autosnippet 2.9.0 → 2.10.0

package/skills/autosnippet-analysis/SKILL.md

@@ -73,7 +73,7 @@ description: Deep project analysis — full scan + semantic field enrichment + g
 }
 ```
 
-Use `autosnippet_submit_candidates` for batch submission.
+Use `autosnippet_submit_knowledge_batch` for batch submission.
 
 #### Phase 4: Guard Deep Audit (Optional)
 对特定文件调用 `autosnippet_guard_audit_files` 做深度规范审计。
@@ -101,7 +101,7 @@ Use `autosnippet_submit_candidates` for batch submission.
 ### Workflow for Enrichment
 
 **方式 A: 你（Agent）直接补全**
-在分析代码时，直接填写所有 6 个语义字段到 `autosnippet_submit_candidate` 调用中。
+在分析代码时，直接填写所有 6 个语义字段到 `autosnippet_submit_knowledge` 调用中。
 
 **方式 B: 对已有候选调用 AI 补全**
 ```
@@ -109,6 +109,13 @@ autosnippet_enrich_candidates({ candidateIds: ["id1", "id2", ...] })
 ```
 AI 会分析每条候选的代码，自动填充缺失字段（已有字段不会被覆盖）。
 
+**方式 C: Bootstrap 候选 AI 润色**
+```
+autosnippet_bootstrap_refine({ candidateIds: ["id1", "id2", ...] })
+```
+使用 AI 对 Bootstrap 候选做内容润色（改善 summary、补充 insight、推断 relations、调整 confidence）。
+必须在 `autosnippet_enrich_candidates` 之后调用（确保字段完整后再润色）。
+
 ---
 
 ## Capability 3: 候选 → Recipe 查漏补缺
@@ -119,11 +126,12 @@ AI 会分析每条候选的代码，自动填充缺失字段（已有字段不
 
 ### Workflow
 
-1. 调用 `autosnippet_enrich_candidates` 对目标候选做 AI 补全
-2. 调用 `autosnippet_validate_candidate` 检查结构完整性
-3. 调用 `autosnippet_check_duplicate` 检查重复
-4. 报告补全情况 + 缺失字段 + 重复风险
-5. 如缺失字段 AI 无法填充，提示用户手动补充
+1. 调用 `autosnippet_enrich_candidates` 对目标候选做结构补齐（诊断缺失字段）
+2. 调用 `autosnippet_bootstrap_refine` 对候选做 AI 润色（summary/insight/relations）
+3. 调用 `autosnippet_validate_candidate` 检查结构完整性
+4. 调用 `autosnippet_check_duplicate` 检查重复
+5. 报告补全情况 + 缺失字段 + 重复风险
+6. 如缺失字段 AI 无法填充，提示用户手动补充
 
 ### Recipe 必备字段检查清单
 

package/skills/autosnippet-candidates/SKILL.md

@@ -221,18 +221,18 @@ Severity: `error` (must fix) | `warning` (should fix) | `info` (suggestion)
 3. Call `autosnippet_context_search` to mark similarity and fill `relations`
 4. (Optional) `autosnippet_validate_candidate` pre-validate
 5. (Optional) `autosnippet_check_duplicate` dedup hint
-6. `autosnippet_submit_candidate` or `autosnippet_submit_candidates` to submit
+6. `autosnippet_submit_knowledge` or `autosnippet_submit_knowledge_batch` to submit
 
 ## Batch Target Scan Flow
 1. `autosnippet_get_targets` → select targetName
 2. `autosnippet_get_target_files(targetName)`
 3. Parallel scan → per-file checklist extraction → aggregate / dedup / score
-4. `autosnippet_submit_candidates` batch submit — **all V2 fields are preserved**
+4. `autosnippet_submit_knowledge_batch` batch submit — **all V2 fields are preserved**
 
 ## Draft File Flow (alternative)
 1. Create draft folder (e.g. `.autosnippet-drafts/`) outside `AutoSnippet/`
 2. Generate one .md per pattern in draft folder
-3. Call `autosnippet_submit_draft_recipes` with filePaths and `deleteAfterSubmit: true`
+3. Call `autosnippet_submit_knowledge_batch` with filePaths and `deleteAfterSubmit: true`
 4. Delete draft folder after submit
 
 ---
@@ -247,9 +247,9 @@ Severity: `error` (must fix) | `warning` (should fix) | `info` (suggestion)
 | `autosnippet_context_search` | Find similar existing Recipes → fill `relations` |
 | `autosnippet_validate_candidate` | Pre-validate candidate fields |
 | `autosnippet_check_duplicate` | Cosine similarity dedup check |
-| `autosnippet_submit_candidate` | Submit single candidate (**all V2 fields**) |
-| `autosnippet_submit_candidates` | Batch submit candidates (**all V2 fields preserved**) |
-| `autosnippet_submit_draft_recipes` | Submit .md draft files as candidates |
+| `autosnippet_submit_knowledge` | Submit single candidate (**all V2 fields**) |
+| `autosnippet_submit_knowledge_batch` | Batch submit candidates (**all V2 fields preserved**) |
+| `autosnippet_submit_knowledge_batch` | Submit .md draft files as candidates |
 
 ---
 

package/skills/autosnippet-coldstart/SKILL.md

@@ -226,7 +226,7 @@ description: Cold-start knowledge base initialization. Full 9-dimension analysis
 
 ## Phase 5: 批量提交
 
-将所有分析结果通过 `autosnippet_submit_candidates` 批量提交：
+将所有分析结果通过 `autosnippet_submit_knowledge_batch` 批量提交：
 
