
@@ -1376,6 +1388,7 @@ ul.doc-loop, ol.doc-branch {
background: #1e1e1e;
color: #d4d4d4;
font: 0.82rem/1.45 var(--mono);
min-height: 2.4rem;
}
.code-body code {
font: inherit;
@@ -1579,17 +1592,23 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
function initForeignCards() {{
var live = !!window.MARQDO_VIEW_LIVE;
document.querySelectorAll(".code-card").forEach(function (card) {{
var foot = card.querySelector(".code-foot");
var btn = card.querySelector(".code-run");
var cmd = card.querySelector(".code-cmd");
var out = card.querySelector(".code-out");
var body = card.querySelector(".code-body code");
if (!btn || !out || !body) return;
// Narrative ```text``` cards have no Run chrome — leave as-is.
if (!btn) return;
if (!live) {{
btn.disabled = true;
btn.title = "Run is available in live marqdo view";
if (cmd) cmd.disabled = true;
// Static: hide fake Run (view-static-interactive D5 / view-prose-annot-gaps P4).
if (foot) foot.hidden = true;
else {{
btn.hidden = true;
if (cmd) cmd.hidden = true;
}}
return;
}}
if (!out || !body) return;
btn.addEventListener("click", function () {{
btn.disabled = true;
out.hidden = false;
@@ -2280,6 +2299,45 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
}});
}}
var MAX_VISIBLE_ANNOTS = 4;
function setFrameHidden(el, hide) {{
el.hidden = !!hide;
if (hide) el.classList.add("is-frame-hidden");
else el.classList.remove("is-frame-hidden");
}}
/** After frame filter: overflow only among chips visible in the selected frame. */
function recomputeAnnotOverflow(scope) {{
var root = scope || document.getElementById("structure") || document;
root.querySelectorAll("[data-annot-band]").forEach(function (band) {{
band.classList.remove("is-expanded");
var chips = band.querySelectorAll(".annot-chip");
var visible = [];
chips.forEach(function (chip) {{
chip.classList.remove("annot-chip-overflow");
if (chip.hidden || chip.classList.contains("is-frame-hidden")) return;
visible.push(chip);
}});
var overflow = 0;
visible.forEach(function (chip, i) {{
if (i >= MAX_VISIBLE_ANNOTS) {{
chip.classList.add("annot-chip-overflow");
overflow++;
}}
}});
var more = band.querySelector("[data-annot-more]");
if (more) {{
if (overflow > 0) {{
more.hidden = false;
more.textContent = "+" + overflow;
}} else {{
more.hidden = true;
}}
}}
}});
}}
/** Source structure always stays visible; only chips + arm classes follow the frame. */
function initCallPickers() {{
document.querySelectorAll(".call-picker").forEach(function (sel) {{
@@ -2295,7 +2353,7 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
root.querySelectorAll(".annot-chip[data-call-id]").forEach(function (chip) {{
var cid = String(chip.getAttribute("data-call-id") || "");
if (!allowed[cid]) return;
chip.hidden = cid !== callId;
setFrameHidden(chip, cid !== callId);
}});
var trace = readViewTrace();
applyParamChipValues(fnName, callId, trace);
@@ -2318,6 +2376,7 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
}});
}});
}}
recomputeAnnotOverflow(root);
initAnnotLeaders();
}}
if (sel.dataset.bound !== "1") {{
@@ -2328,12 +2387,6 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
}});
}}
function setFrameHidden(el, hide) {{
el.hidden = !!hide;
if (hide) el.classList.add("is-frame-hidden");
else el.classList.remove("is-frame-hidden");
}}
function initLoopPickers() {{
document.querySelectorAll(".loop-picker").forEach(function (sel) {{
function apply() {{
@@ -2355,6 +2408,7 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
if (lid && loopId && String(lid) !== loopId) return;
setFrameHidden(chip, String(chipIter) !== iter);
}});
recomputeAnnotOverflow(scope);
initAnnotLeaders();
}}
if (sel.dataset.bound !== "1") {{
@@ -2365,6 +2419,26 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
}});
}}
function ensureBlockLeaders(block) {{
var leaders = block.querySelector(":scope > .annot-leaders");
if (!leaders) {{
leaders = document.createElement("div");
leaders.className = "annot-leaders";
leaders.setAttribute("aria-hidden", "true");
