Commit Graph
48 Commits
Author SHA1 Message Date
cflmy 1c90312c02 feat(ai): M10 把「接不了就申报」写进 agent 提示词(standing + how-to-act)
两处:
1) how-to-act 协议段(每轮都带):无法真正接通的capability 不许静默伪造、不许声称完成,
   改写成带状态的遮蔽,例如 [按关键词检索,接口未接,待开发](None);
   桩必须能跑并写清返回值(None 亦可);收尾用 labels.search 自查。
2) 默认 standing(p):追加一句"接不了就在描述里申报(待开发/模拟),别假装成功"。

为什么两处:standing 定基调(每轮开头),how-to-act 定动作(每轮),
只写一处会被长上下文冲淡或缺动机。
设计要点:给范式而非规则(可直接抄)、给返回值要求(对应 label.empty_payload)、
给自查动作(labels.search)、用英文与既有协议一致。

验证:marqdo check ext/ai/agent.mq.md(occlusion 模式)通过;
grep 确认两处文本落地;全量回归 587 passed / 0 failed(不改逻辑)。

文档:doc/design/change-m10-engineering-habits.md
2026-10-22 15:17:43 +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 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 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 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 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 fe15603a91 fix(llm): hoist ask result fields out of the return table
Table cells shadowed self/column names (model, usage), so [model](self)
and [total_tokens](usage) raised "footnote index needs list or map" and
blocked live agent.step / qdagent BUILD.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-28 22:08:48 +08:00
cflmyandCursor 56ab52356f feat(knowledge): land Engineering Knowledge Compiler (EKC) from 008
Give agents Engineering First Information before codegen: L0–L4 catalog,
reuse/preflight (REUSE→ADAPT→CREATE), graph, and CLI surfaces so large
repos stop reinventing existing capabilities.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-28 16:44:39 +08:00
cflmyandCursor 0ed57b6d54 feat(agent): ext APIs, goldens, and docs for MCP/resume/small-llm routing
Wire mcp_connect/list/call, resume_*, and demote Jev in author docs.
Offline goldens cover MCP client, resume, and small-llm/jev-alias route.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 20:26:55 +08:00
cflmyandCursor f1e82ce4e5 feat(llm): optional plugins/llm Transport ABI for chat completions
OpenAI transport prefers llm_chat_completions when the native plugin is
resolvable, and falls back to lib/net HTTP/SSE so CI stays green without the .so.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 20:26:55 +08:00
cflmyandCursor 2d412e3bba feat(llm): LLMResult.tool_calls from OpenAI-shaped responses
Parse choices[0].message.tool_calls into the ask Result; empty list when
absent. Offline gold covers tool_calls_from.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 20:07:59 +08:00
cflmyandCursor 46d78c6e15 feat(core): Phase 2 Metadata scope lift (no body ${})
Inject bound frontmatter keys into the entry env as ordinary variables.
Reject body ${env…}; bytecode entry prologue seeds the same locals.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 20:05:25 +08:00
cflmyandCursor 81a14f9b38 feat(llm): sys.module_source + entry prompt / meta offline paths
Expose entry source via sys.module_source; ask/stream fall back to
prompt_body of the entry document. Add meta-binding and entry-prompt golds.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 19:52:57 +08:00
cflmyandCursor 88dc6d5a54 feat(core): Artifact Metadata Binding (${env|arg|sys|secret})
Load-time resolve of frontmatter bindings into Module.metadata; Secret
masking; sys.meta / --bind; llm ctor prefers meta over env_get.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 19:51:24 +08:00
cflmyandCursor 4a2d89406b fix(agent): preserve llm_free resources under plan force
Stop plan from clobbering an existing resources workbook with an explore
skeleton. Add llm.prompt helpers and agent-plan-preserve gold.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 19:34:05 +08:00
cflmyandCursor d8d435fead feat(ai-stack): metrics, named handles, prompt docs, backends, zero-LLM
Finish ai-stack steps 1–5: model_turn usage, fast/reasoning, type:prompt
load, openai/ollama split, and agent-zero-llm gold. Fix compiled-skill
Marqdo syntax so llm_free await no longer falls through to plan+HTTP.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 19:30:25 +08:00
cflmyandCursor 1f8e61f3bf feat(llm): semantic ask/stream/collect surface (005)
Lift ext/ai/llm from OpenAI HTTP client to an intelligence primitive;
keep complete as compat. Add unified ai-stack roadmap linking agent + llm.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 13:35:38 +08:00
cflmyandCursor a9ee858f53 feat(agent): Skill Compilation + Adaptive Routing (v2/P4)
Episode → maybe_learn → llm_free skills, plus abstract router backends
(marqdo/small-llm/jev?/llm/auto) via plugins/agent; no core host changes.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-24 13:28:50 +08:00
cflmyandCursor 5c37d54070 refactor(ext): prefer link-index get in web and ai modules
