autosnippet 3.2.18 → 3.2.21

package/skills/autosnippet-recipes/SKILL.md

@@ -5,147 +5,65 @@ description: Provides this project's Recipe-based context to the agent. Recipes
 
 # AutoSnippet Recipe Context (Project Context)
 
-> Self-check and Fallback: MCP tools return unified JSON Envelope. Before heavy ops call `autosnippet_health`. On failure do not retry in same turn; use static context or narrow scope.
-
 This skill provides the agent with this project's context from AutoSnippet Recipes. Recipes are the project's standard knowledge base: code patterns, usage guides, and structured relations.
 
 ---
 
-## Core Rule: Agent Permission Boundary
-
-**Agent CANNOT directly produce or modify Recipes.** Agent's role is:
+## Knowledge Base Overview
 
-| Allowed | Forbidden |
-|---------|-----------|
-| Submit Recipe candidates (submit_candidate / submit_candidates / submit_draft_recipes) | Directly create Recipe |
-| Validate candidates (validate_candidate / check_duplicate) | Modify existing Recipe content |
-| Search/query Recipes (list_recipes / get_recipe / context_search / list_facts) | Publish/deprecate/delete Recipe |
-| Confirm usage (confirm_usage) - record adoption/application telemetry | Modify Recipe quality scores |
-| Enhance candidate info (add rationale/steps/codeChanges etc.) | Bypass candidate review to write to recipes/ |
+| Part | Location | Purpose |
+|------|----------|---------|
+| **Recipes** | `AutoSnippet/recipes/*.md` | Standard code patterns + usage guides; used for AI context, Guard, search |
+| **Snippets** | `AutoSnippet/snippets/*.json` | Code snippets synced to IDE via `asd install` |
+| **Candidates** | `AutoSnippet/.autosnippet/candidates.json` | AI-scanned candidates; review in Dashboard then approve |
+| **Context index** | `AutoSnippet/.autosnippet/context/` | Vector index built by `asd embed`; semantic search via `autosnippet_search(mode=context)` |
 
-Recipe creation, review, publish, update, deprecate, delete are **human-only via Dashboard or HTTP API**.
+**Recipe** = one `.md` file = one specific usage pattern or code snippet. **kind**: `rule` (Guard enforced) / `pattern` (best practice) / `fact` (structural knowledge). **Recipe over project code**: When both exist, prefer Recipe as curated standard.
 
 ---
 
-## V3 Recipe Model
+## Agent Permission Boundary
 
-Recipe is the core knowledge unit. V3 uses a unified structured model:
-
-- **kind**: rule (mandatory norm, Guard enforced) | pattern (best practice) | fact (structural knowledge)
-- **knowledgeType**: code-pattern | architecture | best-practice | naming | error-handling | performance | security | testing | api-usage | workflow | dependency | rule
-- **complexity**: beginner | intermediate | advanced
-- **scope**: universal | project-specific | target-specific
-- **content**: { pattern, rationale, steps[], codeChanges[], verification, markdown }
-- **relations**: { inherits[], implements[], calls[], dependsOn[], dataFlow[], conflicts[], extends[], related[] }
-- **constraints**: { boundaries[], preconditions[], sideEffects[], guards[] }
-- **status**: draft -> active -> deprecated
+| Allowed | Forbidden |
+|---------|-----------|
+| Submit candidates (`autosnippet_submit_knowledge` / `_batch`) | Directly create/modify Recipes |
+| Search/query (`autosnippet_search` / `autosnippet_knowledge`) | Publish/deprecate/delete |
+| Confirm usage (`confirm_usage`) | Write to `AutoSnippet/recipes/` |
 
