
capture
by evmts
Advanced AI Agents in React
SKILL.md
name: capture description: Capture conversation content to reviews/, issues/, TODO.md, or Prompt.md with auto-classification allowed-tools:
- Read
- Write
- Bash
- Grep user-invocable: true
/capture - Content Classification and Capture
Automatically classify and save conversation content to the appropriate location.
When to Trigger
Invoke when user wants to:
- Save a code review
- Create an issue/feature request
- Add TODO items
- Save content to Prompt.md
- Capture important conversation context for later
Keywords: "capture", "save this", "add to", "record", "log this"
Targets
| Target | Directory | Description |
|---|---|---|
| review | reviews/ | Code reviews of commits/changes |
| issue | issues/ | Planned future features/enhancements |
| todo | TODO.md | Immediate action items (appended) |
| prompt | Prompt.md | Context/prompts for future sessions |
Classification Algorithm
Priority Order
- Prompt (explicit): "put this in Prompt.md"
- Review: commit hash + (issue|bug|broken) + file:line refs
- Issue: "add feature", "implement", future tense, no commit refs
- TODO: "must do", "fix now", checkbox patterns, urgent language
Decision Flow
Content Analysis
│
├── Explicit "Prompt.md" mention? ──────────────→ PROMPT
│
├── Commit hash present?
│ └── Yes + negative language ──────────────→ REVIEW
│
├── Feature/implement keywords + no commits? ──→ ISSUE
│
├── Urgent/checkbox/imperative patterns? ──────→ TODO
│
└── Default (confidence < 0.6) ─────────────────→ ASK USER
Usage Patterns
Pattern 1: Auto-classify
# User provides content, let classification decide
bun scripts/capture.ts "Commit abc1234 has a bug in auth.ts:45"
# → Classified as REVIEW (commit hash + bug + file ref)
Pattern 2: Force type
# Override classification with explicit type
bun scripts/capture.ts --type issue "Add WebSocket support"
Pattern 3: From conversation context
When the user says "capture this" or "save the above", extract relevant content from the conversation and run:
bun scripts/capture.ts "extracted content here"
Pattern 4: With metadata
bun scripts/capture.ts --type review --commit abc1234 "Review content"
bun scripts/capture.ts --type todo --priority high "Fix auth bug"
bun scripts/capture.ts --type issue --title "WebSocket Support" "description..."
Template Formats
Review (reviews/YYYYMMDD_HHMMSS_<hash>.md)
# Code Review for Commit [hash]
**Date:** YYYY-MM-DD HH:MM:SS
**Commit Message:** [message]
---
### Summary
[extracted summary]
### Issues Found
[extracted issues]
### Suggested Improvements
[extracted suggestions]
Issue (issues/<kebab-case-title>.md)
# [Title]
<metadata>
<priority>critical|high|medium|low</priority>
<category>feature|bugfix|enhancement</category>
<status>draft</status>
</metadata>
---
## Executive Summary
[1-3 sentences]
## Problem Statement
[detailed problem]
## Proposed Solution
[solution design]
## Acceptance Criteria
- [ ] [criteria]
TODO (appended to TODO.md)
Items appended under appropriate priority section (High/Medium/Low).
Prompt (Prompt.md)
User formatting preserved. Create or append based on context.
CLI Options
--type <review|issue|todo|prompt> Force target type
--commit <hash> Associate with commit (for reviews)
--priority <high|medium|low> Set priority (for TODO)
--title <title> Override title (for issues)
--stdin Read content from stdin
--dry-run Preview without writing
Edge Cases
Low Confidence (< 60%)
When classification confidence is low:
- Report all scores with reasoning
- Ask user: "Which type? (review/issue/todo/prompt)"
- Use user's choice
File Conflicts
- Reviews/Issues: Add counter suffix (
-1,-2, etc.) - TODO.md: Always append (never conflict)
- Prompt.md: Append if exists, create if not
Missing Git Context
If not in git repo:
- Warn user
- Continue without commit context
- Use timestamp-based filenames for reviews
Example Session
User: Capture this review - commit f691852 has poor commit hygiene...
Claude: I'll capture this as a code review.
[Runs: bun scripts/capture.ts --type review --commit f691852 "poor commit hygiene..."]
✅ Captured to reviews/20260118_143000_f691852.md
Classification: review (confidence: 92%)
Reasoning:
- Commit hash detected: f691852
- Negative language: "poor"
Preview:
────────────────────────────────────────────────────
# Code Review for Commit f691852
**Date:** 2026-01-18 14:30:00
**Commit Message:** test commit
...
────────────────────────────────────────────────────
Implementation
Shared utilities: src/utils/capture.ts
Script: scripts/capture.ts
Functions:
classifyContent(ctx)- Pattern matching and confidence scoringgenerateReviewTemplate(content, meta)- Review markdowngenerateIssueTemplate(content, meta)- Issue markdowngenerateTodoItem(content, priority)- TODO append logicgeneratePromptMd(content)- Prompt.md contentwriteCapture(generated)- File I/O
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon