diff --git a/doc/design/ext-abi.md b/doc/design/ext-abi.md index 510a649..0c292be 100644 --- a/doc/design/ext-abi.md +++ b/doc/design/ext-abi.md @@ -83,7 +83,7 @@ cargo build -p marqdo_plugin_agent ``` - Demo (ABI v1): `demo_add`, `demo_echo` — [`tests/lib/plugin-demo.mq.md`](../../tests/lib/plugin-demo.mq.md). -- Agent (ABI v2): layout + session bag + context queries + **OKF agent-kb** (`agent_goal_sig`, `agent_goal_slug`, `agent_kb_lookup`, `agent_kb_promote`, `agent_kb_record_hit`, `agent_workbook_solidify`, `agent_kb_task_files`) — [`plugins/agent`](../../plugins/agent); used by [`ext/ai/agent.mq.md`](../../ext/ai/agent.mq.md). Lookup: `plugin.native_path name=agent` or [ext-cli.md](ext-cli.md) install / `MARQDO_AGENT_PLUGIN`. +- Agent (ABI v2): layout + session bag + context queries + **OKF agent-kb** (`agent_goal_sig`, `agent_goal_slug`, `agent_kb_lookup`, `agent_kb_canonicalize`, `agent_kb_list_tasks`, `agent_kb_add_alias`, `agent_kb_promote`, `agent_kb_record_hit`, `agent_workbook_solidify`, `agent_kb_task_files`) — [`plugins/agent`](../../plugins/agent); used by [`ext/ai/agent.mq.md`](../../ext/ai/agent.mq.md). Lookup: `plugin.native_path name=agent` or [ext-cli.md](ext-cli.md) install / `MARQDO_AGENT_PLUGIN`. ## Layering diff --git a/doc/design/ext-agent-parent.md b/doc/design/ext-agent-parent.md index e5cdb5f..de1f17f 100644 --- a/doc/design/ext-agent-parent.md +++ b/doc/design/ext-agent-parent.md @@ -103,9 +103,11 @@ ### 6.2 标准库与造工具 -- **标准库**:父通过 `lib_catalog` 知晓能力;把 `> lib/….mq.md` 与 `##` 写进工作簿(PATCH)或 scratch 工具文件后即可被子任务 / 后续 `CALL` 使用。 +- **标准库直调(白名单)**:`CALL:lib.fs.read_text`(及 `exists` / `list_dir` / `json.parse` / `json.stringify` / `sys.cwd`)+ 可选 `ARGS:{…}` JSON 行;拒绝 `net.*` / `sys.exec` / 写删类。 +- **目录**:`lib_catalog` 列出可 CALL 的 dotted 名。 - **造工具**:优先 `CONTINUE` + 短 PATCH 在工作簿内新增 `##`;或 `CALL:scratch_tool_write` 沉淀独立工具文件。 -- **非目标(本阶段)**:把全部 `lib/*` 自动反射成 JSON tools schema;原生 function calling。 +- **流式**:父 CALL/READ 发 `tool_start` / `tool_end`(view Tool 卡)。 +- **非目标**:全量 `lib/*` JSON schema 反射;OpenAI function calling。 --- diff --git a/doc/design/ext-agent-plan.md b/doc/design/ext-agent-plan.md index 2699243..b602866 100644 --- a/doc/design/ext-agent-plan.md +++ b/doc/design/ext-agent-plan.md @@ -2,8 +2,8 @@ | | | |---|---| -| 状态 | **Accepted;D2a 已落地;D2b–D2d(观察→补丁式多步)实现中** | -| 日期 | 2026-08-07 | +| 状态 | **Accepted;D2a–D2f 已落地;父 Plan-and-Move / OKF 软命中见配套文档** | +| 日期 | 2026-08-11 | | 父文档 | [ext-agent.md](ext-agent.md) | | 相关 | [**ext-agent-parent.md**](ext-agent-parent.md)(**父 Plan-and-Move 行动面**) · [stdlib-writeback.md](stdlib-writeback.md) · [stdlib-subtask.md](stdlib-subtask.md) · [module-namespace.md](module-namespace.md) · [ai-skill.md](ai-skill.md) · [objects.md](objects.md) | diff --git a/doc/design/okf.md b/doc/design/okf.md index beab816..c811d35 100644 --- a/doc/design/okf.md +++ b/doc/design/okf.md @@ -346,7 +346,9 @@ Prefer spawning `resource` over re-planning. 1. **精确**:slug 试探 + frontmatter `sig` → 扫目录 `sig:` → 旧 hex 文件名。 2. **别名(Accepted)**:精确 miss 后,扫 Task FM `aliases:`(YAML 列表或 `[a, b]`);`normalize_goal(query)` 与某条 alias **精确相等**则命中同一 Skill/resource。lookup 返回 `match: exact|alias`;`plan` 对 alias 写 `cache=soft-hit`。 -3. **不做**默认 embedding / 编辑距离主路径。父裁决 soft_match 仍见 [okf-near-match.md](../roadmap/okf-near-match.md)。 +3. **规范句(Accepted)**:仍 miss 时对 goal 做轻量 `agent_kb_canonicalize`(剥站立前缀、去尾标点),再精确/别名各试一次;命中返回 `match: canonical`,`cache=soft-hit`。 +4. **父裁决 soft_match(Accepted,默认关)**:`plan soft_match=True` 且仍 miss 时,把 `agent_kb_list_tasks` 候选交给父模型:`DECISION: REUSE` + `SLUG:` 或 `DECISION: NEW`;REUSE 后可 `agent_kb_add_alias` 把当前 goal 写入别名。 +5. **不做**默认 embedding / 编辑距离主路径。向量层见 [okf-near-match.md](../roadmap/okf-near-match.md) E。 `agent_kb_promote` 可选 `aliases=`(字符串或列表)写入 Task;再晋升时合并既有 aliases。 @@ -357,6 +359,7 @@ Prefer spawning `resource` over re-planning. | `reuse` / `复用` | `True` | 允许知识包快路径 | | `optimize` / `优化` | `False` | 跳过知识包,强制再规划 | | `force` / `强制` | `False` | 忽略知识包 | +| `soft_match` / `软命中` | `False` | miss 后父裁决 REUSE/NEW(须显式打开) | | `promote` / `晋升` | `True` | DONE 后固化并写入/更新概念与 resource | | `kb_dir` / `知识库目录` | `.marqdo/agent-kb` | bundle 根 | | `workbook_dir` / `工作簿目录` | `None` | 默认直接写 `kb_dir/resources/.mq.md`;显式传入则用该目录下的临时 `workbook--.mq.md` | @@ -432,7 +435,7 @@ agent-kb 运行时实现归属 **agent 插件**(`agent_goal_sig` / `agent_kb_l | **O2** | 本文锁定的任务知识包:`Marqdo Task` / `Agent Skill`;`plan` reuse/promote | **done**(host kb_* + plan 快路径) | | **O3** | catalog 增强:`verified`/`sources` 可选;模块页可点击依赖;扫入 `agent-kb` 概念供人浏览 | **done** | | **O4** | `log.md` 策展、可选 `kb query`、Attested-style「再现跑通」收据(非 BigQuery) | 远期 | -| **O5** | 相近任务软命中:Task `aliases` 第二趟已落地;规范句 / 父裁决 / 向量仍见 [okf-near-match.md](../roadmap/okf-near-match.md) | **aliases done** | +| **O5** | 相近任务软命中:aliases + canonicalize + `soft_match` 父裁决已落地;向量仍见 [okf-near-match.md](../roadmap/okf-near-match.md) E | **A/B/C done** | O2 设计落盘后,智能体侧实现文档可薄化为指向本文 §7,避免两套真理。 diff --git a/doc/design/view.md b/doc/design/view.md index 3d01333..707e5e9 100644 --- a/doc/design/view.md +++ b/doc/design/view.md @@ -116,7 +116,8 @@ marqdo view output tests/structure/hello.mq.md -o /tmp/hello-doc |------|------|------| | **Thinking** | `reasoning` | 可折叠;流式追加到同一块;默认展开直至本轮 `delta`/`done` 开始,之后可收起 | | **Answer** | `delta` | **主文**;token 连续拼进同一段落(禁止每 token 换行);主色正文 | -| **Decision** | `decision` | 时间线条目标记(`RUN` / `CONTINUE` / `DONE` + 可选 summary) | +| **Decision** | `decision` | 时间线条目标记(`RUN` / `CONTINUE` / `DONE` / `REUSE` / `NEW` + 可选 summary) | +| **Tool** | `tool_start` / `tool_end` | 父 Plan-and-Move 的 CALL/READ:名称 + kind;`tool_end` 可带截断 `result` | | **Child / Workbook** | `round` | **独立卡片**:轮次、退出码、workbook 链(`/file?path=`)、可选 `result` 摘要;与父思考/答案视觉分离 | | **Status** | `run_start` / `done` / `error` | 顶栏状态 + 错误行;`done.result` 仅在无 Answer 块时作摘要,不与 reasoning 混排 | diff --git a/doc/roadmap/agent-streaming.md b/doc/roadmap/agent-streaming.md index 8d76765..6435bfe 100644 --- a/doc/roadmap/agent-streaming.md +++ b/doc/roadmap/agent-streaming.md @@ -64,7 +64,7 @@ ext/ai/agent step/plan 订阅事件 → 可选透传或写回「过程槽」 |--------|------|----------| | `reasoning` | 模型思考增量(如 DeepSeek `reasoning_content`) | `text` | | `delta` | 模型答案增量文本 | `text` | -| `tool_start` / `tool_end` | 工具轮 | `name`, `result?