 ---
 
-## Instructions for the agent
-
-1. **Project context**: Read `references/project-recipes-context.md` in this skill folder for **Recipe 轻量索引**（title/trigger/category/summary 表格）。如需 Recipe 全文，调用 MCP `autosnippet_knowledge(operation=get, id)` 或 `autosnippet_search(query)`。索引缺失时，可直接读 `AutoSnippet/recipes/` 目录。
-
-2. **Finding code on demand**: Look up matching Recipe by title/summary/usage guide, use its code as standard to suggest. Cite the Recipe title.
-
-3. **Recipe over code search**: When both Recipe and code search find matches, prefer Recipe as source of truth.
-
-4. **Search - three ways**:
-   - In-context: `references/project-recipes-context.md` 轻量索引按 title/trigger/summary 匹配
-   - Terminal: `asd search <keyword>` or `asd search --semantic <keyword>`
-   - MCP: `autosnippet_search` (mode=auto 或 mode=context) with query and optional limit
-
-5. **Browsing Recipes via MCP**:
-   - `autosnippet_knowledge(operation=list)` - list with kind/language/category/knowledgeType/status/complexity filters
-   - `autosnippet_knowledge(operation=get, id)` - get single Recipe by ID (full content/relations/constraints)
-   - `autosnippet_knowledge(operation=list, kind=fact)` - list kind=fact structural knowledge
-
-6. **Confirming usage**: Call `autosnippet_knowledge(operation=confirm_usage, id, usageType)` when user adopts a Recipe. Telemetry only.
-
-7. **Updating context**: After user changes Recipes, tell them to run `asd install:cursor-skill` to regenerate references.
-
----
+## How to Find Recipes
 
-## How to use this context
+1. **In-context index**: Read `references/project-recipes-context.md` in this skill folder
+2. **MCP browse**: `autosnippet_knowledge(operation=list)` with kind/language/category filters
+3. **MCP get**: `autosnippet_knowledge(operation=get, id)` for full content
+4. **MCP search**: `autosnippet_search(mode=auto)` for unified BM25+semantic search
+5. **Terminal**: `asd search <keyword>`
 
-1. For project standards/Guard/conventions: Use Recipe content as source of truth.
-2. Recipe priority: Prefer Recipe over codebase implementations. Cite Recipe code.
-3. For "how we do X here": Base answer on Recipe content.
-4. For drafting candidates: Follow autosnippet-candidates flow. Never write to `AutoSnippet/recipes/`.
-5. For Audit/as:audit: Suggestions should match Recipe content.
-6. Usage Guide depth: Include deps, steps, error handling, perf, security, pitfalls, related Recipes.
-7. Placeholders: Prefer Xcode placeholders (e.g. `<#URL#>`, `<#Token#>`).
+**Recipe over code search**: When both find matches, prefer Recipe as source of truth. Cite Recipe title.
 
 ---
 
-## MCP Tools Reference (20 个整合工具)
-
-### Query (Agent can freely use)
-
-| Tool | Description |
-|------|-------------|
-| `autosnippet_health` | Health check + KB stats |
-| `autosnippet_search` | Unified search (`mode`: auto/context/keyword/semantic) |
-| `autosnippet_knowledge` | Knowledge browse (`operation`: list/get/insights/confirm_usage) |
-| `autosnippet_graph` | Knowledge graph (`operation`: query/impact/path/stats) |
-| `autosnippet_structure` | Project structure (`operation`: targets/files/metadata) |
-| `autosnippet_capabilities` | Service capability discovery |
-
-### Candidate Submit (Agent core capability)
-
-| Tool | Description |
-|------|-------------|
-| `autosnippet_submit_knowledge` | Submit single candidate (**strict validation** — missing required fields rejected immediately, no fallback). Must provide ALL fields in one call: title, language, content(+rationale), kind, doClause, dontClause, whenClause, coreCode, category, trigger, description, headers, usageGuide, knowledgeType, reasoning |
-| `autosnippet_submit_knowledge_batch` | Batch submit candidates (per-item strict validation + dedup + rate-limit) |
-| `autosnippet_save_document` | Save development document (design doc, debug report, ADR) — title + markdown only |
-
-### Guard & Scan & Bootstrap
-
-| Tool | Description |
-|------|-------------|
-| `autosnippet_guard` | Code Guard check (`code` single / `files[]` batch — auto-routed) |
-| `autosnippet_bootstrap` | Cold-start Mission Briefing (no params — returns project analysis + dimension tasks) |
-| `autosnippet_dimension_complete` | Dimension analysis completion (dimensionId + analysisText required) |
-
-### Wiki
+## How to Use This Context
 