 ```json
 {
@@ -448,12 +448,16 @@ description: Cold-start knowledge base initialization. Full 9-dimension analysis
 | Tool | 用途 |
 |------|------|
 | `autosnippet_bootstrap_knowledge` | 启动冷启动扫描（本 Skill 核心工具） |
-| `autosnippet_submit_candidates` | 批量提交候选 |
-| `autosnippet_submit_candidate` | 提交单条候选 |
+| `autosnippet_enrich_candidates` | ① 结构补齐：诊断候选字段缺失 |
+| `autosnippet_bootstrap_refine` | ② 内容润色：AI 精炼候选质量（summary/insight/relations） |
+| `autosnippet_submit_knowledge_batch` | 批量提交候选 |
+| `autosnippet_submit_knowledge` | 提交单条候选 |
 | `autosnippet_validate_candidate` | 校验候选字段 |
 | `autosnippet_check_duplicate` | 去重检查 |
 | `autosnippet_context_search` | 查找已有知识（避免重复） |
 | `autosnippet_scan_project` | 轻量探查（不写库） |
+| `autosnippet_list_skills` | 列出可用 Skill 列表 |
+| `autosnippet_load_skill` | 加载指定 Skill 文档获取指引 |
 
 ## Related Skills
 

package/skills/autosnippet-concepts/SKILL.md

@@ -87,13 +87,14 @@ This is a conceptual map. Skills stay semantic; MCP provides capability.
 | 知识图谱 | `autosnippet_graph_query`, `autosnippet_graph_impact`, `autosnippet_graph_path`, `autosnippet_graph_stats` |
 | 候选预检 | `autosnippet_validate_candidate` |
 | 去重建议 | `autosnippet_check_duplicate` |
-| 候选提交 | `autosnippet_submit_candidate`, `autosnippet_submit_candidates`, `autosnippet_submit_draft_recipes` |
+| 候选提交 | `autosnippet_submit_knowledge`, `autosnippet_submit_knowledge_batch`, `autosnippet_submit_knowledge_batch` |
 | AI 补全 | `autosnippet_enrich_candidates` |
 | Guard 检查 | `autosnippet_guard_check`, `autosnippet_guard_audit_files` |
 | 合规报告 | `autosnippet_compliance_report`, `autosnippet_recipe_insights` |
 | 使用确认 | `autosnippet_confirm_usage` |
 | 项目扫描 | `autosnippet_scan_project` |
-| 冷启动 | `autosnippet_bootstrap_knowledge` |
+| 冷启动 | `autosnippet_bootstrap_knowledge`, `autosnippet_bootstrap_refine` |
+| Skills 管理 | `autosnippet_list_skills`, `autosnippet_load_skill`, `autosnippet_create_skill`, `autosnippet_update_skill`, `autosnippet_delete_skill`, `autosnippet_suggest_skills` |
 | 自检/能力 | `autosnippet_health`, `autosnippet_capabilities` |
 
 ### Failure Handling (Examples)
@@ -336,7 +337,7 @@ When creating or extracting Recipes:
 
 ### Candidate-only Rule (重要)
 
-- **If the user asks for candidates**: Extract structured items and submit via MCP **`autosnippet_submit_candidates`**.
+- **If the user asks for candidates**: Extract structured items and submit via MCP **`autosnippet_submit_knowledge_batch`**.
 