Migrate footnote-style collection access to [key](base) (including nested
chains) in ext/web and ext/ai EN/ZH sources.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-14 11:55:09 +08:00
cflmyandCursor b5c07385e9 feat(markup): finish v0.3 stock migration and green gold
Migrate remaining .mq.md (tests/ext/public/examples), stop elevating local assigns to params, inject form_target in Go libweb, and skip live LLM gold without a real key.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-13 23:43:44 +08:00
cflmyandCursor fc543f1ca5 feat(web,agent): hosting H1–H4 — proxy, invoke, MCP stdio, cross-module db
Close GAP-01–05 with ext/web streaming reverse proxy and HTTP→## invoke,
transitive object method lookup, corpus_search plugin ensure, and agent MCP
server stdio so downstream apps can drop Python side relays.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-07 22:14:51 +08:00
cflmyandCursor a4c346d4a2 fix(agent,llm): avoid host_* writeback and tolerate SSE reasoning
Alias writeback as wb so the writeback= param does not shadow lib calls; ignore unknown stream event types so live SSE models keep working.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-07 20:52:44 +08:00
cflmyandCursor fe14d1f7e3 fix(agent): route writeback through lib instead of host_*
ext/** must not call host_*; use writeback.record / 自写回.写回 so the host-boundary gold passes.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-07 20:40:24 +08:00
cflmyandCursor e3e2cfa9d1 fix(agent): writeback param shadow + Wave B1 dump/harness
step/plan writeback=True shadowed lib/writeback; persist via host_writeback_record. Align ZH context builder, add dump_step_context, lock offline harness.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-04 11:24:50 +08:00
cflmyandCursor d1b4e62d4d feat(agent): Wave B0 constitution demos + harness
Lock code-as-doc / call_site / OKF as first-class: reorder step
context, ship agent-pong and okf-flywheel examples, offline harness,
and skill/docs refresh. Dev LLM template via .env.example only.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-04 11:17:25 +08:00
cflmyandCursor 9a8545b725 feat(agent): A4 corpus search and MCP fixture evidence tools
Add local keyword corpus_search and offline MCP fixture adapters as parent CALL tools; keep authority on workbook .mq.md.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-28 17:47:59 +08:00
cflmyandCursor ee3116513b feat(agent): A3 context budget and progressive READ disclose
Budget source/skill in step prompts with truncation markers; allow READ:source|skill deepen; parent READ:skill.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-28 17:26:16 +08:00
cflmyandCursor 729357b7c0 feat(agent): A2 process visibility — events always on plan result
Record decision/round/done without requiring stream; render process timeline on view plan cards; keep SSE gated on stream=True.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-28 17:06:59 +08:00
cflmyandCursor 07ab7d52c6 feat(agent): A1 OKF reuse flywheel — list_tasks curation and plan match
Enrich list_tasks for soft_match curation; expose match/score on plan hits; add offline plan-hit gold.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-28 16:53:27 +08:00
cflmyandCursor d0b184939f fix(agent): harden workbook patches against whole-file rewrites
Reject large whole-file FIND spans, narrow soft=True to miss-only, check
decompose CONTINUE patch counts, and unit-test workbook_solidify.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-28 16:22:39 +08:00
cflmyandCursor bd5555afaa Unify bare-identifier semantics in *…* / **…** (Python namespace)
Inside italic/bold code segments every bare word is a variable — including
call arguments and method-call receivers — so text literals must be quoted
and receivers no longer need backticks. A bare dotted callee `a.b` resolves
as a method call when `a` is a local variable, otherwise as a library path
(Python-style unified namespace), in both the tree-walk and bytecode backends.

Migrate ~70 tests plus ext/web, ext/quantum, ext/ai and the example projects
to the new idiom; update SKILL.md and the design docs; fix the web/quantum
example data model (db/index.mq.md). Unit and gold suites pass; only the 3
network-live tests fail offline.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-18 11:20:46 +08:00
cflmyandCursor 1bf414c745 Harden plan parent: first DECISION, soft patches, and empty-tool step retry.
Stop import/fence prompt breakage, prefer modern workbook imports, accept
common patch shapes, and recover when workers CALL disallowed tools.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-16 12:07:17 +08:00
cflmyandCursor ecc92f473a Unify frontmatter imports as import bind:target and drop > imports.