-| Tool | Description |
-|------|-------------|
-| `autosnippet_wiki_plan` | Plan Wiki doc generation (scan project → topic data packages) |
-| `autosnippet_wiki_finalize` | Finalize Wiki (meta.json + dedup + validation) |
-
-### Task Management
-
-| Tool | Description |
-|------|-------------|
-| `autosnippet_task` | Unified task & decision management (`operation`: prime/create/claim/close/fail/defer/progress/decompose/record_decision/revise_decision/unpin_decision/list_decisions) |
-
-### Skills Management
-
-| Tool | Description |
-|------|-------------|
-| `autosnippet_skill` | Skill management (`operation`: list/load/create/update/delete/suggest) |
+1. **Project standards/Guard**: Use Recipe content as source of truth
+2. **"How we do X here"**: Base answer on Recipe content
+3. **Suggesting code**: Cite Recipe's code snippet, not raw search results
+4. **Guard/Audit**: `// as:audit` or MCP `autosnippet_guard` — both use Recipes as standard
+5. **Confirm adoption**: `autosnippet_knowledge(operation=confirm_usage, id, usageType)` when user uses a Recipe
 
 ---
 
-## How Recipes are used in the project
+## Auto-Extracting Headers for New Candidates
 
-| Use | How |
-|-----|-----|
-| Audit | `// as:audit` in source; `asd watch` runs AI review against Recipes. Or MCP `autosnippet_guard` |
-| Search | `asd search keyword` or MCP `autosnippet_search` (mode=auto/context/keyword/semantic) |
-| AI Assistant | Dashboard RAG and AI chat use Recipes as context |
-| Xcode | Recipes linked to Snippets; synced to Xcode CodeSnippets |
-| Guard | kind=rule Recipes enforced by Guard checks during audit |
-| Insights | `autosnippet_knowledge(operation=insights, id)` for quality scores, usage stats, and relation summary |
-| Graph | `autosnippet_graph(operation=query/impact/path)` for relationship analysis |
+1. **From code** (Recommended): Extract all import statements from user's code
+2. **From existing Recipes**: Check index for matching modules, then `autosnippet_knowledge(operation=get, id)` for full content
+3. **Via semantic search**: `autosnippet_search(mode=context)` with query like "import ModuleName"
 
 ---
 
-## Auto-Extracting Headers for New Candidates
-
-1. From code (Recommended): Extract all import statements from user's code
-2. From existing Recipes: Check `references/project-recipes-context.md` index for matching modules, then call MCP `autosnippet_knowledge(operation=get, id)` for full content
-3. Via semantic search: Call `autosnippet_search(mode=context)` with query like "import ModuleName headers"
+## Related Skills
 
-Use Option 1 first, then verify with Option 2 for consistency.
+- **autosnippet-create**: Submit knowledge candidates (V3 fields, validation, lifecycle)
+- **autosnippet-guard**: Code compliance checking against Recipe standards
+- **autosnippet-structure**: Project structure and knowledge graph

package/skills/autosnippet-structure/SKILL.md

@@ -5,83 +5,51 @@ description: Discover project structure (targets, files, dependency graph) and b
 
 # AutoSnippet — Structure & Dependencies & Knowledge Graph
 
-> Self-check & Fallback: MCP 工具返回统一 JSON Envelope（{ success, errorCode?, message?, data?, meta }）。重操作前调用 `autosnippet_health`；失败时不在同一轮重试，转用静态上下文或缩小范围后再试。
-
 Use this skill when the user asks about **project structure**, **module targets**, **dependency graph**, or **knowledge graph relationships**.
 
-## When to use
-
-- "有哪些 Target？"
-- "某个 Target 包含哪些文件？"
-- "依赖关系/谁依赖谁？"
-- "SPM 结构/模块拓扑"
-- "这两个 Recipe 之间有什么关联？"
-- "知识图谱统计 / 关联密度"
-- "修改这个 Recipe 会影响哪些？"
-
 ---
 