block.insertBefore(leaders, block.firstChild);
}}
return leaders;
}}
function leaderColor(kind) {{
if (kind === "bind") return "#5ac8fa";
if (kind === "return") return "#30d158";
if (kind === "call" || kind === "expr" || kind === "code") return "#ff9f0a";
if (kind === "branch") return "#bf5af2";
if (kind === "loop") return "#64d2ff";
return "#8e8e93";
}}
function initAnnotLeaders() {{
var bands = document.querySelectorAll("[data-annot-band]");
if (!bands.length) return;
@@ -2372,68 +2446,122 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
var narrow = window.matchMedia && window.matchMedia("(max-width: 640px)").matches;
function layoutBand(band) {{
var leaders = band.querySelector(".annot-leaders");
var chips = band.querySelectorAll(".annot-chip:not(.annot-chip-overflow):not([hidden]):not(.is-frame-hidden)");
if (!leaders) return;
var block = band.closest(".doc-block");
if (!block) return;
var leaders = ensureBlockLeaders(block);
leaders.innerHTML = "";
if (reduce || narrow || chips.length > 4) {{
band.classList.add("is-keyed");
chips.forEach(function (chip, i) {{
chip.setAttribute("data-annot-key", String(i + 1));
}});
return;
}}
band.classList.remove("is-keyed");
var bandRect = band.getBoundingClientRect();
var chipsEl = band.querySelector(".annot-chips");
var chips = Array.prototype.slice.call(
band.querySelectorAll(".annot-chip:not(.annot-chip-overflow):not([hidden]):not(.is-frame-hidden)")
);
// Pair chips with anchors; reorder DOM left→right under underlines so
// L-leaders do not cross (crowded reverse order looked like missing lines).
var items = [];
chips.forEach(function (chip) {{
if (chip.hidden || chip.classList.contains("is-frame-hidden")) return;
var id = chip.getAttribute("data-annot-id") || "";
var kind = chip.getAttribute("data-annot-kind") || "";
var anchor = id
? document.querySelector('.mq-anchor[data-annot-id="' + id + '"]')
? (block.querySelector('.mq-anchor[data-annot-id="' + id + '"]')
|| document.querySelector('.mq-anchor[data-annot-id="' + id + '"]'))
: null;
// Fallback: match mark kind on this block (frame chips may share a span).
if (!anchor) {{
var block = band.closest(".doc-block");
anchor = block ? block.querySelector(".mq-anchor") : null;
var cls = kind === "bind" ? "mq-decl"
: kind === "return" ? "mq-return"
: (kind === "call" || kind === "expr" || kind === "code") ? "mq-code"
: "";
if (cls) anchor = block.querySelector(".mq-anchor." + cls);
}}
if (!anchor) return;
var a = anchor.getBoundingClientRect();
var c = chip.getBoundingClientRect();
var x0 = a.left + a.width / 2 - bandRect.left;
var y0 = 0;
var x1 = c.left + Math.min(12, c.width / 2) - bandRect.left;
var y1 = c.top - bandRect.top + c.height / 2;
var left = Math.min(x0, x1);
var width = Math.max(2, Math.abs(x1 - x0));
var height = Math.max(2, y1 - y0);
var el = document.createElement("div");
el.className = "annot-leader annot-" + kind;
el.style.left = left + "px";
el.style.top = y0 + "px";
el.style.width = width + "px";
el.style.height = height + "px";
if (x1 < x0) {{
el.style.borderLeft = "none";
el.style.borderRight = "1.5px solid";
el.style.borderBottom = "1.5px solid";
}}
leaders.appendChild(el);
if (id) anchor.setAttribute("data-annot-id", id);
items.push({{
chip: chip,
anchor: anchor,
kind: kind,
x: anchor.getBoundingClientRect().left
}});
}});
items.sort(function (a, b) {{ return a.x - b.x; }});
if (chipsEl && items.length) {{
items.forEach(function (it) {{ chipsEl.appendChild(it.chip); }});
var more = band.querySelector("[data-annot-more]");
if (more) chipsEl.appendChild(more);
}}
if (reduce || narrow || items.length > MAX_VISIBLE_ANNOTS) {{
block.classList.add("is-keyed");
band.classList.add("is-keyed");
items.forEach(function (it, i) {{
it.chip.setAttribute("data-annot-key", String(i + 1));
}});
return;
}}
block.classList.remove("is-keyed");
band.classList.remove("is-keyed");
// Mid-bottom of mark → mid-top of chip: vertical, horizontal rail, short stub.