 ---
 
@@ -376,7 +377,7 @@ When `asd ui` is running in the project root, use the HTTP API for on-demand sem
 | Capability | Description | Skill |
 |------------|-------------|-------|
 | **Recipe lookup** | Read `references/project-recipes-context.md` 轻量索引，需全文调 MCP `autosnippet_get_recipe(id)` / `autosnippet_context_search`. Recipe over source | autosnippet-recipes |
-| **Create Recipe** | Dashboard New Recipe; or write to `_draft_recipe.md` and watch auto-adds; or MCP `autosnippet_submit_draft_recipes` | autosnippet-create |
+| **Create Recipe** | Dashboard New Recipe; or write to `_draft_recipe.md` and watch auto-adds; or MCP `autosnippet_submit_knowledge_batch` | autosnippet-create |
 | **Search & insert** | `ass` shortcut or `// as:search`, `asd search`, Dashboard search | autosnippet-search |
 | **Audit review** | `// as:audit`; watch runs AI review against knowledge base | autosnippet-guard |
 | **Dependency graph** | `AutoSnippet/AutoSnippet.spmmap.json`; `asd spm-map` to update; MCP graph tools for querying | autosnippet-structure |
@@ -593,8 +594,8 @@ authority: 3
 
 ### How to add new knowledge
 
-1. **Single code / single Recipe**: Copy to clipboard → open Dashboard (run `asd ui` if not running) → Use Copied Code, paste, review, save; or write `_draft_recipe.md` and let watch auto-add to Candidates. Or use `autosnippet_submit_draft_recipes` via MCP.
-2. **Multiple drafts (recommended)**: Create a **draft folder** (e.g. `.autosnippet-drafts`), **one .md file per Recipe**—do not put everything in one big file. Call MCP **`autosnippet_submit_draft_recipes`** with those file paths to submit to Candidates, then review in Dashboard **Candidates**. **After submit, delete the draft folder** (use `deleteAfterSubmit: true` or `rm -rf .autosnippet-drafts`).
+1. **Single code / single Recipe**: Copy to clipboard → open Dashboard (run `asd ui` if not running) → Use Copied Code, paste, review, save; or write `_draft_recipe.md` and let watch auto-add to Candidates. Or use `autosnippet_submit_knowledge_batch` via MCP.
+2. **Multiple drafts (recommended)**: Create a **draft folder** (e.g. `.autosnippet-drafts`), **one .md file per Recipe**—do not put everything in one big file. Call MCP **`autosnippet_submit_knowledge_batch`** with those file paths to submit to Candidates, then review in Dashboard **Candidates**. **After submit, delete the draft folder** (use `deleteAfterSubmit: true` or `rm -rf .autosnippet-drafts`).
 3. **Intro-only docs**: Recipe candidates can be intro-only (frontmatter + usage guide, no code); after approval they become Recipes and **do not generate a Snippet**—used only for search and Guard context.
 
 ### How to use knowledge once it’s in the base

package/skills/autosnippet-create/SKILL.md

@@ -12,8 +12,8 @@ This skill tells the agent how to **submit module usage code** (that Cursor has
 ## Instructions for the agent (read this first)
 