` | +| `tool_start` / `tool_end` | 工具轮(父 plan CALL/READ 已发;view Tool 卡) | `name`, `kind?`, `result?` | | `round` | plan 轮次边界(子工作簿跑完) | `round`, `workbook`, `exit_code?`, `result?`(子返回值摘要) | | `decision` | DONE / CONTINUE / RUN | `decision`, `summary?` | | `done` | 本调用结束 | `result`(答案正文;不含 thinking) | diff --git a/doc/roadmap/okf-near-match.md b/doc/roadmap/okf-near-match.md index afacfd1..76af199 100644 --- a/doc/roadmap/okf-near-match.md +++ b/doc/roadmap/okf-near-match.md @@ -2,8 +2,8 @@ | | | |---|---| -| 状态 | **B aliases 已落地**(A 规范句 / C 父裁决 / E 向量仍待) | -| 日期 | 2026-08-09 | +| 状态 | **A 规范句 + B aliases + C 父裁决(soft_match)已落地**;E 向量仍待 | +| 日期 | 2026-08-11 | | 相关 | [okf.md](../design/okf.md) §7 · [ext-agent-plan.md](../design/ext-agent-plan.md) §4.2 · [ext-agent.md](../design/ext-agent.md) | | 触发观察 | `man-test`:相近 goal 生成两份独立 resource | @@ -11,23 +11,20 @@ ## 1. 问题 -今日 agent-kb **只做精确命中**: +v1 agent-kb **默认只做精确命中**(仍是默认快路径): 1. `normalize_goal`:去首尾空白、合并连续空白(**不做**近义改写)。 2. `sig` = 规范化字符串的稳定短哈希;`slug` = 可读路径名。 -3. `agent_kb_lookup`:按 `sig` / slug 精确对上才复用;否则新建 Task / Skill / `resources/.mq.md`。 +3. `agent_kb_lookup`:精确 → aliases → **canonicalize 后再精确/别名**;仍 miss 且 `plan soft_match=True` 时父裁决 REUSE/NEW。 -因此字面不同、含义相近的两次 `plan` 会被当成**两个任务**。实测例子: +历史痛点(已由 A/B/C 缓解): -| goal(调用方输入) | 产物 slug(示意) | -|--------------------|-------------------| -| `你是一个智能体,帮我规划明天的行程` | `你是一个智能体-帮我规划明天的行程` | -| `帮我规划明行程` | `帮我规划明行程` | +| goal(调用方输入) | 旧行为 | +|--------------------|--------| +| `你是一个智能体,帮我规划明天的行程` | 独立 slug,与下方不命中 | +| `帮我规划明天的行程` | 另一份 resource | -调用方直觉:「这不是同一件事吗?」 -框架行为:「指纹不同 → 不命中 → 再探索 / 再固化。」 - -这在 v1 **是合理且有意的**(见 §2),但缺少一层「相近含义」通道时,知识包会碎片化,复用率下降。 +调用方直觉:「这不是同一件事吗?」——现可用 canonicalize(剥站立前缀)或显式 `soft_match` 父裁决。 --- @@ -44,14 +41,14 @@ --- -## 3. 目标(未来能力) +## 3. 目标 -在 **不破坏** 精确 `sig` 契约的前提下,可选地: +在 **不破坏** 精确 `sig` 契约的前提下: -1. 识别「同一任务意图的不同措辞」→ 复用已有 Skill / resource;或 -2. 至少提示调用方 / 父智能体:「可能命中已有任务 X,确认后复用」。 +1. **已落地**:A 规范句、B aliases、C `soft_match` 父裁决(默认关)。 +2. **仍待**:E 向量(仅有证据再开)。 -非目标(本规划明确不做为默认): +非目标(仍不做为默认): - 把 agent-kb 做成通用向量知识库 / RAG 产品。 - 用不可审计的静默相似度覆盖精确 `sig`。 @@ -63,32 +60,26 @@ 相近命中 **不等于** 必须上向量检索。按与 Marqdo 契合度排序: -### A. 规范化后再精确命中(首选探索) +### A. 规范化后再精确命中 — **已落地** | | | |--|--| -| 做法 | lookup 前把 goal 收成**规范句**(规则表,或一次小模型 / 父 LLM 调用),再算现有 `sig` | -| 优点 | 索引与今日相同;无 ANN;结果可写回 Task(`canonical` / `aliases`) | -| 风险 | 规范器不稳会导致漂移;需冻结「规范句」写入 task 页,避免每次改写 | -| 落地形态 | `agent_kb_canonicalize`(插件)或 `plan` 内可选 `canonicalize=True` | +| 做法 | `agent_kb_canonicalize`:剥站立前缀、去尾 `?`/`?`/`。`,再 `normalize_goal`;lookup 第三趟 | +| 落地 | 插件 + `match: canonical`;`cache=soft-hit` | -### B. 别名 / 标签表(文档原生) +### B. 别名 / 标签表 — **已落地** | | | |--|--| -| 做法 | Task frontmatter 增加 `aliases: […]` / `tags: […]`;lookup:先 `sig`,再扫别名精确匹配 | -| 优点 | 完全可读可 git;promote / 人工策展可补别名 | -| 风险 | 冷启动靠人;自动抽别名仍可能要模型 | -| 落地形态 | 扩 `concepts/tasks/*.md`;`kb_lookup` 第二趟扫 aliases | +| 做法 | Task FM `aliases:`;lookup 第二趟精确匹配 | +| 落地 | `match: alias`;promote / `agent_kb_add_alias` | -### C. 父智能体软裁决(miss 后) +### C. 父智能体软裁决(miss 后)— **已落地(默认关)** | | | |--|--| -| 做法 | 精确 miss 时,把 `index.md`(或 task 标题列表)交给父模型:`REUSE:` / `NEW` | -| 优点 | 无新基础设施;与「父是开发大师」一致;决策可写回 | -| 风险 | 多一次 LLM;需严格协议防乱指;`llm_free` 快路径仍走精确命中 | -| 落地形态 | `plan` 在 reuse miss 且 `soft_match=True` 时分支 | +| 做法 | `plan soft_match=True`:`agent_kb_list_tasks` → 父 `DECISION: REUSE` + `SLUG:` 或 `DECISION: NEW` | +| 落地 | REUSE → 现有 hit 分支;可选 append alias | ### D. 字面近似(弱,仅拼写容错) @@ -110,29 +101,30 @@ --- -## 5. 推荐演进顺序 +## 5. 演进顺序 ```text -精确 sig(现状,永不删除) - → A 规范句 或 B aliases(文档层,可金样) - → C soft_match 父裁决(可选,打网) +精确 sig(永不删除) + → A 规范句 + B aliases(已落地) + → C soft_match 父裁决(已落地,默认关) → E 向量(仅有证据再开) ``` -验收原则: +验收原则(仍适用): -1. 默认行为与今日一致(未开开关 = 仅精确命中)。 -2. 软命中必须留下可读痕迹(写回 `cache=soft-hit`、选用的 slug、可选置信说明)。 +1. 默认行为与精确路径一致(`soft_match` 默认 False;canonicalize 无模型、低风险)。 +2. 软命中留下 `cache=soft-hit` / `match` 痕迹。 3. 假阳性可被 `force=True` / 显式 `workbook=` 覆盖。 4. 实现仍在 **agent 插件 + `ext/ai`**,不进核心 `src/host`。 --- -## 6. 调用方近期规避(无需等实现) +## 6. 调用方建议 -- 固定规范 goal 字符串(应用层常量),不要把「你是一个智能体」等站立语拼进 goal。 +- 固定规范 goal 字符串;站立语可由 canonicalize 剥掉,但仍宜少拼。 - 续跑同一主题时传 `workbook=` 或已知 resource 路径。 -- 策展:合并重复 task 页,把次要措辞写入将来的 `aliases`(今日可先记在 task 正文)。 +- 近义但字面差大时显式 `soft_match=True`(多一次父 LLM)。 +- 策展:合并重复 task,次要措辞写入 `aliases`。 --- @@ -140,8 +132,6 @@ | 条目 | 关系 | |------|------| -| [okf.md](../design/okf.md) O2 | 任务包精确 reuse **已完成**;本文是 O2 之上的软命中延伸 | -| [okf.md](../design/okf.md) O4 | `kb query` / 策展可与 B(aliases)同批考虑 | -| [ext-agent-plan.md](../design/ext-agent-plan.md) | 多轮仍只走子文件接力;软命中只影响「选哪个子文件」,不改变模式 C | - -本文不锁定实现工期;开工前应先选 A/B/C 之一写一小段 Accepted 补丁进 `okf.md` §7.3,再改插件与金样。 +| [okf.md](../design/okf.md) O2 / O5 | 精确 reuse + A/B/C 软命中;E 向量仍待 | +| [okf.md](../design/okf.md) O4 | `kb query` / 策展可继续扩 aliases | +| [ext-agent-plan.md](../design/ext-agent-plan.md) | 多轮仍只走子文件接力;软命中只影响「选哪个子文件」 | diff --git a/ext/ai/agent.mq.md b/ext/ai/agent.mq.md index 47f5d02..5e5d457 100644 --- a/ext/ai/agent.mq.md +++ b/ext/ai/agent.mq.md @@ -344,13 +344,131 @@ Drop `` blocks (including marqdo-out bodies) so parent excerpts stay ## lib_catalog -Short catalog of English stdlib modules the parent may wire into workbooks or scratch tools. +Callable dotted names (whitelist for `CALL:lib…`) plus static module file list for workbook imports. -*`out` = > json.parse text={"note":"Import via frontmatter > lib/.mq.md then call ## members. Prefer PATCH workbook or CALL:scratch_tool_write over inventing prose.","files":["fs.mq.md","json.mq.md","subtask.mq.md","sys.mq.md","writeback.mq.md","text.mq.md","time.mq.md","math.mq.md","net.mq.md","table.mq.md","plugin.mq.md","foreign.mq.md"]} * +*`out` = > json.parse text={"note":"Direct CALL:lib.. + optional ARGS:{json} for whitelist only. Or import via frontmatter > lib/.mq.md then PATCH / scratch_tool_write.","files":["fs.mq.md","json.mq.md","subtask.mq.md","sys.mq.md","writeback.mq.md","text.mq.md","time.mq.md","math.mq.md","net.mq.md","table.mq.md","plugin.mq.md","foreign.mq.md"],"callable":[{"name":"lib.fs.read_text","desc":"Read UTF-8 text file"},{"name":"lib.fs.exists","desc":"Path exists?"},{"name":"lib.fs.list_dir","desc":"List directory entries"},{"name":"lib.json.parse","desc":"Parse JSON text"},{"name":"lib.json.stringify","desc":"Serialize JSON"},{"name":"lib.sys.cwd","desc":"Process working directory"}]} * **`out`** --- +## extract_call_args + + `reply`=None + +Parse `ARGS:{…}` JSON line, or a bare `{…}` line after CALL. + +*`esc` = > json.parse text={"nl":"\n","mark":"ARGS:","brace":"{"} * +*`nl` = > json.get value=`esc` key=nl * +*`mark` = > json.get value=`esc` key=mark * +*`brace` = > json.get value=`esc` key=brace * +*`empty` = > json.parse text={} * +1. `reply` + *`text` = > trim value=`reply` * + *`lines` = > split value=`text` sep=`nl` * + *`found` = None* + - [`line`](`lines`) + 1. `found` + *`_` = 1* + 2. * + *`t` = > trim value=`line` * + *`chunks` = > split value=`t` sep=`mark` * + *`cn` = > len value=`chunks` * + 1. `cn` > 1 + *`rest` = > at value=`chunks` index=1 * + *`found` = > trim value=`rest` * + 2. * + *`chunks` = > split value=`t` sep=`brace` * + *`cn` = > len value=`chunks` * + *`prefix` = > at value=`chunks` index=0 * + 1. `cn` > 1 + 1. `prefix` + *`_` = 1* + 2. * + *`found` = `t`* + 2. * + *`_` = 1* + 1. `found` + **> json.parse text=`found`** + 2. * + **`empty`** +2. * + **`empty`** + +--- + +## tool_result_brief + + `value`=None + + `max_chars`=1024 + +Truncate a tool result for `tool_end` events (~1KiB). + +1. `value` + *`s` = > json.stringify value=`value` * + **> text_head_lines text=`s` max_lines=40 max_chars=`max_chars`** +2. * + **** + +--- + +## plan_append_tool + + `events` + + `type` + + `name` + + `kind` + + `result`=None + + `stream`=False + +Publish `tool_start` / `tool_end` when `stream=True`. + +1. `stream` + *`ev` = > json.parse text={} * + *`ev` = > json.set map=`ev` key=type value=`type` * + *`ev` = > json.set map=`ev` key=name value=`name` * + *`ev` = > json.set map=`ev` key=kind value=`kind` * + 1. `result` + *`ev` = > json.set map=`ev` key=result value=`result` * + 2. * + *`_` = 1* + *`events` = > json.append list=`events` item=`ev` * + > sys.stream_publish event=`ev` +2. * + *`_` = 1* + +**`events`** + +--- + +## run_lib_call + + `name` + + `reply`=None + +Whitelist `CALL:lib..` (+ ARGS). Rejects net/exec/write/delete. + +*`args` = > extract_call_args reply=`reply` * +*`errs` = > json.parse text={"deny":"Lib CALL not allowed: "} * + +1. `name` == lib.fs.exists + *`path` = > json.get value=`args` key=path * + **> fs.exists path=`path`** +2. `name` == lib.fs.read_text + *`path` = > json.get value=`args` key=path * + **> fs.read_text path=`path`** +3. `name` == lib.fs.list_dir + *`path` = > json.get value=`args` key=path * + **> fs.list_dir path=`path`** +4. `name` == lib.json.parse + *`text` = > json.get value=`args` key=text * + **> json.parse text=`text`** +5. `name` == lib.json.stringify + *`value` = > json.get value=`args` key=value * + **> json.stringify value=`value`** +6. `name` == lib.sys.cwd + **> sys.cwd** +7. * + *`deny` = > json.get value=`errs` key=deny * + **`deny` + `name`** + +--- + ## scratch_tool_write + `name` + `text` @@ -510,7 +628,7 @@ Parse NAME: id plus a triple-angle body after CALL:scratch_tool_write. + `observation`=None + `reply`=None -Parent Plan-and-Move tools (stdlib-aware helpers in this module). +Parent Plan-and-Move tools (helpers + whitelist `CALL:lib…`). 1. `name` == workbook_read **> workbook_read path=`path` depth=deep** @@ -531,8 +649,13 @@ Parent Plan-and-Move tools (stdlib-aware helpers in this module). 2. * **> json.get value=`errs` key=b** 5. * - *`msg` = Parent tool not allowed: * - **`msg` + `name`** + *`dots` = > split value=`name` sep=. * + *`head` = > at value=`dots` index=0 * + 1. `head` == lib + **> run_lib_call name=`name` reply=`reply`** + 2. * + *`msg` = Parent tool not allowed: * + **`msg` + `name`** --- @@ -742,17 +865,97 @@ Parse `DECISION: DONE` / `CONTINUE` / `RUN` (Chinese: `决定:完成` / `继 **CONTINUE** 3. `found` == RUN **RUN** -4. `found` == 完成 +4. `found` == REUSE + **REUSE** +5. `found` == NEW + **NEW** +6. `found` == 完成 **DONE** -5. `found` == 继续 +7. `found` == 继续 **CONTINUE** -6. `found` == 运行 +8. `found` == 运行 **RUN** -7. * +9. `found` == 复用 + **REUSE** +10. `found` == 新建 + **NEW** +11. * **`found`** --- +## extract_soft_match_slug + + `reply` + +Parse `SLUG: …` / `标识:…` after a soft-match REUSE decision. + +*`esc` = > json.parse text={"nl":"\n","fw":":"} * +*`nl` = > json.get value=`esc` key=nl * +*`fw` = > json.get value=`esc` key=fw * +*`text` = > trim value=`reply` * +*`lines` = > split value=`text` sep=`nl` * +*`found` = None* + +- [`line`](`lines`) + *`t` = > trim value=`line` * + *`parts` = > split value=`t` sep=: * + *`n` = > len value=`parts` * + 1. `n` > 1 + *`head` = > at value=`parts` index=0 * + *`head` = > trim value=`head` * + *`rest` = > at value=`parts` index=1 * + *`rest` = > trim value=`rest` * + 1. `head` == SLUG + *`found` = `rest`* + 2. `head` == 标识 + *`found` = `rest`* + 3. * + *`_` = 1* + 2. * + *`parts` = > split value=`t` sep=`fw` * + *`n` = > len value=`parts` * + 1. `n` > 1 + *`head` = > at value=`parts` index=0 * + *`head` = > trim value=`head` * + *`rest` = > at value=`parts` index=1 * + *`rest` = > trim value=`rest` * + 1. `head` == 标识 + *`found` = `rest`* + 2. `head` == SLUG + *`found` = `rest`* + 3. * + *`_` = 1* + 2. * + *`_` = 1* + +**`found`** + +--- + +## build_soft_match_prompt + + `goal` + + `tasks` + +Short parent prompt: REUSE+SLUG or NEW only. + +*`esc` = > json.parse text={"a":"\n\n--- goal ---\n","b":"\n\n--- candidate tasks (slug | title) ---\n","c":"\n\n--- how to act ---\nExact reuse miss. Reply with ONE protocol only:\nDECISION: REUSE\nSLUG: \nor\nDECISION: NEW\nNo other prose.