var blockRect = block.getBoundingClientRect();
var bw = Math.max(1, block.offsetWidth);
var bh = Math.max(1, block.offsetHeight);
var svgNS = "http://www.w3.org/2000/svg";
var svg = document.createElementNS(svgNS, "svg");
svg.setAttribute("width", String(bw));
svg.setAttribute("height", String(bh));
svg.setAttribute("viewBox", "0 0 " + bw + " " + bh);
svg.setAttribute("aria-hidden", "true");
items.forEach(function (it) {{
var a = it.anchor.getBoundingClientRect();
var c = it.chip.getBoundingClientRect();
var x0 = a.left + a.width / 2 - blockRect.left;
var y0 = a.bottom - blockRect.top + 1;
var x1 = c.left + c.width / 2 - blockRect.left;
var y1 = c.top - blockRect.top;
if (y1 < y0 + 8) {{
y1 = y0 + 12;
}}
// Short stub into chip top-center; rail sits above that.
var stub = Math.min(10, Math.max(6, (y1 - y0) * 0.28));
var yRail = y1 - stub;
if (yRail < y0 + 4) {{
yRail = (y0 + y1) / 2;
}}
var path = document.createElementNS(svgNS, "path");
path.setAttribute("class", "annot-leader annot-" + it.kind);
path.setAttribute("stroke", leaderColor(it.kind));
// Orthogonal: down from mark mid → across → short down to chip mid.
path.setAttribute(
"d",
"M " + x0.toFixed(1) + " " + y0.toFixed(1) +
" L " + x0.toFixed(1) + " " + yRail.toFixed(1) +
" L " + x1.toFixed(1) + " " + yRail.toFixed(1) +
" L " + x1.toFixed(1) + " " + y1.toFixed(1)
);
svg.appendChild(path);
}});
leaders.appendChild(svg);
}}
function layoutAll() {{
bands.forEach(layoutBand);
}}
layoutAll();
// Second pass after fonts/layout settle (first paint often has zero chip width).
requestAnimationFrame(function () {{ layoutAll(); }});
if (document.documentElement.dataset.annotLeadersRo === "1") return;
document.documentElement.dataset.annotLeadersRo = "1";
if (typeof ResizeObserver !== "undefined") {{
var ro = new ResizeObserver(function () {{ layoutAll(); }});
bands.forEach(function (b) {{ ro.observe(b); }});
bands.forEach(function (b) {{
ro.observe(b);
var blk = b.closest(".doc-block");
if (blk) ro.observe(blk);
}});
var surface = document.querySelector(".document-surface");
if (surface) ro.observe(surface);
}}
window.addEventListener("resize", layoutAll);
document.querySelectorAll("[data-annot-more]").forEach(function (btn) {{
if (btn.dataset.bound === "1") return;
btn.dataset.bound = "1";
btn.addEventListener("click", function () {{
var band = btn.closest("[data-annot-band]");
if (!band) return;
@@ -2443,6 +2571,8 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String {
}});
}});
document.querySelectorAll(".annot-chip").forEach(function (chip) {{
if (chip.dataset.focusBound === "1") return;
chip.dataset.focusBound = "1";
chip.addEventListener("mouseenter", function () {{
var id = chip.getAttribute("data-annot-id");
if (!id) return;
+7 -2
View File
@@ -1259,12 +1259,14 @@ fn render_stmt(
<span class="badge">code</span>
<code class="expr">`{name}` =</code>
<span class="code-lang">{lang}</span>
</div>
<pre class="code-body"><code class="language-{lang}">{src}</code></pre>
<div class="code-foot">
<div class="code-actions">
<input type="text" class="code-cmd" placeholder="{placeholder}" spellcheck="false" autocomplete="off"/>
<button type="button" class="code-run">Run</button>
</div>
</div>
<pre class="code-body"><code class="language-{lang}">{src}</code></pre>
<pre class="code-out" hidden></pre>
</div>"#,
lang = lang,
@@ -1526,6 +1528,7 @@ pub(crate) fn render_formula_assign_card(name: &str, tex: &str) -> String {
}
/// Foreign code assign card for document surface embedding (single visual layer).