 1. **Goal**: When you (Cursor) have **finished writing or refining** module usage code, or the user says "把这段提交到 web / 加入知识库", guide them to **submit that code to the Dashboard** so it becomes a **Recipe** in `AutoSnippet/recipes/`.
-2. **Draft workflow**: Prefer **creating a draft folder** (e.g. `.autosnippet-drafts`), **one .md file per Recipe**, multiple draft files—**do not use one big file**. Call MCP **`autosnippet_submit_draft_recipes`** with filePaths (the .md files in the draft folder) to submit to Candidates. **After submit, delete the draft folder** (use `deleteAfterSubmit: true` or run `rm -rf .autosnippet-drafts`). Single-item flow can use `_draft_recipe.md` and watch will auto-add to Candidates.
-3. **When user asks for "candidates"**: Use MCP **`autosnippet_submit_candidates`** for structured items (title/summary/trigger/code/usageGuide); use **`autosnippet_submit_draft_recipes`** for Markdown draft files (prefer draft folder + multiple files; delete draft folder after submit).
+2. **Draft workflow**: Prefer **creating a draft folder** (e.g. `.autosnippet-drafts`), **one .md file per Recipe**, multiple draft files—**do not use one big file**. Call MCP **`autosnippet_submit_knowledge_batch`** with filePaths (the .md files in the draft folder) to submit to Candidates. **After submit, delete the draft folder** (use `deleteAfterSubmit: true` or run `rm -rf .autosnippet-drafts`). Single-item flow can use `_draft_recipe.md` and watch will auto-add to Candidates.
+3. **When user asks for "candidates"**: Use MCP **`autosnippet_submit_knowledge_batch`** for structured items (title/summary/trigger/code/usageGuide); use **`autosnippet_submit_knowledge_batch`** for Markdown draft files (prefer draft folder + multiple files; delete draft folder after submit).
 4. **One Recipe = one scenario**: If you are drafting content, **split** into multiple Recipes by scenario. Never combine multiple usage patterns into one Recipe file or one candidate.
 5. **Recipe candidates can be intro-only**: Intro-only docs (no code block) can be submitted as candidates; after approval they become Recipes and **do not generate a Snippet**—used only for search and Guard context.
 6. **MUST follow standard Recipe format**: Use the complete template from **autosnippet-concepts** skill. Include all required fields: frontmatter with `title`, `trigger`, `category` (one of 8 standard values), `language` (must be `swift` or `objectivec`), `summary_cn`, `summary_en`, `headers` (complete import statements), plus `## Snippet / Code Reference` and `## AI Context / Usage Guide` section.
@@ -51,11 +51,11 @@ This skill tells the agent how to **submit module usage code** (that Cursor has
      ```
    - See [templates/recipes-setup/README.md](../../templates/recipes-setup/README.md) for detailed format guide & examples.
 7. **Auto-fill headers from project context**: Before submitting, **check `references/project-recipes-context.md`** (轻量索引) to find similar Recipes by title/trigger/category, then call MCP **`autosnippet_get_recipe(id)`** to get full content including headers. Copy the exact import format for `headers` field. If needed, call MCP **`autosnippet_context_search`** with the module name to find similar Recipes and extract their header patterns. This ensures consistency and correctness.
-8. **Primary flow (MCP preferred)**: Code is ready → Agent writes to `_draft_recipe.md` or calls `autosnippet_submit_draft_recipes` / `autosnippet_submit_candidates` → candidates appear in Dashboard Candidates → user reviews and approves → Recipe is added to the knowledge base.
+8. **Primary flow (MCP preferred)**: Code is ready → Agent writes to `_draft_recipe.md` or calls `autosnippet_submit_knowledge_batch` / `autosnippet_submit_knowledge_batch` → candidates appear in Dashboard Candidates → user reviews and approves → Recipe is added to the knowledge base.
 9. **Alternative (Dashboard browser)**: Code is ready → user opens Dashboard (`asd ui` running) → **New Recipe** → **Use Copied Code** (paste the code) → AI fills title/summary/trigger/headers → **user reviews and approves** → saved to knowledge base.
 10. **Alternative (in editor)**: User adds **`// as:create`** in the source file, copies the code, saves → **watch** (from `asd watch` or `asd ui`) auto-adds to Candidates → user opens Dashboard **Candidates** to review and save.
 9. **Draft & clipboard auto-add**: When you write to **`_draft_recipe.md`** (project root) or user uses **`// as:create`** with clipboard content, **watch** automatically reads the draft/clipboard, adds it to **Candidates** (target `_draft` or `_watch`), and shows a **friendly prompt** (e.g. "已创建候选「xxx」，请在 Candidates 页审核" in notification and console). User only needs to open Dashboard **Candidates** to review and save — no manual copy-paste required.
-10. **Multiple recipes**: Prefer **one .md file per Recipe** in a draft folder (e.g. `.autosnippet-drafts/`), call **`autosnippet_submit_draft_recipes`** with the list of file paths, then **delete the draft folder** after submit. Do not use one big file for many Recipes.
+10. **Multiple recipes**: Prefer **one .md file per Recipe** in a draft folder (e.g. `.autosnippet-drafts/`), call **`autosnippet_submit_knowledge_batch`** with the list of file paths, then **delete the draft folder** after submit. Do not use one big file for many Recipes.
 11. **Project root** = directory with `AutoSnippet/AutoSnippet.boxspec.json`. All commands run from the project root.
 