-## MCP Tools — Project Structure
+## Project Structure Tools
 
 | Tool | Purpose | Key Input |
 |------|---------|-----------|
-| `autosnippet_structure(operation=targets)` | List all module Targets (with file count, language stats, inferred role) | `includeSummary` (default true) |
-| `autosnippet_structure(operation=files)` | Get source files for a Target | `targetName` (required), `includeContent`, `contentMaxLines`, `maxFiles` |
-| `autosnippet_structure(operation=metadata)` | Get Target metadata (dependencies, package info, inferred role, graph edges) | `targetName` (required) |
+| `autosnippet_structure(operation=targets)` | List all module Targets (file count, language stats, inferred role) | `includeSummary` |
+| `autosnippet_structure(operation=files)` | Get source files for a Target | `targetName` (required) |
+| `autosnippet_structure(operation=metadata)` | Target metadata (dependencies, package info, graph edges) | `targetName` (required) |
 
-### Workflow: Discover project structure
-1. `autosnippet_structure(operation=targets)` → see all Targets with roles
-2. `autosnippet_structure(operation=files, targetName)` → drill into specific Target
-3. `autosnippet_structure(operation=metadata, targetName)` → get dependencies and relations
+### Workflow
+1. `targets` → see all Targets with roles
+2. `files` → drill into specific Target
+3. `metadata` → get dependencies and relations
 
 ---
 
-## MCP Tools — Knowledge Graph
+## Knowledge Graph Tools
 
-The knowledge graph captures **relationships between Recipes** (dependencies, extensions, conflicts, etc.).
+Captures **relationships between Recipes** (dependencies, extensions, conflicts, etc.).
 
 | Tool | Purpose | Key Input |
 |------|---------|-----------|
-| `autosnippet_graph(operation=query)` | Get all relations for a Recipe node | `nodeId` (Recipe ID), `relation` (filter), `direction` (out/in/both) |
-| `autosnippet_graph(operation=impact)` | Impact analysis: what downstream depends on this Recipe | `nodeId`, `maxDepth` (default 3) |
-| `autosnippet_graph(operation=path)` | Find shortest path between two Recipes (BFS) | `fromId`, `toId`, `maxDepth` (1-10) |
-| `autosnippet_graph(operation=stats)` | Global graph statistics: edge count, relation type distribution | (none) |
-
-### Workflow: Explore knowledge graph
-1. `autosnippet_graph(operation=stats)` → overview: how many edges, relation types
-2. `autosnippet_graph(operation=query, nodeId)` → see all connections for a specific Recipe
-3. `autosnippet_graph(operation=impact, nodeId)` → before changing a Recipe, check downstream impact
-4. `autosnippet_graph(operation=path, fromId, toId)` → discover indirect relationships between Recipes
-
-### When to use graph tools
-- User asks "修改这个 Recipe 会影响什么？" → `autosnippet_graph(operation=impact)`
-- User asks "这两个模块有什么关联？" → `autosnippet_graph(operation=path)`
-- User wants overview of knowledge connections → `autosnippet_graph(operation=stats)`
-- User wants to see all dependencies/extends/conflicts for a Recipe → `autosnippet_graph(operation=query)`
+| `autosnippet_graph(operation=query)` | Get all relations for a Recipe node | `nodeId`, `relation`, `direction` |
+| `autosnippet_graph(operation=impact)` | Impact analysis: downstream dependencies | `nodeId`, `maxDepth` |
+| `autosnippet_graph(operation=path)` | Shortest path between two Recipes (BFS) | `fromId`, `toId` |
+| `autosnippet_graph(operation=stats)` | Global graph statistics | — |
 
----
-
-## Dependency Structure File
-
-The SPM dependency structure is stored in `AutoSnippet/AutoSnippet.spmmap.json`:
-- **`graph.packages`**: Package declarations with targets
-- **`graph.edges`**: "from depends on to" relationships
-- **Update**: Run `asd spm-map` from project root to refresh (supports SPM / Node / Go / JVM / Python / Dart / Rust)
+### When to use
+- "修改这个 Recipe 会影响什么？" → `impact`
+- "这两个模块有什么关联？" → `path`
+- "知识图谱统计" → `stats`
 
 ---
 
-## Notes
+## Dependency Structure File
 
-- Keep results concise: show target name, type, path, and key dependencies.
-- If a user requests code generation, switch to **autosnippet-recipes** or **autosnippet-candidates** after clarifying the structure.
-- Use Dashboard dep graph visualization for team presentations (requires `asd ui`).
+SPM dependency structure: `AutoSnippet/AutoSnippet.spmmap.json`
+- Run `asd spm-map` to refresh (supports SPM / Node / Go / JVM / Python / Dart / Rust)
 