/// Head → body → foot(actions) so the source is never collapsed under the Run row.
pub(crate) fn render_code_assign_card(name: &str, lang: &str, source: &str) -> String {
let placeholder = escape(&crate::foreign::default_cmd_display(lang));
format!(
@@ -1534,12 +1537,14 @@ pub(crate) fn render_code_assign_card(name: &str, lang: &str, source: &str) -> S
<span class="badge">code</span>
<span class="code-name">{name}</span>
<span class="code-lang">{lang}</span>
</div>
<pre class="code-body"><code class="language-{lang}">{src}</code></pre>
<div class="code-foot">
<div class="code-actions">
<input type="text" class="code-cmd" placeholder="{placeholder}" spellcheck="false" autocomplete="off"/>
<button type="button" class="code-run">Run</button>
</div>
</div>
<pre class="code-body"><code class="language-{lang}">{src}</code></pre>
<pre class="code-out" hidden></pre>
</div>"#,
lang = escape(lang),
+77
View File
@@ -0,0 +1,77 @@
---
title: agent / llm soft transport errors
description: >-
Regression (2026-10): unreachable OpenAI upstream must not abort the VM.
`llm.ask` / `agent.step` / `agent.run` return structured status=error.
import agent:ext/ai/agent.mq.md
import json:lib/json.mq.md
import llm:ext/ai/llm.mq.md
import plugin:lib/plugin.mq.md
import sys:lib/sys.mq.md
---
# main
> sys.env_set name="OPENAI_API_KEY" value="offline-transport-soft"
> sys.env_set name="OPENAI_BASE_URL" value="http://127.0.0.1:9"
**p = > plugin.native_path name="agent"**
1. `p`
> plugin.load path=`p`
2. *
> print text=no-agent-plugin
> sys.exit code=1
**model = > llm.llm**
**ask = > model.ask prompt="ping soft transport"**
**fin = > json.get value=`ask` key="finish"**
**aerr = > json.get value=`ask` key="error"**
1. `fin` == "error"
1. `aerr`
> print text=llm-ask-soft-ok
2. *
> print text=llm-ask-missing-error
> print text=`ask`
> sys.exit code=1
2. *
> print text=llm-ask-hard-or-ok-unexpected
> print text=`ask`
> sys.exit code=1
`tools` =
| 工具 |
|------|
**ag = > agent.agent model=`model` tools=`tools` standing="transport soft probe"**
**step = > `ag`.step task="unreachable upstream probe" writeback=False max_reads=1**
**sst = > json.get value=`step` key="status"**
**serr = > json.get value=`step` key="error"**
1. `sst` == "error"
1. `serr`
> print text=agent-step-soft-ok
2. *
> print text=agent-step-missing-error
> print text=`step`
> sys.exit code=1
2. *
> print text=agent-step-unexpected
> print text=`step`
> sys.exit code=1
**out = > `ag`.run task="unreachable upstream probe" efi=False learn=False verify=False writeback=False reuse=False near_match=False max_rounds=1**
**rst = > json.get value=`out` key="status"**
**rerr = > json.get value=`out` key="error"**
1. `rst` == "error"
1. `rerr`
> print text=agent-run-soft-ok
2. *
> print text=agent-run-missing-error
> print text=`out`
> sys.exit code=1
2. *
> print text=agent-run-unexpected
> print text=`out`
> sys.exit code=1
*"agent-llm-transport-soft-ok"*
+76
View File
@@ -0,0 +1,76 @@
---
title: agent.run cold OKF — resolve llm_free + self.preflight
description: >-
Regression for qdagent-facing bugs (2026-10):
(A) miss resolve must expose llm_free=false;
(B) `self`.preflight must resolve when agent is imported by a consumer.