 ---
@@ -68,7 +68,7 @@ This skill tells the agent how to **submit module usage code** (that Cursor has
 
 - **Code scenario**: Usage code is in current file or clipboard. If not copied, prompt user to copy the code block you provide.
 ...rated a full Recipe (frontmatter, Snippet, Usage Guide), **prefer writing to draft file** `_draft_recipe.md` at project root. On save, **watch automatically reads the draft**, adds it to **Candidates** (target `_draft`), and shows a **friendly prompt** ("已创建候选「xxx」，请在 Candidates 页审核"). User opens Dashboard **Candidates** to review and save — no manual copy needed. Or output in copyable format in chat and guide user to copy → Dashboard → Use Copied Code → paste → review → save. **Do not write to `AutoSnippet/recipes/` or `AutoSnippet/snippets/`.**
-- **Candidate output rule**: When the user asks for candidates, **do not create files under `AutoSnippet/`**. Use **`autosnippet_submit_candidates`** for structured items; use **`autosnippet_submit_draft_recipes`** for draft .md files (prefer draft folder + multiple files; delete draft folder after submit).
+- **Candidate output rule**: When the user asks for candidates, **do not create files under `AutoSnippet/`**. Use **`autosnippet_submit_knowledge_batch`** for structured items; use **`autosnippet_submit_knowledge_batch`** for draft .md files (prefer draft folder + multiple files; delete draft folder after submit).
 
 ### Step 2: Open the web (Dashboard) in browser
 
@@ -78,7 +78,7 @@ This skill tells the agent how to **submit module usage code** (that Cursor has
 ### Step 3: Submit via Dashboard
 
 1. **Browser**: Run terminal command `open "http://localhost:3000/?action=create&source=clipboard"` (macOS) to open the browser.
-2. **Draft file (preferred for Agent)**: Write full Recipe MD to `_draft_recipe.md` at project root → watch auto-adds to Candidates. Or call MCP **`autosnippet_submit_draft_recipes`** with draft file paths.
+2. **Draft file (preferred for Agent)**: Write full Recipe MD to `_draft_recipe.md` at project root → watch auto-adds to Candidates. Or call MCP **`autosnippet_submit_knowledge_batch`** with draft file paths.
 3. **Manual fallback**: User opens **`http://localhost:3000`** in browser → **New Recipe** → **Use Copied Code** → paste.
 3. Pasted code: **Full Recipe MD** (with `---` frontmatter, `## Snippet / Code Reference`, `## AI Context / Usage Guide`) is parsed directly, **no AI rewrite**. Plain code still goes through AI analysis and fill.
 4. **User reviews and approves** — 人工审核 title/summary/category/trigger 及内容，确认无误后再保存。
@@ -92,7 +92,7 @@ This skill tells the agent how to **submit module usage code** (that Cursor has
 
 ## Alternative: Copy then jump to Web (Xcode and Cursor)
 
-**Option A — MCP Submit (Cursor 首选)**: After user confirms code, Agent calls **`autosnippet_submit_draft_recipes`** (for .md drafts) or **`autosnippet_submit_candidates`** (for structured items) to submit directly to Candidates. User reviews in Dashboard.
+**Option A — MCP Submit (Cursor 首选)**: After user confirms code, Agent calls **`autosnippet_submit_knowledge_batch`** (for .md drafts) or **`autosnippet_submit_knowledge_batch`** (for structured items) to submit directly to Candidates. User reviews in Dashboard.
 
 **Option B — Browser (Dashboard)**: Run `open "http://localhost:3000/?action=create&source=clipboard"` to open Dashboard New Recipe page; page reads clipboard and fills.
 
@@ -148,9 +148,9 @@ Watch parses all such blocks and adds each as a separate candidate; prompt may s
 