\n","nl":"\n","sep":" | "} * +*`a` = > json.get value=`esc` key=a * +*`b` = > json.get value=`esc` key=b * +*`c` = > json.get value=`esc` key=c * +*`nl` = > json.get value=`esc` key=nl * +*`sep` = > json.get value=`esc` key=sep * +*`lines` = > json.parse text=[] * +- [`t`](`tasks`) + *`slug` = > json.get value=`t` key=slug * + *`title` = > json.get value=`t` key=title * + *`row` = `slug` + `sep` + `title` * + *`lines` = > json.append list=`lines` item=`row` * +*`catalog` = > join value=`lines` sep=`nl` * +*`p` = Soft-match: same task intent? * +**`p` + `a` + `goal` + `b` + `catalog` + `c`** + +--- + ## extract_plan_summary + `reply` @@ -824,7 +1027,7 @@ Parent Plan-and-Move prompt. Short protocol only — no long monologues. *`compact` = > compact_plan_observation observation=`observation` * *`obs_s` = > json.stringify value=`compact` * *`goal_s` = > json.stringify value=`goal` * -*`esc` = > json.parse text={"a":"\n\n--- standing ---\n","b":"\n\n--- goal ---\n","c":"\n\n--- observation ---\n","d":"\n\n--- skill brief ---\n","tools":"\n\n--- parent tools ---\nCALL:workbook_read | workbook_excerpt | lib_catalog | scratch_tool_write\nREAD:source | stderr | stdout | slots\nCreate tools: CONTINUE PATCH add ##, or CALL:scratch_tool_write with NAME line and fenced body.\n","e_rev":"\n\n--- how to act ---\nPlan-and-Move parent. Reply with ONE protocol only (no long reasoning).\n1) exit_code=0 and has_value → DECISION: DONE + one-line SUMMARY. solidify_on_done. Do not CONTINUE just because has_worker_step.\n2) Need more evidence → READ:kind or CALL:tool (then you will be re-prompted).\n3) Failure / wrong value → DECISION: CONTINUE + short PATCH (<20 lines). Never paste user prose into REPLACE.\nProtocols:\nCALL:name\nREAD:kind\nDECISION: DONE\nSUMMARY: one line\nDECISION: CONTINUE\nPATCH with FIND/REPLACE fences\n","e_dec":"\n\n--- how to act ---\nPRE-RUN decompose. Reply with ONE protocol only (no long reasoning).\n1) Skeleton OK → DECISION: RUN.\n2) Need evidence → READ:source or CALL:workbook_read / lib_catalog.\n3) Reshape → DECISION: CONTINUE + short PATCH (<20 lines).\n4) Fixed answer without LLM → PATCH to return, then RUN or DONE.\nProtocols:\nCALL:name\nREAD:kind\nDECISION: RUN\nDECISION: DONE\nSUMMARY: one line\nDECISION: CONTINUE\nPATCH with FIND/REPLACE fences\n","f":"\n\n--- explore ---\n"} * +*`esc` = > json.parse text={"a":"\n\n--- standing ---\n","b":"\n\n--- goal ---\n","c":"\n\n--- observation ---\n","d":"\n\n--- skill brief ---\n","tools":"\n\n--- parent tools ---\nCALL:workbook_read | workbook_excerpt | lib_catalog | scratch_tool_write\nCALL:lib.fs.read_text|exists|list_dir | lib.json.parse|stringify | lib.sys.cwd (+ ARGS:{json})\nREAD:source | stderr | stdout | slots\nCreate tools: CONTINUE PATCH add ##, or CALL:scratch_tool_write with NAME line and fenced body.\n","e_rev":"\n\n--- how to act ---\nPlan-and-Move parent. Reply with ONE protocol only (no long reasoning).\n1) exit_code=0 and has_value → DECISION: DONE + one-line SUMMARY. solidify_on_done. Do not CONTINUE just because has_worker_step.\n2) Need more evidence → READ:kind or CALL:tool (then you will be re-prompted).\n3) Failure / wrong value → DECISION: CONTINUE + short PATCH (<20 lines). Never paste user prose into REPLACE.\nProtocols:\nCALL:name\nREAD:kind\nDECISION: DONE\nSUMMARY: one line\nDECISION: CONTINUE\nPATCH with FIND/REPLACE fences\n","e_dec":"\n\n--- how to act ---\nPRE-RUN decompose. Reply with ONE protocol only (no long reasoning).\n1) Skeleton OK → DECISION: RUN.\n2) Need evidence → READ:source or CALL:workbook_read / lib_catalog.\n3) Reshape → DECISION: CONTINUE + short PATCH (<20 lines).\n4) Fixed answer without LLM → PATCH to return, then RUN or DONE.\nProtocols:\nCALL:name\nREAD:kind\nDECISION: RUN\nDECISION: DONE\nSUMMARY: one line\nDECISION: CONTINUE\nPATCH with FIND/REPLACE fences\n","f":"\n\n--- explore ---\n"} * *`a` = > json.get value=`esc` key=a * *`b` = > json.get value=`esc` key=b * *`c` = > json.get value=`esc` key=c * @@ -889,11 +1092,17 @@ LLM Plan → CALL/READ/DECISION loop. Returns `{decision,reply,observation,event *`kind` = > json.get value=`act` key=kind * *`name` = > json.get value=`act` key=name * 1. `kind` == call + *`evs_all` = > plan_append_tool events=`evs_all` type=tool_start name=`name` kind=call stream=`stream` * *`tool_out` = > run_parent_tool name=`name` path=`path` observation=`last_obs` reply=`last_reply` * + *`brief` = > tool_result_brief value=`tool_out` * + *`evs_all` = > plan_append_tool events=`evs_all` type=tool_end name=`name` kind=call result=`brief` stream=`stream` * *`last_obs` = > json.set map=`last_obs` key=tool_result value=`tool_out` * *`left` = `left` - 1* 2. `kind` == read + *`evs_all` = > plan_append_tool events=`evs_all` type=tool_start name=`name` kind=read stream=`stream` * *`last_obs` = > plan_read_deepen observation=`last_obs` kind=`name` path=`path` * + *`brief` = > tool_result_brief value=`last_obs` * + *`evs_all` = > plan_append_tool events=`evs_all` type=tool_end name=`name` kind=read result=`brief` stream=`stream` * *`left` = `left` - 