Offline — dummy API key; no live network required for assertions A/B.
import agent:ext/ai/agent.mq.md
import fs:lib/fs.mq.md
import json:lib/json.mq.md
import llm:ext/ai/llm.mq.md
import plugin:lib/plugin.mq.md
import sys:lib/sys.mq.md
---
# main
> sys.env_set name="OPENAI_API_KEY" value="offline-cold-efi-dummy"
**p = > plugin.native_path name="agent"**
1. `p`
> plugin.load path=`p`
2. *
> print text=no-agent-plugin
> sys.exit code=1
**kb = ".marqdo/agent-kb-cold-efi"**
**ex = > fs.exists path=`kb`**
1. `ex`
> fs.remove_tree path=`kb`
2. *
**_ = 1**
> fs.make_dir path=`kb`
**resolved = > agent_resolve task="qdagent-cold-unknown-task-xyz-20261016" kb_dir=`kb`**
**lf = > json.get value=`resolved` key="llm_free"**
**mode = > json.get value=`resolved` key="mode"**
1. `lf` == False
> print text=resolve-miss-llm_free-ok
2. *
> print text=resolve-miss-llm_free-MISSING
> print text=`resolved`
> sys.exit code=1
1. `mode` == "unknown"
> print text=resolve-miss-mode-ok
2. *
> print text=resolve-miss-mode-unexpected
> sys.exit code=1
**model = > llm.llm**
`tools` =
| 工具 |
|------|
**ag = > agent.agent model=`model` tools=`tools` standing="cold efi preflight probe"**
**pf = > `ag`.preflight task="qdagent-cold-unknown-task-xyz-20261016" marqdo_dir=".marqdo"**
**st = > json.get value=`pf` key="status"**
1. `st`
> print text=self-preflight-ok
> print text=`st`
2. *
> print text=self-preflight-failed
> print text=`pf`
> sys.exit code=1
**via = > `ag`.resolve task="qdagent-cold-unknown-task-xyz-20261016" kb_dir=`kb`**
**via_lf = > json.get value=`via` key="llm_free"**
1. `via_lf` == False
> print text=method-resolve-llm_free-ok
2. *
> print text=method-resolve-llm_free-MISSING
> sys.exit code=1
*"agent-run-cold-efi-ok"*
+49
View File
@@ -0,0 +1,49 @@
---
title: agent trial sandbox opt-in — host unchanged until commit
description: >
O1: agent.trial write stays upper until commit=True.
Prefer link index over json.get.
import agent:ext/ai/agent.mq.md
import sandbox:ext/ai/sandbox.mq.md
import fs:lib/fs.mq.md
---
# main
Reset leftover workspace from prior runs.
**had = fs.exists path="_sbx_optin"**
1. `had`
> fs.remove path="_sbx_optin"
2. *
**_ = 0**
> fs.make_dir path="_sbx_optin"
Trial with `commit=False`: upper write must not touch the host tree.
**dry = agent.trial root="_sbx_optin" write_path="proof.txt" text=optin-ok commit=False**
**dry_ok = [ok](dry)**
**dry_committed = [committed](dry)**
> print text=`dry_ok`
> print text=`dry_committed`
**before = fs.exists path="_sbx_optin/proof.txt"**
> print text=`before`
Trial with `commit=True`: host sees the file after success.
**wet = agent.trial root="_sbx_optin" write_path="proof.txt" text=optin-ok commit=True**
**wet_ok = [ok](wet)**
**wet_committed = [committed](wet)**
> print text=`wet_ok`
> print text=`wet_committed`
**after = fs.read_text path="_sbx_optin/proof.txt"**
> print text=`after`
Profile row is attached (link-index readable).
**prof = [profile](wet)**
**pname = [name](prof)**
> print text=`pname`
+35
View File
@@ -0,0 +1,35 @@
---
title: promote requires sandbox proof (O2)
description: >
Good workbook promotes; broken workbook is rejected (sandbox_proof_failed).