 ---
 
 ## Related Skills
 
-- **autosnippet-recipes**: Recipe content used as project standards.
-- **autosnippet-candidates**: After understanding structure, generate candidates per Target.
-- **autosnippet-analysis**: Deep project analysis uses structure + graph tools.
-
-```
+- **autosnippet-recipes**: Recipe content used as project standards
+- **autosnippet-create**: After understanding structure, submit knowledge candidates

package/templates/cursor-rules/autosnippet-skills.mdc

@@ -1,45 +1,29 @@
 ---
-description: AutoSnippet Project Skills — 项目级知识增强文档索引与使用指引
+description: AutoSnippet Project Skills — knowledge enhancement documents for AI Agent
 alwaysApply: true
 ---
 
 # AutoSnippet Project Skills
 
-Project Skills 是针对当前项目自动生成或手动创建的 AI Agent 知识增强文档。
-它们存放在 `AutoSnippet/skills/` 目录下，跟随项目 Git 管理。
+Project Skills are AI knowledge documents in `AutoSnippet/skills/`, tracked in Git.
 
-## 如何发现与加载
+## Discovery & Loading
 
-1. **列出所有 Skills**：调用 `autosnippet_list_skills` 查看当前可用的 Skills（含内置 + 项目级）
-2. **加载完整内容**：调用 `autosnippet_load_skill(skillName)` 获取具体 Skill 的完整文档
-3. **按需加载**：不确定该用哪个 Skill 时，先加载 `autosnippet-intent` 做意图路由
-4. **创建 Skill**：调用 `autosnippet_create_skill(name, description, content)` 创建项目级 Skill
-5. **Skill 推荐**：调用 `autosnippet_suggest_skills` 获取基于使用模式的 Skill 创建建议
+- **List**: `autosnippet_skill({ operation: "list" })` — all available Skills (builtin + project)
+- **Load**: `autosnippet_skill({ operation: "load", name: "<skill>" })` — full Skill content
+- **Create**: `autosnippet_skill({ operation: "create", name, description, content })` — new project Skill
 
-## Bootstrap 自动生成的 Project Skills
+## Built-in Skills (5)
 
-冷启动（`autosnippet_bootstrap_knowledge`）Phase 5.5 会自动为 4 个宏观维度生成 Project Skills：
+| Skill | Purpose |
+|-------|---------|
+| `autosnippet-create` | Submit knowledge candidates (V3 fields, validation) |
+| `autosnippet-recipes` | Project standard context (Recipe search, retrieval) |
+| `autosnippet-guard` | Code compliance checking against Recipes |
+| `autosnippet-structure` | Project structure discovery & knowledge graph |
+| `autosnippet-devdocs` | Save development documents (ADRs, debug reports) |
 
-| Skill | 维度 | 说明 |
-|-------|------|------|
-| `project-code-standard` | 代码规范 | 命名约定、文件组织规范 |
-| `project-architecture` | 架构模式 | 分层架构、模块边界、依赖图 |
-| `project-profile` | 项目特征 | 技术栈、模块结构、代码指标 |
-| `project-agent-guidelines` | Agent 约束 | 强制规则、TODO/FIXME、WARNING |
+## Priority
 
-生成后的 Skill 文件位于 `AutoSnippet/skills/<name>/SKILL.md`，自动跟随 Git。
-调用 `autosnippet_load_skill` 可加载完整内容。
-
-## 推荐工作流
-
-冷启动完整流程：
-1. `autosnippet_bootstrap_knowledge` — 9 维度分析 + 自动生成 Project Skills
-2. `autosnippet_enrich_candidates` — 补全缺失语义字段
-3. `autosnippet_bootstrap_refine` — AI 润色 summary/insight/relations
-4. 逐 Target 深入分析 → `autosnippet_submit_candidates`
-
-## 使用优先级
-
-- **Project Skills > 内置 Skills**：同名时项目级覆盖内置
-- **Skill > Recipe**：宏观架构/规范知识优先查 Skill，微观代码模式查 Recipe
-- **Skills 是只读参考**：不要直接修改 SKILL.md，应通过 `autosnippet_create_skill` 工具创建或更新
+- **Project Skills > Built-in Skills** — same name → project overrides builtin
+- **Skills = read-only reference** — load via MCP tool, do not edit SKILL.md directly