 | User / Cursor situation | Action |
 |-------------------------|--------|
-| "把这段提交到 web / 加入知识库" (code just written) | 1) 提示用户复制代码；2) 写入 `_draft_recipe.md` 或调用 **`autosnippet_submit_draft_recipes`**；3) 若需 Dashboard，运行 `open "http://localhost:3000/?action=create&source=clipboard"`；4) 用户粘贴并保存。 |
+| "把这段提交到 web / 加入知识库" (code just written) | 1) 提示用户复制代码；2) 写入 `_draft_recipe.md` 或调用 **`autosnippet_submit_knowledge_batch`**；3) 若需 Dashboard，运行 `open "http://localhost:3000/?action=create&source=clipboard"`；4) 用户粘贴并保存。 |
 | Agent 起草了 Recipe/Snippet 内容 | **Prefer**: Write to `_draft_recipe.md` (project root) → save → **watch 自动读取并加入候选池**，并给出友好提示 → 用户打开 Dashboard **Candidates** 审核并保存。无需手动复制。 |
-| Code in current file, want to submit from editor | **MCP**：`autosnippet_submit_draft_recipes`（写草稿 .md 后提交）。或 Add **`// as:create`**, copy, save；确保 **`asd watch`** / **`asd ui`** 运行；有剪贴板时自动加入候选并友好提示。 |
+| Code in current file, want to submit from editor | **MCP**：`autosnippet_submit_knowledge_batch`（写草稿 .md 后提交）。或 Add **`// as:create`**, copy, save；确保 **`asd watch`** / **`asd ui`** 运行；有剪贴板时自动加入候选并友好提示。 |
 | Code already in a file (path known) | 打开 **`http://localhost:3000`** → New Recipe → enter path → **Scan File** → review → save. |
 | Agent 起草内容无法完整复制 | Write to `_draft_recipe.md` → save → **watch 自动读取草稿、加入候选并友好提示** → 用户打开 **Candidates** 审核保存。 |
 | Batch from Target | **`asd ais <Target>`** → 打开 **`http://localhost:3000`** → **Candidates** → approve. |
@@ -162,13 +162,13 @@ Watch parses all such blocks and adds each as a separate candidate; prompt may s
 | Tool | Use |
 |------|-----|
 | `autosnippet_context_search` | On-demand semantic search of knowledge base; pass `query`, `limit?` |
-| `autosnippet_submit_draft_recipes` | Submit draft .md as candidates: prefer draft folder + multiple files (not one big file); supports intro-only docs (no code—no Snippet). Pass `filePaths`, optional `targetName`, `deleteAfterSubmit`. **Delete the draft folder after submit** (e.g. `deleteAfterSubmit: true` or `rm -rf .autosnippet-drafts`). |
-| `autosnippet_submit_candidates` | Submit structured items (title, summary, trigger, language, code, usageGuide) for batch scan, etc. |
-| `autosnippet_submit_candidate` | Submit single structured candidate with full V2 fields. |
+| `autosnippet_submit_knowledge_batch` | Submit draft .md as candidates: prefer draft folder + multiple files (not one big file); supports intro-only docs (no code—no Snippet). Pass `filePaths`, optional `targetName`, `deleteAfterSubmit`. **Delete the draft folder after submit** (e.g. `deleteAfterSubmit: true` or `rm -rf .autosnippet-drafts`). |
+| `autosnippet_submit_knowledge_batch` | Submit structured items (title, summary, trigger, language, code, usageGuide) for batch scan, etc. |
+| `autosnippet_submit_knowledge` | Submit single structured candidate with full V2 fields. |
 | `autosnippet_validate_candidate` | Pre-validate candidate quality before submission. |
 | `autosnippet_check_duplicate` | Check for duplicate Recipes before submission. |
 
-**Fallback when Dashboard not open**: Agent can always use `autosnippet_submit_draft_recipes` or `autosnippet_submit_candidates` directly via MCP — no browser needed.
+**Fallback when Dashboard not open**: Agent can always use `autosnippet_submit_knowledge_batch` or `autosnippet_submit_knowledge_batch` directly via MCP — no browser needed.
 
 ---
 

package/skills/autosnippet-intent/SKILL.md

@@ -42,10 +42,11 @@ Use this skill when the user's intent is unclear or overlaps multiple capabiliti
 - Insights: `autosnippet_recipe_insights`, `autosnippet_compliance_report`
 - Graph: `autosnippet_graph_query`, `autosnippet_graph_impact`, `autosnippet_graph_path`, `autosnippet_graph_stats`
 - Structure: `autosnippet_get_targets`, `autosnippet_get_target_files`, `autosnippet_get_target_metadata`
-- Candidates: `autosnippet_validate_candidate`, `autosnippet_check_duplicate`, `autosnippet_submit_candidate`, `autosnippet_submit_candidates`, `autosnippet_submit_draft_recipes`, `autosnippet_enrich_candidates`
+- Candidates: `autosnippet_validate_candidate`, `autosnippet_check_duplicate`, `autosnippet_submit_knowledge`, `autosnippet_submit_knowledge_batch`, `autosnippet_submit_knowledge_batch`, `autosnippet_enrich_candidates`
 - Guard: `autosnippet_guard_check`, `autosnippet_guard_audit_files`, `autosnippet_scan_project`
 - Telemetry: `autosnippet_confirm_usage`
-- Bootstrap: `autosnippet_bootstrap_knowledge`
+- Bootstrap: `autosnippet_bootstrap_knowledge`, `autosnippet_bootstrap_refine`
+- Skills: `autosnippet_list_skills`, `autosnippet_load_skill`, `autosnippet_create_skill`, `autosnippet_update_skill`, `autosnippet_delete_skill`, `autosnippet_suggest_skills`
 