1* 3. `kind` == decision *`decision` = `name`* @@ -1193,6 +1402,7 @@ With `stream=True`, the model call uses SSE; `echo=True` prints delta text to st + `reuse`=True + `optimize`=False + `force`=False + + `soft_match`=False + `promote`=True + `kb_dir`=.marqdo/agent-kb + `improve_every`=3 @@ -1203,7 +1413,7 @@ With `stream=True`, the model call uses SSE; `echo=True` prints delta text to st Multi-step with OKF agent-kb. Default workbook is `kb_dir/resources/.mq.md`. While task file count `< explore_n` and skill is not llm_free, force a new explore variant under `kb_dir/explore//`. Code-first: llm_free hits skip parent LLM. File children return via `# main`; `plan` exposes that as `result`. -Non-hit path runs parent **decompose** before the first child spawn (`DECISION: RUN` / `CONTINUE`+patch / solidified `DONE`). Then `await` → revise loop. With `stream=True`, emit parent `delta` / `decision` / `round` / `done` on `events`. `echo=True` prints `plan:decompose` / `plan:await` / deltas. `trace=True` writes events to writeback slot `trace`. Quiet child subtasks stay quiet. +Non-hit path: optional `soft_match=True` parent REUSE/NEW; else **decompose** before first child spawn (`DECISION: RUN` / `CONTINUE`+patch / solidified `DONE`). Then `await` → revise loop. With `stream=True`, emit parent `delta` / `decision` / `tool_start`/`tool_end` / `round` / `done` on `events`. `echo=True` prints `plan:decompose` / `plan:await` / deltas. `trace=True` writes events to writeback slot `trace`. Quiet child subtasks stay quiet. *`tools` = > json.get value=`self` key=tools * *`cache` = miss* @@ -1227,7 +1437,13 @@ Non-hit path runs parent **decompose** before the first child spawn (`DECISION: *`match_kind` = > json.get value=`hit` key=match * 1. `match_kind` == alias *`cache_label` = soft-hit* - 2. * + 2. `match_kind` == canonical + *`cache_label` = soft-hit* + *`hit_slug` = > json.get value=`hit` key=slug * + > agent_kb_add_alias kb_dir=`kb_dir` slug=`hit_slug` alias=`goal` + 3. `match_kind` == soft + *`cache_label` = soft-hit* + 4. * *`cache_label` = hit* *`lf` = > json.get value=`hit` key=llm_free * 1. `lf` @@ -1311,6 +1527,71 @@ Non-hit path runs parent **decompose** before the first child spawn (`DECISION: 2. * *`path` = None* 2. * + 1. `soft_match` + *`listed` = > agent_kb_list_tasks kb_dir=`kb_dir` * + *`tasks` = > json.get value=`listed` key=tasks * + *`tn` = > len value=`tasks` * + 1. `tn` > 0 + *`soft_prompt` = > build_soft_match_prompt goal=`goal` tasks=`tasks` * + *`model` = > json.get value=`self` key=model * + 1. `stream` + *`evs` = > `model`.complete prompt=`soft_prompt` stream=True echo=`echo` * + *`soft_reply` = > llm.stream_result events=`evs` * + *`events` = > plan_merge_deltas events=`events` from=`evs` stream=`stream` * + 2. * + *`soft_reply` = > `model`.complete prompt=`soft_prompt` * + *`soft_reply` = > trim value=`soft_reply` * + *`soft_dec` = > extract_plan_decision reply=`soft_reply` * + *`events` = > plan_append_decision events=`events` decision=`soft_dec` stream=`stream` * + 1. `soft_dec` == REUSE + *`reuse_slug` = > extract_soft_match_slug reply=`soft_reply` * + 1. `reuse_slug` + *`hit` = > agent_kb_lookup kb_dir=`kb_dir` slug=`reuse_slug` goal=`goal` tools=`tools` * + 1. `hit` + > agent_kb_add_alias kb_dir=`kb_dir` slug=`reuse_slug` alias=`goal` + *`cache_label` = soft-hit* + *`path` = > json.get value=`hit` key=resource * + *`aw` = > await_workbook path=`path` * + *`code` = > json.get value=`aw` key=code * + *`child_val` = > json.get value=`aw` key=value * + *`last_obs` = > json.get value=`aw` key=observation * + 1. `code` == 0 + > agent_kb_record_hit kb_dir=`kb_dir` goal=`goal` tools=`tools` improve_every=`improve_every` + *`cache` = soft-hit* + *`out` = > json.parse text={"status":"ok"} * + *`out` = > json.set map=`out` key=status value=ok * + *`out` = > json.set map=`out` key=goal value=`goal` * + *`out` = > json.set map=`out` key=workbook value=`path` * + *`out` = > json.set map=`out` key=rounds value=1 * + *`out` = > json.set map=`out` key=cache value=soft-hit * + *`sk` = > json.get value=`hit` key=skill * + *`out` = > json.set map=`out` key=skill value=`sk` * + *`st` = > json.get value=`hit` key=status * + *`out` = > json.set map=`out` key=skill_status value=`st` * + *`sum` = OKF soft_match REUSE; spawned resource * + *`out` = > json.set map=`out` key=summary value=`sum` * + *`out` = > json.set map=`out` key=observation value=`last_obs` * + *`out` = > json.set map=`out` key=result value=`child_val` * + *`events` = > plan_append_round events=`events` round=1 workbook=`path` exit_code=`code` result=`child_val` stream=`stream` echo=`echo` * + 1. `writeback` + *`body` = > json.stringify value=`out` * + > writeback.record value=`body` key=ok + 2. * + *`_` = 1* + *`out` = > plan_finish_stream out=`out` events=`events` stream=`stream` trace=`trace` result=`child_val` * + **`out`** + 2. * + *`path` = None* + 2. * + *`_` = 1* + 2. * + *`_` = 1* + 2. * + *`_` = 1* + 2. * + *`_` = 1* + 2. * + *`_` = 1* 1. `nfiles` < `explore_n` *`explore` = 1* *`cache` = explore* diff --git a/ext/ai/智能体.mq.md b/ext/ai/智能体.mq.md index 02a87e0..0f119db 100644 --- a/ext/ai/智能体.mq.md +++ b/ext/ai/智能体.mq.md @@ -215,6 +215,7 @@ description: 文档驱动智能体 — 单步 / 多步;工具经 lib/子任务 + `复用`=True + `优化`=False + `强制`=False + + `软命中`=False + `晋升`=True + `知识库目录`=.marqdo/agent-kb + `改进周期`=3 @@ -223,10 +224,10 @@ description: 文档驱动智能体 — 单步 / 多步;工具经 lib/子任务 + `打印增量`=False + `轨迹`=False -多步工作簿默认写入 OKF `知识库目录/resources/.mq.md`;仅当传入 `工作簿目录` 时才用临时 runs。转发至英文 `agent.plan`。`流式` / `打印增量` / `轨迹` 对应 `stream` / `echo` / `trace`。 +多步工作簿默认写入 OKF `知识库目录/resources/.mq.md`;仅当传入 `工作簿目录` 时才用临时 runs。转发至英文 `agent.plan`。`流式` / `打印增量` / `轨迹` 对应 `stream` / `echo` / `trace`;`软命中` → `soft_match`。 *`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`.plan goal=`目标` workbook_dir=`工作簿目录` workbook=`工作簿` max_rounds=`最大轮次` confirm=`确认分解` writeback=`写回` skeleton=`骨架` reuse=`复用` optimize=`优化` force=`强制` promote=`晋升` kb_dir=`知识库目录` improve_every=`改进周期` explore_n=`探索次数` stream=`流式` echo=`打印增量` trace=`轨迹`** +**> `en`.plan goal=`目标` workbook_dir=`工作簿目录` workbook=`工作簿` max_rounds=`最大轮次` confirm=`确认分解` writeback=`写回` skeleton=`骨架` reuse=`复用` optimize=`优化` force=`强制` soft_match=`软命中` promote=`晋升` kb_dir=`知识库目录` improve_every=`改进周期` explore_n=`探索次数` stream=`流式` echo=`打印增量` trace=`轨迹`** diff --git a/plugins/agent/src/kb.rs b/plugins/agent/src/kb.rs index 5742af7..484d58d 100644 --- a/plugins/agent/src/kb.rs +++ b/plugins/agent/src/kb.rs @@ -23,6 +23,46 @@ pub fn normalize_goal(raw: &str) -> String { out } +/// Light rule-based canonicalize (no model): strip standing prefixes, trailing `?`/`?`/`。`. +pub fn canonicalize_goal_str(raw: &str) -> String { + let mut s = raw.trim().to_string(); + const PREFIXES: &[&str] = &[ + "你是一个智能体,", + "你是一个智能体,", + "你是一个智能体:", + "你是一个智能体:", + "你是一个智能体", + "作为一个智能体,", + "作为一个智能体,", + "作为一个智能体:", + "作为一个智能体:", + "作为一个智能体", + "You are an agent, ", + "You are an agent,", + "You are an agent. ", + "You are an agent.", + "You are an agent ", + "You are an agent", + "As an agent, ", + "As an agent,", + "As an agent ", + "As an agent", + ]; + for p in PREFIXES { + if let Some(rest) = s.strip_prefix(p) { + s = rest.trim_start().to_string(); + break; + } + } + while s.ends_with('?') || s.ends_with('?') || s.ends_with('。') { + s.pop(); + while s.ends_with(|c: char| c.is_whitespace()) { + s.pop(); + } + } + normalize_goal(&s) +} + pub fn goal_sig_hex(material: &str) -> String { let mut h: u64 = 5381; for b in material.as_bytes() { @@ -561,24 +601,21 @@ pub fn goal_slug(args: &Value) -> Result { Ok(Value::String(goal_slug_str(goal))) } -pub fn kb_lookup(args: &Value) -> Result { - let kb = arg_text(args, "kb_dir")?; +pub fn kb_canonicalize(args: &Value) -> Result { let goal = arg_text(args, "goal")?; - let goal_n = normalize_goal(goal); - let query_sig = goal_sig_hex(&material_for(goal, opt_tools(args))); - let slug_hint = goal_slug_str(goal); - let root = resolve_path(kb); - let mut match_kind = "exact"; - let resolved = resolve_skill_paths(&root, &query_sig, &slug_hint).or_else(|| { - match_kind = "alias"; - resolve_skill_paths_by_alias(&root, &goal_n) - }); - let Some((_task, skill_path, _res, slug)) = resolved else { - return Ok(Value::Null); - }; + Ok(Value::String(canonicalize_goal_str(goal))) +} +fn lookup_hit_payload( + kb: &str, + root: &Path, + skill_path: &Path, + slug: &str, + match_kind: &str, + query_sig: &str, +) -> Result { let skill_src = - fs::read_to_string(&skill_path).map_err(|e| format!("kb_lookup read skill: {e}"))?; + fs::read_to_string(skill_path).map_err(|e| format!("kb_lookup read skill: {e}"))?; let mut status = extract_fm_field(&skill_src, "status").unwrap_or_else(|| "stable".into()); if status == "deprecated" || status == "draft" { return Ok(Value::Null); @@ -592,7 +629,7 @@ pub fn kb_lookup(args: &Value) -> Result { let hits = extract_fm_field(&skill_src, "hits") .and_then(|s| s.parse::().ok()) .unwrap_or(0); - let skill_sig = extract_fm_field(&skill_src, "sig").unwrap_or_else(|| query_sig.clone()); + let skill_sig = extract_fm_field(&skill_src, "sig").unwrap_or_else(|| query_sig.to_string()); let res_path = root.join("resources").join(format!("{slug}.mq.md")); if res_path.is_file() { @@ -624,15 +661,162 @@ pub fn kb_lookup(args: &Value) -> Result { })) } +fn resolve_skill_paths_by_slug( + root: &Path, + slug: &str, +) -> Option<(PathBuf, PathBuf, PathBuf, String)> { + let slug = slug.trim(); + if slug.is_empty() { + return None; + } + let task = root.join("concepts/tasks").join(format!("{slug}.md")); + let skill = root.join("concepts/skills").join(format!("{slug}.md")); + let res = root.join("resources").join(format!("{slug}.mq.md")); + if task.is_file() && skill.is_file() && res.is_file() { + Some((task, skill, res, slug.to_string())) + } else { + None + } +} + +pub fn kb_lookup(args: &Value) -> Result { + let kb = arg_text(args, "kb_dir")?; + let root = resolve_path(kb); + let tools = opt_tools(args); + + // Soft-match REUSE: resolve by explicit slug. + if let Some(slug) = args + .get("slug") + .and_then(|v| v.as_str()) + .map(str::trim) + .filter(|s| !s.is_empty()) + { + let Some((_task, skill_path, _res, slug)) = resolve_skill_paths_by_slug(&root, slug) else { + return Ok(Value::Null); + }; + let query_sig = args + .get("goal") + .and_then(|v| v.as_str()) + .map(|g| goal_sig_hex(&material_for(g, tools))) + .unwrap_or_default(); + return lookup_hit_payload(kb, &root, &skill_path, &slug, "soft", &query_sig); + } + + let goal = arg_text(args, "goal")?; + let goal_n = normalize_goal(goal); + let query_sig = goal_sig_hex(&material_for(goal, tools)); + let slug_hint = goal_slug_str(goal); + let mut match_kind = "exact"; + let resolved = resolve_skill_paths(&root, &query_sig, &slug_hint) + .or_else(|| { + match_kind = "alias"; + resolve_skill_paths_by_alias(&root, &goal_n) + }) + .or_else(|| { + let canon = canonicalize_goal_str(goal); + if canon.is_empty() || canon == goal_n { + return None; + } + match_kind = "canonical"; + let sig2 = goal_sig_hex(&material_for(&canon, tools)); + let slug2 = goal_slug_str(&canon); + resolve_skill_paths(&root, &sig2, &slug2) + .or_else(|| resolve_skill_paths_by_alias(&root, &canon)) + }); + let Some((_task, skill_path, _res, slug)) = resolved else { + return Ok(Value::Null); + }; + + lookup_hit_payload(kb, &root, &skill_path, &slug, match_kind, &query_sig) +} + +pub fn kb_list_tasks(args: &Value) -> Result { + let kb = arg_text(args, "kb_dir")?; + let limit = match args.get("limit") { + None | Some(Value::Null) => 40usize, + Some(Value::Number(n)) => n.as_u64().unwrap_or(40) as usize, + Some(Value::String(s)) => s.parse().unwrap_or(40), + _ => 40, + } + .clamp(1, 200); + let root = resolve_path(kb); + let tasks_dir = root.join("concepts/tasks"); + let mut tasks: Vec = Vec::new(); + if tasks_dir.is_dir() { + let mut ents: Vec<_> = fs::read_dir(&tasks_dir) + .map_err(|e| format!("list tasks: {e}"))? + .flatten() + .collect(); + ents.sort_by_key(|e| e.file_name()); + for ent in ents { + if tasks.len() >= limit { + break; + } + let p = ent.path(); + if p.extension().and_then(|e| e.to_str()) != Some("md") { + continue; + } + let Ok(src) = fs::read_to_string(&p) else { + continue; + }; + let slug = p + .file_stem() + .map(|s| s.to_string_lossy().into_owned()) + .unwrap_or_default(); + if slug.is_empty() { + continue; + } + let title = extract_fm_field(&src, "title").unwrap_or_else(|| slug.clone()); + tasks.push(json!({ "slug": slug, "title": title })); + } + } + let count = tasks.len() as i64; + Ok(json!({ "tasks": tasks, "count": count })) +} + +pub fn kb_add_alias(args: &Value) -> Result { + let kb = arg_text(args, "kb_dir")?; + let slug = arg_text(args, "slug")?; + let alias = arg_text(args, "alias")?; + let alias_n = normalize_goal(alias); + if alias_n.is_empty() { + return Ok(json!({ "ok": false, "reason": "empty_alias" })); + } + let root = resolve_path(kb); + let path = root.join("concepts/tasks").join(format!("{slug}.md")); + if !path.is_file() { + return Ok(json!({ "ok": false, "reason": "missing_task" })); + } + let src = fs::read_to_string(&path).map_err(|e| format!("read task: {e}"))?; + let sig = extract_fm_field(&src, "sig").unwrap_or_default(); + let goal = extract_fm_field(&src, "description") + .or_else(|| extract_fm_field(&src, "title")) + .unwrap_or_default(); + let now = now_rfc3339(); + write_task_concept(&path, &sig, slug, &goal, &now, &[alias_n.clone()])?; + Ok(json!({ "ok": true, "slug": slug, "alias": alias_n })) +} + pub fn kb_record_hit(args: &Value) -> Result { let kb = arg_text(args, "kb_dir")?; let goal = arg_text(args, "goal")?; let goal_n = normalize_goal(goal); - let sig = goal_sig_hex(&material_for(goal, opt_tools(args))); + let tools = opt_tools(args); + let sig = goal_sig_hex(&material_for(goal, tools)); let slug_hint = goal_slug_str(goal); let root = resolve_path(kb); let Some((_t, skill_path, _r, slug)) = resolve_skill_paths(&root, &sig, &slug_hint) .or_else(|| resolve_skill_paths_by_alias(&root, &goal_n)) + .or_else(|| { + let canon = canonicalize_goal_str(goal); + if canon.is_empty() || canon == goal_n { + return None; + } + let sig2 = goal_sig_hex(&material_for(&canon, tools)); + let slug2 = goal_slug_str(&canon); + resolve_skill_paths(&root, &sig2, &slug2) + .or_else(|| resolve_skill_paths_by_alias(&root, &canon)) + }) else { return Ok(Value::Null); }; @@ -910,4 +1094,12 @@ mod tests { assert_eq!(a, goal_sig_hex("hello")); assert_eq!(a.len(), 12); } + + #[test] + fn canonicalize_strips_standing_prefix() { + let c = canonicalize_goal_str("你是一个智能体,帮我规划明天的行程"); + assert_eq!(c, "帮我规划明天的行程"); + let c2 = canonicalize_goal_str("你是一个智能体,帮我规划明天的行程?"); + assert_eq!(c2, "帮我规划明天的行程"); + } } diff --git a/plugins/agent/src/lib.rs b/plugins/agent/src/lib.rs index 8e2d29d..2ef48ad 100644 --- a/plugins/agent/src/lib.rs +++ b/plugins/agent/src/lib.rs @@ -877,7 +877,10 @@ macro_rules! kb_ffi { kb_ffi!(agent_goal_sig, kb::goal_sig, "agent_goal_sig"); kb_ffi!(agent_goal_slug, kb::goal_slug, "agent_goal_slug"); +kb_ffi!(agent_kb_canonicalize, kb::kb_canonicalize, "agent_kb_canonicalize"); kb_ffi!(agent_kb_lookup, kb::kb_lookup, "agent_kb_lookup"); +kb_ffi!(agent_kb_list_tasks, kb::kb_list_tasks, "agent_kb_list_tasks"); +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_record_hit, kb::kb_record_hit, "agent_kb_record_hit"); kb_ffi!( @@ -962,9 +965,18 @@ pub unsafe extern "C" fn marqdo_plugin_init(host: *const MarqdoHostApi) -> c_int if register(host, "agent_goal_slug", "goal", agent_goal_slug) != 0 { return 1; } + if register(host, "agent_kb_canonicalize", "goal", agent_kb_canonicalize) != 0 { + return 1; + } if register(host, "agent_kb_lookup", "kb_dir,goal", agent_kb_lookup) != 0 { return 1; } + if register(host, "agent_kb_list_tasks", "kb_dir", agent_kb_list_tasks) != 0 { + return 1; + } + if register(host, "agent_kb_add_alias", "kb_dir,slug,alias", agent_kb_add_alias) != 0 { + return 1; + } if register( host, "agent_kb_promote", diff --git a/src/view/html.rs b/src/view/html.rs index a44edaa..e079908 100644 --- a/src/view/html.rs +++ b/src/view/html.rs @@ -644,6 +644,37 @@ section.block > h2 { color: #1d1d1f; letter-spacing: 0.02em; } +.stream-tool { + border: 1px solid var(--line); + border-left: 3px solid #94a3b8; + border-radius: 8px; + background: #f1f5f9; + padding: 0.45rem 0.7rem; + margin: 0.25rem 0; +} +.stream-tool-head { + display: flex; + flex-wrap: wrap; + align-items: baseline; + gap: 0.35rem 0.65rem; + font: 0.78rem/1.3 var(--sans); + color: var(--muted); +} +.stream-tool-head strong { + color: #334155; + font-weight: 600; +} +.stream-tool-meta { + font: 0.72rem/1.3 var(--mono); + color: #64748b; +} +.stream-tool-body { + margin-top: 0.3rem; + font: 0.78rem/1.4 var(--mono); + color: #475569; + white-space: pre-wrap; + word-break: break-word; +} .stream-child { border: 1px solid var(--line); border-left: 3px solid #2563eb; @@ -1156,6 +1187,35 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String { resetTurnRefs(); scrollIfPinned(); }} + function addToolCard(ev) {{ + flushPendingAll(); + ensureTurn(); + var card = document.createElement("div"); + card.className = "stream-tool"; + var head = document.createElement("div"); + head.className = "stream-tool-head"; + var title = document.createElement("strong"); + var phase = (ev.type === "tool_end") ? "Tool end" : "Tool start"; + title.textContent = phase; + head.appendChild(title); + var meta = document.createElement("span"); + meta.className = "stream-tool-meta"; + var bits = []; + if (ev.name) bits.push(ev.name); + if (ev.kind) bits.push(ev.kind); + meta.textContent = bits.join(" · "); + head.appendChild(meta); + card.appendChild(head); + if (ev.type === "tool_end" && ev.result != null && ev.result !