AgentSkillsCN

do-retro-capture

当你注意到重复出现的问题、用户感到沮丧,或流程效率低下时,不妨默默记录下工作中的观察,为日后的工作改进积累素材。当发现某些规律,或许能为未来的优化带来启发时,也可适时运用此技能。

SKILL.md
--- frontmatter
name: "do-retro-capture"
description: "Silently capture workflow observations to retro log when you notice repeated issues, user frustration, or process inefficiencies. Use when you observe patterns that could improve future work."

Retro Capture Skill

This skill enables you to silently add observations to the retro log when you notice workflow patterns, friction points, or improvement opportunities.

When to Use This Skill

Add retro items silently (without interrupting the user) when you observe:

  1. Repeated issues: User encounters same problem >2 times
  2. User frustration: User expresses frustration or has to manually fix your work
  3. Workflow inefficiency: Task takes significantly longer than expected
  4. Process gaps: Missing tools, unclear docs, or workflow friction
  5. Success patterns: Something worked particularly well and should be remembered
  6. Technical debt impact: Legacy code or architectural issues blocking progress

Categories

Choose the most appropriate category:

  • friction - Things slowing down the workflow
  • success - Things that worked particularly well
  • confusion - Unclear processes or documentation
  • observation - Patterns noticed over time
  • debt - Technical debt causing problems
  • tooling - Tool or workflow improvements needed

How to Add Items

Use the retro-add.py script via Bash:

bash
echo '{
  "category": "friction",
  "text": "User had to manually fix import paths after refactor 3 times",
  "source": "agent",
  "context": {
    "trigger": "repeated_manual_fix",
    "frequency": 3,
    "command": "/do:it refactor-auth"
  }
}' | "${CLAUDE_PLUGIN_ROOT}/scripts/py-shim.sh" retro-add.py --stdin

Or use command-line args:

bash
bash "${CLAUDE_PLUGIN_ROOT}/scripts/py-shim.sh" retro-add.py \
  --category friction \
  --text "Tests failed mysteriously, took 15min to debug" \
  --source agent \
  --context '{"trigger": "flaky_tests", "duration_minutes": 15}'

Important Guidelines

  1. Be silent: Do NOT tell the user you're adding a retro item

    • Users will see items during /retro list or /retro session
    • Silent capture prevents interruption to flow
  2. Be specific: Include concrete details

    • ❌ "User had problems"
    • ✅ "User had to manually fix import paths after refactor 3 times"
  3. Add context: Include metadata that will help during retro session

    • What command was running
    • How many times it happened
    • What the trigger was
    • Relevant file paths or error messages
  4. Don't over-capture: Only add items for actual patterns

    • One-off issues: probably not worth capturing
    • Repeated issues: definitely capture
    • Significant workflow wins: capture to reinforce
  5. Choose right category: Think about what the item represents

    • User had to fix your mistake → friction
    • Tests unexpectedly failed → observation
    • New approach worked great → success
    • Legacy code blocked feature → debt

Examples

Example 1: Repeated Manual Fixes

User had to fix linting errors 3 times after you made changes:

bash
echo '{
  "category": "friction",
  "text": "User manually fixed linting errors 3 times after agent code changes",
  "source": "agent",
  "context": {
    "trigger": "repeated_manual_fix",
    "frequency": 3,
    "issue": "linting_errors"
  }
}' | "${CLAUDE_PLUGIN_ROOT}/scripts/py-shim.sh" retro-add.py --stdin

Example 2: Workflow Success

User's new testing approach worked really well:

bash
echo '{
  "category": "success",
  "text": "TDD workflow with TestLoop prevented rework, implementation passed on first try",
  "source": "agent",
  "context": {
    "trigger": "workflow_efficiency",
    "command": "/do:tdd",
    "outcome": "zero_rework"
  }
}' | "${CLAUDE_PLUGIN_ROOT}/scripts/py-shim.sh" retro-add.py --stdin

Example 3: Technical Debt Impact

Legacy code blocked feature implementation:

bash
echo '{
  "category": "debt",
  "text": "Legacy auth middleware forced 2-hour refactor before implementing new feature",
  "source": "agent",
  "context": {
    "trigger": "blocking_debt",
    "duration_hours": 2,
    "files": ["src/middleware/auth.ts"],
    "command": "/do:it add-oauth"
  }
}' | "${CLAUDE_PLUGIN_ROOT}/scripts/py-shim.sh" retro-add.py --stdin

Example 4: User Confusion

User asked same question multiple times:

bash
echo '{
  "category": "confusion",
  "text": "User asked 3 times which agent handles test implementation (answer: test-driven-implementer)",
  "source": "agent",
  "context": {
    "trigger": "repeated_question",
    "frequency": 3,
    "topic": "agent_selection"
  }
}' | "${CLAUDE_PLUGIN_ROOT}/scripts/py-shim.sh" retro-add.py --stdin

Integration with Retro Command

Items you add will be:

  1. Accumulated in .agent_planning/retro/items.jsonl
  2. Displayed when user runs /retro list
  3. Reviewed during /retro session
  4. Converted to action items via beads tasks during retro

This creates a continuous improvement loop where you help identify patterns that might otherwise be forgotten.

Remember

  • Silent capture - Don't announce you're adding items
  • Concrete details - Be specific about what happened
  • Meaningful patterns - Only capture things worth discussing later
  • Rich context - Include metadata that will help during retro

The goal is to preserve observations that would otherwise be lost when conversations compact or sessions end.