package/templates/instructions/agent-static.md

@@ -0,0 +1,24 @@
+# {{projectName}} — Agent Instructions
+
+> Auto-generated by AutoSnippet.
+
+This project uses **AutoSnippet** for knowledge management.
+Access the knowledge base through MCP tools.
+
+## MCP Tools
+
+- `autosnippet_task` — Task & decision management (prime/create/claim/close/record_decision)
+- `autosnippet_search` — Search knowledge (mode: auto/context/keyword/semantic)
+- `autosnippet_knowledge` — Knowledge CRUD (operation: list/get/insights)
+- `autosnippet_submit_knowledge` — Submit knowledge candidate
+- `autosnippet_guard` — Code compliance check
+- `autosnippet_skill` — Skill management (list/load)
+- `autosnippet_structure` — Project structure discovery
+- `autosnippet_bootstrap` — Project cold-start & scan
+- `autosnippet_health` — Service health & KB stats
+
+## Constraints
+
+1. Do NOT modify knowledge base files directly (`AutoSnippet/recipes/`, `.autosnippet/`).
+2. Prefer Recipes as project standards; source code is supplementary.
+3. Create or update knowledge only through MCP tools.

package/templates/instructions/conventions.md

@@ -0,0 +1,42 @@
+## YOU Operate autosnippet_task — The User Doesn't
+
+Users speak naturally; you translate to task operations. Never tell users to call autosnippet_task.
+
+## Task Rules (MANDATORY)
+
+1. **Prime EVERY message** — `autosnippet_task({ operation: "prime" })` FIRST.
+2. **Create task for non-trivial work** — ≥2 files OR ≥10 lines → `create` → `claim` → code → `close`.
+3. **Decision persistence** — User agrees/disagrees → `record_decision` immediately.
+4. **Session end** — Close or defer ALL tasks. Zero in_progress on exit.
+5. **You are the operator** — Never tell users to call autosnippet_task.
+6. **Skip task for**: Quick questions, single-file trivial fixes (<10 lines), code explanation.
+
+| User Says | You Run |
+|---|---|
+| "fix bug" / "implement" | `create` → `claim` → code → `close` |
+| "continue" | resume in-progress → `close` |
+| "pause" / "abandon" | `defer` / `fail` |
+| "break down" | `decompose(id, subtasks)` |
+| "what's next" | `ready()` → present |
+| "agreed" | `record_decision(...)` |
+
+## Knowledge Rules
+
+- **Do NOT modify** `AutoSnippet/recipes/` or `.autosnippet/` directly.
+- **Prefer Recipe** as project standard; source code is supplementary.
+- **Search**: `autosnippet_search({ query: "..." })` — auto mode (BM25 + semantic).
+
+## Essential MCP Tools
+
+- `autosnippet_task` — Task & decision management (**call `prime` first on every message**)
+- `autosnippet_search` — Search knowledge (mode: auto/context/keyword/semantic)
+- `autosnippet_knowledge` — Browse recipes (operation: list/get/insights)
+- `autosnippet_submit_knowledge` — Submit knowledge candidate
+- `autosnippet_guard` — Code compliance check
+- `autosnippet_skill` — Load project skills (list/load)
+- `autosnippet_health` — Service health & KB stats
+
+## Context Pressure
+
+- `CONTEXT_PRESSURE:WARNING` → Summarize completed work, continue.
+- `CONTEXT_PRESSURE:CRITICAL` → Call `prime` immediately to restore context.