 This skill is a router only; it does not perform actions itself.
 

package/skills/autosnippet-lifecycle/SKILL.md

@@ -37,9 +37,9 @@ This skill documents the Recipe lifecycle and clarifies what Agent can and canno
 ## Agent Role: What You CAN Do
 
 ### 1. Submit Candidates
-- `autosnippet_submit_candidate` - single structured candidate
-- `autosnippet_submit_candidates` - batch structured candidates
-- `autosnippet_submit_draft_recipes` - submit .md files as candidates
+- `autosnippet_submit_knowledge` - single structured candidate
+- `autosnippet_submit_knowledge_batch` - batch structured candidates
+- `autosnippet_submit_knowledge_batch` - submit .md files as candidates
 
 ### 2. Validate and Enhance Candidates
 - `autosnippet_validate_candidate` - pre-validate quality
@@ -81,7 +81,7 @@ This skill documents the Recipe lifecycle and clarifies what Agent can and canno
 2. Generate structured candidate with rationale, steps, etc.
 3. `autosnippet_check_duplicate` - verify no duplicate exists
 4. `autosnippet_validate_candidate` - pre-validate quality
-5. `autosnippet_submit_candidate` - submit to candidate pool
+5. `autosnippet_submit_knowledge` - submit to candidate pool
 6. Tell user: "Candidate submitted. Review in Dashboard Candidates page."
 
 ### Workflow 2: User asks about an existing Recipe
@@ -94,7 +94,7 @@ This skill documents the Recipe lifecycle and clarifies what Agent can and canno
 1. Explain: "I cannot modify Recipes directly. I can submit an improved version as a new candidate."
 2. `autosnippet_get_recipe` to read current content
 3. Generate improved candidate based on user's feedback
-4. `autosnippet_submit_candidate` - submit improved version
+4. `autosnippet_submit_knowledge` - submit improved version
 5. Tell user: "Improved version submitted as candidate. Update the Recipe in Dashboard."
 
 ---

package/skills/autosnippet-recipes/SKILL.md

@@ -78,7 +78,7 @@ Recipe is the core knowledge unit. V2 uses a unified structured model:
 
 ---
 
-## MCP Tools Reference (31 tools)
+## MCP Tools Reference (38 tools)
 
 ### Query (Agent can freely use)
 
@@ -109,9 +109,9 @@ Recipe is the core knowledge unit. V2 uses a unified structured model:
 
 | Tool | Description |
 |------|-------------|
-| autosnippet_submit_candidate | Submit single candidate (supports structured content) |
-| autosnippet_submit_candidates | Batch submit candidates |
-| autosnippet_submit_draft_recipes | Submit draft .md files as candidates |
+| autosnippet_submit_knowledge | Submit single candidate (supports structured content) |
+| autosnippet_submit_knowledge_batch | Batch submit candidates |
+| autosnippet_submit_knowledge_batch | Submit draft .md files as candidates |
 | autosnippet_validate_candidate | Validate candidate quality |
 | autosnippet_check_duplicate | Dedup check |
 | autosnippet_enrich_candidates | AI semantic field enrichment for candidates |
@@ -124,6 +124,18 @@ Recipe is the core knowledge unit. V2 uses a unified structured model:
 | autosnippet_guard_audit_files | Multi-file batch Guard audit |
 | autosnippet_scan_project | Lightweight project scan + Guard audit |
 | autosnippet_bootstrap_knowledge | Cold-start knowledge base initialization (9 dimensions) |
+| autosnippet_bootstrap_refine | AI refinement of bootstrap candidates (Phase 6) |
+
+### Skills Management
+
+| Tool | Description |
+|------|-------------|
+| autosnippet_list_skills | List all available Agent Skills |
+| autosnippet_load_skill | Load Skill document for domain guidance |
+| autosnippet_create_skill | Create project-level Skill |
+| autosnippet_update_skill | Update project-level Skill |
+| autosnippet_delete_skill | Delete project-level Skill |
+| autosnippet_suggest_skills | AI-powered Skill creation suggestions |
 
