AutoSnippet Recipe Candidates — Draft Folder + MCP Submit

⚠️ CRITICAL: This skill is for single file/module scanning (e.g., "扫描 BDSchemeDispatcher").

DO NOT confuse with batch-scan workflow:

• •❌ WRONG: Use autosnippet_submit_candidates for file/module scanning (that is for SPM Target batch scan)
• •✅ RIGHT: Create draft folder → one .md per Recipe (not one big file) → call autosnippet_submit_draft_recipes → delete the draft folder after submit

For SPM Target batch scanning (multiple targets/files), use autosnippet-batch-scan skill instead.

Recommended: Create a draft folder (e.g. .autosnippet-drafts), one .md per Recipe—do not use one big file. After submitting candidates, delete the draft folder (use deleteAfterSubmit: true or rm -rf .autosnippet-drafts). Recipe candidates can be intro-only (no code); those do not generate a Snippet after approval.

When to use this skill

• •User wants to scan specific code files/modules (e.g. "扫描 BDSchemeDispatcher", "分析这个类生成 Recipe 候选")
• •User asks for Recipe candidate generation from existing code
• •User wants to extract reusable code patterns for the knowledge base
• •Critical: User should NOT be asking to create files directly in AutoSnippet/recipes/

❌ Common Mistakes to AVOID

1. •

   DO NOT create files directly in AutoSnippet/recipes/

   • •AutoSnippet/recipes/ is for FINAL, reviewed Recipes only
   • •WRONG: Creating .md in AutoSnippet/recipes/ or subdirs
   • •RIGHT: Create draft folder at project root (e.g. .autosnippet-drafts/)
2. •

   DO NOT use one big file for many Recipes

   • •Prefer draft folder, one .md per Recipe. After submit, delete the draft folder.
   • •RIGHT: .autosnippet-drafts/pattern-1.md, pattern-2.md, … → call autosnippet_submit_draft_recipes → delete draft folder
3. •

   DO NOT use autosnippet_submit_candidates for file/module scanning

   • •autosnippet_submit_candidates is for batch-scan (SPM Target) with structured items
   • •RIGHT: Draft folder + Markdown files + autosnippet_submit_draft_recipes
4. •

   DO NOT touch AutoSnippet/ when generating candidates

   • •Only work with a draft folder outside AutoSnippet; never delete or modify existing Recipe files

✅ Correct Flow (4 Steps)

Step 1: Create Draft Folder (Outside AutoSnippet)

# Project root, NOT inside AutoSnippet/
mkdir -p .autosnippet-drafts

Prefer draft folder with multiple files (not one big file). After submit, delete the draft folder.

Suggested names: .autosnippet-drafts/, .recipe-drafts/, tmp-recipes/

Step 2: Generate Markdown Files (One per Recipe)

When scanning code (e.g. BDSchemeDispatcher):

1. •Analyze the code to identify distinct usage patterns (or pure intro docs).
2. •Create ONE .md file per pattern in the draft folder:
   code

   .autosnippet-drafts/
   ├── BDSchemeDispatcher-初始化与分发.md
   ├── BDSchemeDispatcher-插件注册.md
   └── BDSchemeDispatcher-错误处理.md
   

3. •Format: Full Recipe MD (frontmatter with title, trigger, then ## Snippet / Code Reference + code block + ## AI Context / Usage Guide). Intro-only docs may have no code block (frontmatter + Usage Guide only); those do not generate a Snippet after approval.
4. •One pattern per file: Each file = ONE specific usage scenario or one intro doc.

Step 3: Submit via MCP

Call MCP autosnippet_submit_draft_recipes with the list of file paths:

• •filePaths: e.g. [".autosnippet-drafts/pattern-1.md", ".autosnippet-drafts/pattern-2.md", ...]
• •targetName: e.g. BDSchemeDispatcher
• •source, expiresInHours optional; deleteAfterSubmit: true to delete submitted files, or delete the draft folder manually after success.

Step 4: Delete draft folder (required)

After submit, delete the draft folder. Use deleteAfterSubmit: true when calling autosnippet_submit_draft_recipes to remove submitted files, then rm -rf .autosnippet-drafts to remove the folder, or remove the folder manually after confirming submission succeeded.