Link-index reads; no json glue.
import plugin:lib/plugin.mq.md
import fs:lib/fs.mq.md
import sys:lib/sys.mq.md
---
# main
**p = plugin.native_path name="agent"**
1. `p`
> plugin.load path=`p`
2. *
> print text=no-plugin
> sys.exit code=1
**kb = "_sbx_promote_kb"**
**had = fs.exists path=`kb`**
1. `had`
> fs.remove path=`kb`
2. *
**_ = 0**
**prom = agent_kb_promote kb_dir=`kb` goal="sandbox promote good" workbook="sandbox-fixtures/promote-good.mq.md"**
**ok = [promoted](prom)**
> print text=`ok`
**rej = agent_kb_promote kb_dir=`kb` goal="sandbox promote bad" workbook="sandbox-fixtures/promote-bad.mq.md"**
**rej_ok = [promoted](rej)**
**reason = [reason](rej)**
> print text=`rej_ok`
> print text=`reason`
+34
View File
@@ -0,0 +1,34 @@
---
title: O5 quarantine skill status
description: Promote good skill then quarantine — status becomes quarantine.
import plugin:lib/plugin.mq.md
import fs:lib/fs.mq.md
import sys:lib/sys.mq.md
---
# main
**p = plugin.native_path name="agent"**
1. `p`
> plugin.load path=`p`
2. *
> print text=no-plugin
> sys.exit code=1
**kb = "_sbx_quar_kb"**
**had = fs.exists path=`kb`**
1. `had`
> fs.remove path=`kb`
2. *
**_ = 0**
**prom = agent_kb_promote kb_dir=`kb` goal="quarantine demo" workbook="sandbox-fixtures/promote-good.mq.md"**
**ok = [promoted](prom)**
**slug = [slug](prom)**
> print text=`ok`
**q = agent_kb_quarantine kb_dir=`kb` slug=`slug` reason="fixture_regression"**
**qok = [ok](q)**
**st = [status](q)**
> print text=`qok`
> print text=`st`
+19
View File
@@ -0,0 +1,19 @@
---
title: O3 step_route table backends
description: lookup → marqdo; net → llm; default → auto.
import agent:ext/ai/agent.mq.md
---
# main
**a = agent.step_route kind="lookup"**
**ba = [backend](a)**
> print text=`ba`
**b = agent.step_route kind="net"**
**bb = [backend](b)**
> print text=`bb`
**c = agent.step_route kind="default"**
**bc = [backend](c)**
> print text=`bc`
+60
View File
@@ -0,0 +1,60 @@
---
title: sandbox CoW — write stays upper until commit
description: >
Readable trial: seed on disk, mutate in session, host tree unchanged,
then commit / discard. Prefer link index over json.get.
import sandbox:ext/ai/sandbox.mq.md
import fs:lib/fs.mq.md
---
# main
Prepare a tiny workspace under this test directory (reset leftovers from prior runs).
**had = fs.exists path="_sbx_cow"**
1. `had`
> fs.remove path="_sbx_cow"
2. *
**_ = 0**
> fs.make_dir path="_sbx_cow"
> fs.write_text path="_sbx_cow/seed.txt" text=seed-lower
**sb = sandbox.open root="_sbx_cow"**
**id = [id](sb)**
Session edits (upper / whiteout only).
> sandbox.write id=`id` path="seed.txt" text=seed-upper
> sandbox.write id=`id` path="new.txt" text=brand-new
> sandbox.remove id=`id` path="seed.txt"
**pending = sandbox.diff id=`id`**
**n = len value=`pending`**
> print text=`n`
Host tree must still show the lower seed; `new.txt` must not exist yet.
**lower = fs.read_text path="_sbx_cow/seed.txt"**
> print text=`lower`
**ex_new = fs.exists path="_sbx_cow/new.txt"**
> print text=`ex_new`
Commit applies upper + whiteout.
> sandbox.commit id=`id`
**after = fs.read_text path="_sbx_cow/new.txt"**
> print text=`after`
**seed_gone = fs.exists path="_sbx_cow/seed.txt"**
> print text=`seed_gone`
Discard must not touch the host.
**sb2 = sandbox.open root="_sbx_cow"**
**id2 = [id](sb2)**
> sandbox.write id=`id2` path="new.txt" text=should-discard
> sandbox.discard id=`id2`
**kept = fs.read_text path="_sbx_cow/new.txt"**
> print text=`kept`
+34
View File
@@ -0,0 +1,34 @@
---
title: sandbox.run — plugin.load denied; inspect lists notes
description: >
Guest that loads a native plugin must fail with unavailable in sandbox.