package/skills/autosnippet-analysis/SKILL.md

@@ -1,169 +0,0 @@
----
-name: autosnippet-analysis
-description: Deep project analysis — full scan + semantic field enrichment + gap fill before Recipe creation. Both Cursor Agent and internal AI share the same capability set.
----
-
-# AutoSnippet — Deep Project Analysis & Semantic Enrichment
-
-> Self-check & Fallback: MCP 工具返回统一 JSON Envelope（{ success, errorCode?, message?, data?, meta }）。重操作前调用 `autosnippet_health`；失败时不在同一轮重试，转用静态上下文或缩小范围后再试。
-
-## Core Principle
-
-**MCP 工具负责数据采集 + 静态检查；你（Agent）负责语义理解；Agent 直接填写完整字段后通过 `autosnippet_submit_knowledge` 提交。**
-
-**⚠️ 严格提交规则**：`autosnippet_submit_knowledge` 实施严格前置校验——缺少必要字段的提交将被**直接拒绝（不入库）**。必须在单次调用中一次性提供所有必填字段（title, language, content, kind, doClause, dontClause, whenClause, coreCode, category, trigger, description, headers, usageGuide, knowledgeType, reasoning, content.rationale）。不要分步提交或先提交再补全。
-
----
-
-## Capability 1: 全项目扫描 → 提交候选
-
-> **首次接入项目？** 使用 **autosnippet-coldstart** Skill — 完整的 9 维度冷启动流程（含架构分析、项目特征、Agent 注意事项），比本 Skill 的扫描流程更全面。
-> 本 Capability 适用于已有知识库后的**增量补充扫描**。
-
-### When to Use
-用户说："全项目扫描"、"分析我的项目"、"提取架构模式"、"生成知识库"。
-
-### Workflow
-
-#### Phase 1: Collect & Baseline
-1. 调用 `autosnippet_bootstrap`（无参数）获取 Mission Briefing（文件列表 + Guard 审计基线）
-2. 审视 Guard 违规 — 静态规则问题 (命名/API 使用/废弃 API)
-3. 确定高优先级文件 (核心模块/共享工具/Service 层)
-
-#### Phase 2: Semantic Deep-Read
-读取代码文件，逐维度分析：
-
-| 分析维度 | 寻找什么 | knowledgeType | 对应字段 |
-|---------|---------|---------------|---------|
-| **架构模式** | 分层架构、MVVM/MVC、依赖注入 | `architecture` | rationale: 为什么选这种架构 |
-| **代码模式** | 封装方式、公共 API 设计、工厂/单例 | `code-pattern` | steps: 如何采用该模式 |
-| **代码规范** | 命名规范、文件组织、注释风格 | `code-standard` | constraints.preconditions: 前置条件 |
-| **最佳实践** | 错误处理、并发、内存管理 | `best-practice` | rationale: 为什么这是最佳实践 |
-| **调用链** | 关键业务的调用路径 | `call-chain` | steps: 调用顺序 |
-| **数据流** | 状态管理、模块间数据传递 | `data-flow` | steps: 数据流向 |
-| **模块依赖** | 依赖边界、约束 | `module-dependency` | constraints: 边界约束 |
-| **Bug 修复** | 常见修复模式、defensive coding | `solution` | rationale: 为什么这样修 |
-
-#### Phase 3: Submit with Full Semantic Fields
-**每条候选必须填充以下语义字段**（不留空）：
-
-```json
-{
-  "title": "网络层统一错误处理模式",
-  "content": {
-    "pattern": "<实际代码模式>",
-    "rationale": "统一错误处理避免分散的 try-catch，便于日志收集和错误上报"
-  },
-  "language": "swift",
-  "kind": "pattern",
-  "doClause": "Use centralized error handling via APIClient.request() for all network calls",
-  "category": "Network",
-  "trigger": "@network-error-handling",
-  "description": "网络层统一错误处理——所有请求通过 APIClient.request() 封装，统一 catch + 日志上报",
-  "headers": ["import Combine", "import Foundation"],
-  "usageGuide": "### 何时使用\n发起网络请求时统一通过 APIClient\n### 如何使用\n1. 创建 NetworkError 枚举 2. 调用 APIClient.request()",
-  "knowledgeType": "code-pattern",
-  "complexity": "intermediate",
-  "scope": "project-specific",
-  "steps": [
-    { "title": "创建 Error 枚举", "description": "在 Network/ 下创建 NetworkError.swift", "code": "enum NetworkError: Error { ... }" },