Workflow Diagram

User Request: "扫描 BDSchemeDispatcher 生成候选"
    ↓
Create .autosnippet-drafts/ (outside AutoSnippet/)
    ↓
Analyze code → Generate one .md per pattern (not one big file)
    ↓
.autosnippet-drafts/
├── Pattern1.md
├── Pattern2.md
└── Pattern3.md
    ↓
Call MCP: autosnippet_submit_draft_recipes(filePaths, targetName, ...)
    ↓
Candidates appear in Dashboard → User reviews
    ↓
Approved → AutoSnippet/recipes/ (via Dashboard). Intro-only candidates do not generate Snippet. **Then delete draft folder.**

Key Principles

1. •Draft folder: Create draft folder, one .md per Recipe (not one big file). Delete draft folder after submit.
2. •One .md per pattern: Each file = one usage scenario or one intro doc; intro-only candidates do not generate Snippet.
3. •MCP: Use autosnippet_submit_draft_recipes only (not create_candidates_from_staging; that tool was removed).
4. •Review: Candidates go through Dashboard Candidates before becoming Recipes.

Related MCP Tools

• •autosnippet_submit_draft_recipes: Submit draft .md files as candidates (recommended: draft folder + multiple files)
• •autosnippet_submit_candidates: Structured items only (for batch-scan workflow)
• •autosnippet_get_targets: For SPM Target batch scanning (different skill)

Relation to Other Skills

• •autosnippet-batch-scan: For scanning entire SPM Targets; this skill is for specific files/modules
• •autosnippet-create: For single Recipe creation via UI; this is for bulk candidate generation
• •autosnippet-when: Routes to this skill when user wants code-to-candidate generation

Instructions for the Agent

CRITICAL WORKFLOW:

1. •

   STEP 1: Create draft folder

   bash

   mkdir -p .autosnippet-drafts
   

   • •Project root, outside AutoSnippet/. Prefer one .md per Recipe; do not use one big file.
2. •

   STEP 2: Generate Markdown files

   • •Analyze the code (or doc need) and identify patterns
   • •Create ONE .md file per pattern in the draft folder (e.g. .autosnippet-drafts/ModuleName-Pattern.md)
   • •Full Recipe format: frontmatter + ## Snippet / Code Reference + ## AI Context / Usage Guide. Intro-only docs (no code block) do not generate Snippet after approval.
3. •

   STEP 3: Call MCP

   • •autosnippet_submit_draft_recipes with filePaths, targetName, optional source, expiresInHours, deleteAfterSubmit: true (or delete folder manually after).
4. •

   STEP 4: Delete draft folder

   • •Always delete the draft folder after submit. Use deleteAfterSubmit: true to remove submitted files, then rm -rf .autosnippet-drafts (or equivalent), or remove the folder manually once submission succeeded.

NEVER:

• •❌ Create files in AutoSnippet/recipes/
• •❌ Use autosnippet_submit_candidates for file/module scanning (use for batch-scan only)
• •❌ Use one big file for many Recipes (prefer one file per Recipe)
• •❌ Leave the draft folder in place after submit—always delete it

If MCP unavailable: Tell user to run asd ui first.

Example User Interactions

Good Request:

"扫描 BDSchemeDispatcher，生成 Recipe 候选"

Agent Response:

1. •Creates .autosnippet-drafts/ (draft folder first)
2. •Analyzes code, generates one .md per pattern (not one big file)
3. •Calls MCP autosnippet_submit_draft_recipes with the list of file paths (use deleteAfterSubmit: true or delete folder after)
4. •Deletes the draft folder after submit
5. •Tells user to review in Dashboard Candidates. Intro-only candidates do not generate Snippet.

Bad Request (redirect):

"Create BDSchemeDispatcher doc in AutoSnippet/recipes/"

Agent Response: "AutoSnippet/recipes/ is for final, reviewed content. Use the candidate flow: create a draft folder, generate one .md per scenario, call autosnippet_submit_draft_recipes, then delete the draft folder after submit."