`import json:lib/json.mq.md` binds file libraries; `import fmt:time.format`
binds member short names (replaces `> use`). Legacy `> path.mq.md` / `> use`
now hard-error with a hint. Migrates repo docs, skills, ext, lib, public, examples,
man-test and tests; adds a negative gold and updates view/catalog rendering.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-13 10:03:21 +08:00
cflmyandCursor 25b7be6739 Add usable table/map stdlib with put, and show run bindings in view.
Authors can update nested collections via table.put/表.改 instead of json.set chains; view’s right rail adds collapsible Variables from the entry env.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-12 11:49:48 +08:00
cflmyandCursor 91d51f3f0a Treat bare ids as variables in italic/bold, and land table/quantum follow-ups.
Prefer-var parsing in *…* / **…* (quoted strings for text), table cell expressions, digit map footnotes, and quantum steps/draw with updated golds and LLM sources.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-12 11:21:23 +08:00
cflmyandCursor ab8ea8438f Add local n-gram near match for OKF agent-kb reuse.
Sparse bigram cosine closes near-duplicate goal fragmentation; soft_match parents now judge ranked scored candidates instead of a flat task list.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-11 10:07:03 +08:00
cflmyandCursor f240bdf156 Stream child agent events into view SSE without jumpy drip.
Forward quiet file-subtask EventBus frames while wait blocks, coalesce
DOM updates per rAF, and keep one SSE write per event so Stream stays
live during child work instead of freezing or dumping bursts.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-11 09:45:08 +08:00
cflmyandCursor ff84c43c10 Add OKF soft match, CALL tool streaming, and lib whitelist dispatch.
Canonicalize/alias lookup and optional soft_match parent REUSE cut kb
fragmentation; stream tool_start/end for Plan-and-Move visibility; allow
safe CALL:lib.* without FC.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-11 09:26:59 +08:00
cflmyandCursor dbb2765553 Add parent Plan-and-Move: CALL/READ tools and quiet I/O capture.
Give the plan loop actionable observation (source excerpts, stderr/stdout tails) and short stop rules so the parent can inspect children and use stdlib helpers instead of inventing long patches.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-11 09:11:15 +08:00
cflmyandCursor 7e3b193d49 Stop plan loops after child success; harden stream UX.
Prefer DONE + runtime solidify when the child already returned a value, compact parent observations, drop the SSE 8MiB raw-buffer abort, and drip-stream the view panel via POST /api/run.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-10 17:25:24 +08:00
cflmyandCursor 2a9fceb51c Land OKF aliases, view SSE, workbook nav, and explicit parent ctor.
Complete the core agent roadmap: soft KB hits via Task aliases, EventBus-backed view streaming, D4 writeback/workbook navigation, and I3 explicit-super without implicit auto-super.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-10 11:58:26 +08:00
cflmyandCursor d1ddb74235 Plan parent-first decompose, reasoning streaming, TTY input editing.
Reorder non-hit plan pipeline so the parent decomposes before spawning the child, stream DeepSeek thinking (reasoning_content) to TTY, and add Unicode-aware line editing for interactive input to fix CJK backspace corruption.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-10 11:45:58 +08:00
cflmyandCursor b6bb42e460 Implement streaming S0-S2: SSE complete, step/plan events.
Add net/llm SSE primitives and optional stream/echo/trace on step and plan so callers can watch deltas and rounds without changing final return values.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-10 11:03:04 +08:00
cflmyandCursor d1b33c78c6 Make file subtasks return #main value and expose it as plan.result.
Quiet spawn still hides child stdout; workbooks solidify to **returns**, agent-kb lives in the plugin, and multi-step follows OKF file relay rather than print-based answers.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-07 17:53:14 +08:00
cflmyandCursor fbece725e1 Enable default writeback on agent step and lock the plan design.
Document multi-step workbooks in ext-agent-plan, and make step/单步 persist ok/error slots unless writeback is turned off.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-07 13:47:34 +08:00
cflmyandCursor 1f23e2ed93 Add module namespaces and harden L1/agent after the import redesign.
Bind imports to library names with bare path calls (and use), forward optional host args through mq wrappers, stabilize writeback replacement newlines, and document the M1/M2 decisions.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-07 13:28:32 +08:00
cflmyandCursor 63e5be80ab Add v0.2 surface syntax, writeback/subtask stdlib, and view input deferral.
Use + for parameters, ordered-list branches, backtick identifiers, and quoted strings with escapes. Add writeback and subtask host libs, optional parameter defaults, narrative-bold comment fix, and defer view execution until preset stdin is submitted.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-06 17:52:33 +08:00
cflmyandCursor 92d88fe761 Move LLM and agent into ext/ai with host tool catalog and call_fn.
Consolidate official AI extensions under ext/ai/, add agent history/context host APIs, dynamic call_fn, and a first-pass TOOL:name loop so runbooks can list zero-arg tools in a table.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-06 15:28:39 +08:00