No json.get chains — link index + a small expect table.
import sandbox:ext/ai/sandbox.mq.md
---
# main
**sb = sandbox.open root="sandbox-fixtures"**
**id = [id](sb)**
**ret = sandbox.run id=`id` path="guest-plugin.mq.md"**
**ok = [ok](ret)**
**err = [error](ret)**
> print text=`ok`
**parts = split value=`err` sep="unavailable in sandbox"**
**hits = len value=`parts`**
1. `hits` > 1
> print text=deny-ok
2. *
> print text=`err`
**info = sandbox.inspect id=`id`**
**notes = [notes](info)**
**nn = len value=`notes`**
1. `nn` > 0
> print text=inspect-ok
2. *
> print text=inspect-missing
> sandbox.discard id=`id`
@@ -0,0 +1,9 @@
---
import json:lib/json.mq.md
---
# main
**m = > json.parse text={"a":1}**
**x = > json.get value=`m` key="a"**
> print text=`x`
@@ -0,0 +1,8 @@
---
import plugin:lib/plugin.mq.md
---
# main
**p = > plugin.native_path name="agent"**
> plugin.load path=`p`
@@ -0,0 +1,8 @@
---
title: promote-bad
proof: promote-bad.mq.md
---
# main
> this_is_not_a_real_builtin_xyz
@@ -0,0 +1,12 @@
---
title: promote-ok
type: skill
applies_when: demo
fixtures: promote-good.mq.md
sandbox_profile: default
proof: promote-good.mq.md
---
# main
> print text=proof-ok
+22
View File
@@ -0,0 +1,22 @@
---
title: sandbox.run — core lib/json inside the trial
description: Guest fixture imports lib/json; print ok and guest stdout via link index.
import sandbox:ext/ai/sandbox.mq.md
---
# main
Open a session on the fixture directory (guest programs live here).
**sb = sandbox.open root="sandbox-fixtures"**
**id = [id](sb)**
Run the guest that imports `lib/json` and prints `1`.
**ret = sandbox.run id=`id` path="guest-json.mq.md"**
**ok = [ok](ret)**
**out = [stdout](ret)**
> print text=`ok`
> print text=`out`
> sandbox.discard id=`id`
+83
View File
@@ -2951,6 +2951,89 @@ fn ext_agent_zero_llm() {
);
}
#[test]
fn ext_ai_sandbox_cow() {
ensure_agent_plugin_built();
assert_out(
"tests/ext/sandbox-cow.mq.md",
"2
seed-lower
False
brand-new
False
brand-new",
);
}
#[test]
fn ext_ai_sandbox_run_lib() {
ensure_agent_plugin_built();
assert_out(
"tests/ext/sandbox-run-lib.mq.md",
"True
1",
);
}
#[test]
fn ext_ai_sandbox_deny() {
ensure_agent_plugin_built();
assert_out(
"tests/ext/sandbox-deny.mq.md",
"False
deny-ok
inspect-ok",
);
}
#[test]
fn ext_ai_agent_sandbox_optin() {
ensure_agent_plugin_built();
assert_out(
"tests/ext/agent-sandbox-optin.mq.md",
"True
False
False
True
True
optin-ok
default",
);
}
#[test]
fn ext_ai_agent_sandbox_promote() {
ensure_agent_plugin_built();
assert_out(
"tests/ext/agent-sandbox-promote.mq.md",
"True
False
sandbox_proof_failed",
);
}
#[test]
fn ext_ai_agent_step_route() {
ensure_agent_plugin_built();
assert_out(
"tests/ext/agent-step-route.mq.md",
"marqdo
llm
auto",
);
}
#[test]
fn ext_ai_agent_sandbox_quarantine() {
ensure_agent_plugin_built();
assert_out(
"tests/ext/agent-sandbox-quarantine.mq.md",
"True
True
quarantine",
);
}
#[test]
fn ext_llm_named_offline() {
let root = std::path::PathBuf::from(env!("CARGO_MANIFEST_DIR"));