 ### Usage Telemetry
 

package/templates/constitution.yaml

@@ -48,7 +48,7 @@ roles:
       - "read:recipes"
       - "read:guard_rules"
       - "create:candidates"
-      - "submit:candidates"
+      - "submit:knowledge"
       - "read:audit_logs:self"
       - "knowledge:bootstrap"
       - "create:skills"

package/templates/copilot-instructions.md

@@ -23,7 +23,7 @@
 
 ## 强制规则（必须遵守）
 1. **禁止直接修改** 知识库目录内容（`AutoSnippet/recipes/`、`.autosnippet/` 等）。
-2. 创建或入库必须走 **Dashboard** 或 MCP 工具流程（`autosnippet_submit_candidate`、`autosnippet_submit_candidates`）。
+2. 创建或入库必须走 **Dashboard** 或 MCP 工具流程（`autosnippet_submit_knowledge`、`autosnippet_submit_knowledge_batch`）。
 3. **优先使用 Recipe** 作为项目标准；源代码仅作补充。
 4. MCP 检索优先：使用 `autosnippet_search` 或 `autosnippet_context_search` 获取语义检索结果。
 5. MCP 调用失败时，**不要在同一轮重复重试**，回退到已读文档或静态上下文。
@@ -47,9 +47,9 @@
 ### 候选提交 & 校验
 - `autosnippet_validate_candidate` — 提交前预校验（5 层）
 - `autosnippet_check_duplicate` — 相似度检测
-- `autosnippet_submit_candidate` — 单条提交（reasoning 必填）
-- `autosnippet_submit_candidates` — 批量提交（含去重 + 限流）
-- `autosnippet_submit_draft_recipes` — 解析草稿 Markdown 文件
+- `autosnippet_submit_knowledge` — 单条提交（reasoning 必填）
+- `autosnippet_submit_knowledge_batch` — 批量提交（含去重 + 限流）
+- `autosnippet_submit_knowledge_batch` — 解析草稿 Markdown 文件
 - `autosnippet_enrich_candidates` — AI 补全缺失语义字段
 
 ### 项目扫描 & 冷启动
@@ -85,8 +85,8 @@
 
 ## 推荐工作流
 - **查找**：`autosnippet_search`（推荐）或 `autosnippet_context_search`（上下文感知）。
-- **产出候选**：`autosnippet_validate_candidate` 预校验 → `autosnippet_submit_candidate` 提交。
-- **冷启动**：`autosnippet_bootstrap_knowledge` → `autosnippet_enrich_candidates` → `autosnippet_bootstrap_refine` → 逐 Target 深入 → `autosnippet_submit_candidates`。
+- **产出候选**：`autosnippet_validate_candidate` 预校验 → `autosnippet_submit_knowledge` 提交。
+- **冷启动**：`autosnippet_bootstrap_knowledge` → `autosnippet_enrich_candidates` → `autosnippet_bootstrap_refine` → 逐 Target 深入 → `autosnippet_submit_knowledge_batch`。
 - **Skills 创建**：`autosnippet_suggest_skills` 分析 → `autosnippet_create_skill` 固化知识。
 - **采纳反馈**：`autosnippet_confirm_usage`（记录使用量影响排序权重）。
 

package/templates/recipes-setup/README.md

@@ -160,9 +160,9 @@ V2 知识库中的 Recipe 按 `knowledgeType` 自动归入三类 kind，影响
 | `autosnippet_search` | 统合搜索（BM25 + 语义融合） |
 | `autosnippet_context_search` | 智能上下文检索 |
 | `autosnippet_validate_candidate` | 候选结构化预校验 |
-| `autosnippet_submit_candidate` | 单条候选提交（reasoning 必填） |
-| `autosnippet_submit_candidates` | 批量候选提交 |
-| `autosnippet_submit_draft_recipes` | 解析草稿 Markdown 为候选 |
+| `autosnippet_submit_knowledge` | 单条候选提交（reasoning 必填） |
+| `autosnippet_submit_knowledge_batch` | 批量候选提交 |
+| `autosnippet_submit_knowledge_batch` | 解析草稿 Markdown 为候选 |
 | `autosnippet_confirm_usage` | 确认 Recipe 被采纳/应用 |
 | `autosnippet_recipe_insights` | 获取 Recipe 质量洞察（只读） |