AutoSnippet Batch Scan — Cursor as Batch Scanner

This skill describes when and how to use Cursor as a batch scanning tool for an SPM Target: get target list → get file list → extract candidates per file with Cursor → submit to Dashboard Candidates → human review. Skills express semantics (when to use, flow, order of steps); MCP provides capability (the tools autosnippet_get_targets, autosnippet_get_target_files, autosnippet_submit_candidates). Do not expose URLs or HTTP here; call MCP tools by name.

When to use this skill

•User wants to scan a whole Target (or multiple files) with Cursor’s model to extract Recipe-like candidates and send them to Dashboard for review.
•User asks for "用 Cursor 扫一下这个 Target" / "批量提取候选" / "像 asd ais 那样但用 Cursor 模型" — i.e. alternative or complement to asd ais [Target] (which uses project-configured AI).
•Prerequisites: asd ui running (Dashboard on localhost:3000), MCP configured. Same Candidates pool as asd ais; review and approve in Dashboard Candidates page.

Flow (semantics only; capability = MCP tools)

•
Get Target list
Call MCP autosnippet_get_targets (no args). User or agent picks a targetName (e.g. MyModule) from the result.
•
Get file list for that Target
Call MCP autosnippet_get_target_files with targetName. Result is the list of source files (path, name) for that Target.
•
Extract candidates per file
For each file: read file content (in editor or via Cursor), use Cursor's model to extract Recipe-shaped items. Each item should be a SINGLE independent usage pattern (one specific scenario, one focused code snippet). Extract fields: title, summary, trigger, language, code, usageGuide; optional summary_cn, usageGuide_cn, category, headers.

Important extraction rules:
- •✅ One Recipe per specific usage: If a file has 3 different usage patterns (e.g. init, load, refresh), extract 3 separate Recipe items.
- •❌ Don't create mega-Recipes: Avoid combining multiple patterns ("This module does A, B, C...") into one Recipe.
- •✅ Focus on reusable snippets: Each Recipe's code block should be a self-contained, copy-pasteable example for ONE scenario.
- •✅ Clear trigger per scenario: @ModuleInit, @ModuleLoad, @ModuleRefresh — not just @Module.
Collect into an items array.
•
Submit candidates
Call MCP autosnippet_submit_candidates with targetName, items, and optional source (e.g. cursor-scan), expiresInHours (e.g. 24). Candidates are appended to Dashboard for the given target.
•
Review
User opens Dashboard Candidates page to approve or ignore each candidate — same workflow as for asd ais results.

Relation to asd ais

	asd ais [Target]	Cursor batch scan (this skill)
Model	Project AI (.env)	Cursor’s current model
Trigger	CLI	MCP tools + Cursor Agent
Output	Same Candidates pool	Same Candidates pool

Both feed the same Candidates pool; review and save flow in Dashboard is identical.

Instructions for the agent

•Recommend only when it fits: e.g. user wants to scan a Target with Cursor, or wants to compare Cursor vs project AI for extraction.
•Do not hard-code URLs or HTTP: use MCP tools only. If MCP is unavailable, suggest running asd ui and configuring MCP, or using asd ais [Target] instead.
•One flow at a time: get_targets → get_target_files → extract (per file) → submit_candidates. Do not skip steps; each tool is provided by MCP.
•Output via Candidates: Batch scan outputs candidates via autosnippet_submit_candidates.
•Hard rule — Single scenario per Recipe: Each extracted item must represent one specific usage scenario. If multiple patterns are found, split into multiple items.

Relation to other skills

•autosnippet-when: Use this skill when the user’s intent is "batch scan a Target with Cursor"; when routes here.
•autosnippet-concepts: Candidates and knowledge base concepts; Skills = semantics, MCP = capability.
•autosnippet-create: Single Recipe submit (open create page, paste); batch scan is multi-file → multi-candidate → submit in one go.