-    { "title": "封装请求方法", "description": "所有请求统一通过 APIClient.request()", "code": "" }
-  ],
-  "constraints": {
-    "preconditions": ["iOS 15+", "需引入 Combine framework"],
-    "boundaries": ["不可用于 WebSocket 连接"],
-    "sideEffects": []
-  },
-  "tags": ["networking", "error-handling"],
-  "reasoning": {
-    "whyStandard": "项目中 15 个网络模块均采用此模式",
-    "sources": ["Sources/Network/APIClient.swift", "Sources/Network/ErrorHandler.swift"],
-    "confidence": 0.9
-  }
-}
-```
-
-Use `autosnippet_submit_knowledge_batch` for batch submission.
-
-#### Phase 4: Guard Deep Audit (Optional)
-对特定文件调用 `autosnippet_guard` (files[]) 做深度规范审计。
-
----
-
-## Capability 2: 语义字段补全（需要 AI 理解的字段）
-
-### When to Use
-- 候选已有 title/code/language 但缺少深度语义字段
-- 候选要晋升为 Recipe 前，需要补全
-- 用户说："补全候选字段"、"enrich"、"查漏补缺"
-
-### 需要语义理解的 6 个字段
-
-| 字段 | 为什么需要 AI | 示例 |
-|------|-------------|------|
-| **rationale** | 需理解设计意图，回答"为什么这样做" | "该模式使用 Builder 而非直接构造，因为参数超过 5 个且多数可选" |
-| **knowledgeType** | 需理解知识本质（规范/模式/事实） | `architecture` vs `code-pattern` vs `best-practice` |
-| **complexity** | 需评估使用难度 | `beginner`（一行调用）vs `advanced`（需理解泛型+协议组合） |
-| **scope** | 需判断适用范围 | `universal`（通用 Singleton）vs `project-specific`（依赖项目 Config） |
-| **steps** | 需拆解实施步骤 | `[{title: "创建协议", description: "...", code: "..."}]` |
-| **constraints.preconditions** | 需理解前置条件 | `["iOS 15+", "需先配置 Firebase", "依赖 NetworkModule"]` |
-
-### 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_knowledge` 调用中。
-
-**方式 B: 对已有候选调用 AI 补全**
-> 注：`autosnippet_enrich_candidates` 已移入 admin 层级，Agent 推荐使用方式 A 直接填写完整字段。
-
-**方式 C: 候选字段诊断**
-```
-autosnippet_enrich_candidates({ candidateIds: ["id1", "id2", ...] })
-```
-诊断候选缺失字段（不使用 AI），Agent 根据诊断结果自行补全。
-
----
-
-## Capability 3: 候选 → Recipe 查漏补缺
-
-### When to Use
-- 用户准备将候选发布为 Recipe
-- 用户说："准备发布"、"候选检查"、"Recipe 就绪检查"
-
-### Workflow
-
-1. 对目标候选做结构补齐（确保所有字段完整）
-2. 调用 `autosnippet_enrich_candidates` 诊断候选缺失字段，根据结果自行补全
-3. 提交时使用 `autosnippet_submit_knowledge`（内置自动校验 + 去重检查）
-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
-
-### Recipe 必备字段检查清单
-
-| 字段 | 重要性 | AI 可填? |
-|------|-------|---------|
-| title | ★★★ 必填 | ✅ |
-| content (markdown+pattern+rationale) | ★★★ 必填 | ✅ |
-| language | ★★★ 必填 | ✅ |
-| kind (rule/pattern/fact) | ★★★ 必填 | ✅ |
-| doClause | ★★★ 必填 | ✅ |
-| category | ★★★ 必填 | ✅ |
-| trigger | ★★★ 必填 | ✅ |
-| description | ★★★ 必填 | ✅ |
-| headers | ★★★ 必填 | ✅ |
-| usageGuide | ★★★ 必填 | ✅ |
-| knowledgeType | ★★★ 必填 | ✅ |
-| reasoning | ★★★ 必填 | Agent 必须自己填 |
-| complexity | ★★☆ 推荐 | ✅ |
-| scope | ★★☆ 推荐 | ✅ |
-| steps | ★★☆ 推荐 | ✅ |
-| constraints.preconditions | ★★☆ 推荐 | ✅ |
-
----
-
-## Related Skills
-
-- **autosnippet-candidates**: 完整候选字段模型 + 提交指南
-- **autosnippet-structure**: 项目结构发现 (targets / files / dependencies)
-- **autosnippet-guard**: Guard 触发机制 (`// as:audit`)
-- **autosnippet-recipes**: Recipe 内容与查询