== "") {{ + var body = document.createElement("div"); + body.className = "stream-tool-body"; + var res = formatResult(ev.result); + body.textContent = res.length > 1024 ? res.slice(0, 1024) + "…" : res; + card.appendChild(body); + }} + turn.appendChild(card); + scrollIfPinned(); + }} function addChildCard(ev) {{ flushPendingAll(); var card = document.createElement("div"); @@ -1209,6 +1269,8 @@ pub fn layout(title: &str, nav: &str, main: &str) -> String { if (status) status.textContent = "error"; }} else if (t === "decision") {{ addDecision(ev); + }} else if (t === "tool_start" || t === "tool_end") {{ + addToolCard(ev); }} else if (t === "round") {{ addChildCard(ev); }} else if (t === "run_start") {{ diff --git a/tests/ext/agent-kb-alias.mq.md b/tests/ext/agent-kb-alias.mq.md index ac39e03..e7486ce 100644 --- a/tests/ext/agent-kb-alias.mq.md +++ b/tests/ext/agent-kb-alias.mq.md @@ -62,3 +62,32 @@ description: Promote resource; patch Task aliases; variant goal hits same resour *`exact` = > agent_kb_lookup kb_dir=`kb` goal=`goal` * *`mk2` = > json.get value=`exact` key=match * > print text=`mk2` + +*`g_trip` = 帮我规划明天的行程 * +*`wb2` = .marqdo/agent-runs/alias-trip.mq.md * +*`body2` = "---\ntitle: trip\n---\n\n# main\n\n**trip-ok**\n" * +> fs.write_text path=`wb2` text=`body2` +*`prom2` = > agent_kb_promote kb_dir=`kb` goal=`g_trip` workbook=`wb2` * +*`ok2` = > json.get value=`prom2` key=promoted * +1. `ok2` + > print text=trip-promoted +2. * + > print text=trip-promote-fail + +*`g_canon` = 你是一个智能体,帮我规划明天的行程 * +*`c` = > agent_kb_canonicalize goal=`g_canon` * +> print text=`c` +*`hit_c` = > agent_kb_lookup kb_dir=`kb` goal=`g_canon` * +1. `hit_c` + > print text=canonical-ok +2. * + > print text=canonical-miss +*`mk3` = > json.get value=`hit_c` key=match * +> print text=`mk3` + +*`listed` = > agent_kb_list_tasks kb_dir=`kb` * +*`tc` = > json.get value=`listed` key=count * +1. `tc` > 1 + > print text=list-ok +2. * + > print text=list-bad diff --git a/tests/ext/agent-lib-call.mq.md b/tests/ext/agent-lib-call.mq.md new file mode 100644 index 0000000..1b86130 --- /dev/null +++ b/tests/ext/agent-lib-call.mq.md @@ -0,0 +1,45 @@ +--- +title: Parent CALL:lib whitelist (offline) +description: lib.fs.exists + ARGS; reject lib.net; catalog callable. +> ext/ai/agent.mq.md +> lib/json.mq.md +> lib/fs.mq.md +> lib/sys.mq.md +--- + +# main + +*`parts` = > json.parse text={"a":"CALL:lib.fs.exists","b":"ARGS:{\"path\":\".\"}","nl":"\n"} * +*`a` = > json.get value=`parts` key=a * +*`b` = > json.get value=`parts` key=b * +*`nl` = > json.get value=`parts` key=nl * +*`t_ok` = `a` + `nl` + `b` + `nl` * +*`act` = > agent.extract_plan_act reply=`t_ok` * +*`kind` = > json.get value=`act` key=kind * +*`name` = > json.get value=`act` key=name * +> print text=`kind` +> print text=`name` + +*`out` = > agent.run_parent_tool name=`name` path=. reply=`t_ok` * +1. `out` + > print text=exists-ok +2. * + > print text=exists-bad + +*`deny` = > agent.run_parent_tool name=lib.net.get path=. reply=CALL:lib.net.get * +*`sep` = > json.parse text={"s":"not allowed"} * +*`s` = > json.get value=`sep` key=s * +*`dparts` = > split value=`deny` sep=`s` * +*`pn` = > len value=`dparts` * +1. `pn` > 1 + > print text=deny-ok +2. * + > print text=deny-bad + +*`cat` = > agent.lib_catalog * +*`callable` = > json.get value=`cat` key=callable * +*`cn` = > len value=`callable` * +1. `cn` > 3 + > print text=callable-ok +2. * + > print text=callable-bad diff --git a/tests/ext/agent-soft-match.mq.md b/tests/ext/agent-soft-match.mq.md new file mode 100644 index 0000000..badd6a0 --- /dev/null +++ b/tests/ext/agent-soft-match.mq.md @@ -0,0 +1,34 @@ +--- +title: OKF soft_match protocol parse (offline) +description: REUSE/NEW + SLUG extractors; no network. +> ext/ai/agent.mq.md +> lib/json.mq.md +--- + +# main + +*`samples` = > json.parse text={"reuse":"DECISION: REUSE\nSLUG: trip-plan\n","new":"DECISION: NEW\n","zh":"决定:复用\n标识:行程-a\n"} * +*`r1` = > json.get value=`samples` key=reuse * +*`d1` = > agent.extract_plan_decision reply=`r1` * +> print text=`d1` +*`s1` = > agent.extract_soft_match_slug reply=`r1` * +> print text=`s1` + +*`r2` = > json.get value=`samples` key=new * +*`d2` = > agent.extract_plan_decision reply=`r2` * +> print text=`d2` + +*`r3` = > json.get value=`samples` key=zh * +*`d3` = > agent.extract_plan_decision reply=`r3` * +> print text=`d3` +*`s3` = > agent.extract_soft_match_slug reply=`r3` * +> print text=`s3` + +*`tasks` = > json.parse text=[{"slug":"a","title":"Alpha"},{"slug":"b","title":"Beta"}] * +*`prompt` = > agent.build_soft_match_prompt goal=demo tasks=`tasks` * +*`has` = > split value=`prompt` sep=DECISION: REUSE * +*`n` = > len value=`has` * +1. `n` > 1 + > print text=prompt-ok +2. * + > print text=prompt-bad diff --git a/tests/gold.rs b/tests/gold.rs index 56f60ae..34b8bea 100644 --- a/tests/gold.rs +++ b/tests/gold.rs @@ -1423,7 +1423,37 @@ alias-written alias-ok alias same-slug -exact", +exact +trip-promoted +帮我规划明天的行程 +canonical-ok +canonical +list-ok", + ); +} + +#[test] +fn ext_agent_soft_match() { + assert_out( + "tests/ext/agent-soft-match.mq.md", + "REUSE +trip-plan +NEW +REUSE +行程-a +prompt-ok", + ); +} + +#[test] +fn ext_agent_lib_call() { + assert_out( + "tests/ext/agent-lib-call.mq.md", + "call +lib.fs.exists +exists-ok +deny-ok +callable-ok", ); }