AgentSkillsCN

managing-signs

管理经验教训的提取与签章应用工作流。当需要从会话转录中提取经验教训、审查拟议的签章,或将已批准的经验教训应用于 CLAUDE.md 文件时,应使用此技能。可通过“提取经验教训”、“审查签章”、“应用签章”、“管理签章”、“经验教训待办事项清单”等短语进行触发。

SKILL.md
--- frontmatter
name: managing-signs
description: Manages learning extraction and sign application workflows. This skill should be used when extracting learnings from session transcripts, reviewing proposed signs, or applying approved learnings to CLAUDE.md files. Triggers on "extract learnings", "review signs", "apply signs", "manage signs", "learning backlog".

Managing Signs

This skill provides workflows for the learning loop - extracting wisdom from session failures and applying them as permanent signs in CLAUDE.md files.

Capabilities

  • Extract learnings from session transcripts
  • Manage the learning backlog (add, list, review)
  • Apply approved signs to target CLAUDE.md files
  • Integrate with sprint workflows for automated extraction

Quick Start

See the plugin commands for available operations:

  • /m42-signs:add - Manually add a learning to the backlog
  • /m42-signs:list - List learnings in the backlog
  • /m42-signs:extract - Extract learnings from a session transcript
  • /m42-signs:review - Review and approve/reject learnings
  • /m42-signs:apply - Apply approved learnings to CLAUDE.md files

Sprint Workflow Integration

Integrate learning extraction into your sprint workflows for automated capture of development insights.

Transcript Location

During sprint execution, session transcripts are stored at:

code
.claude/sprints/<sprint-id>/transcripts/<phase>-<step>-<subphase>.jsonl

Template Variables

VariableAvailable InDescription
{{sprint.id}}All phasesSprint identifier (e.g., 2026-01-18_feature-name)
{{sprint.name}}All phasesHuman-readable sprint name
{{step.id}}For-each phasesStep identifier (e.g., step-1)
{{step.prompt}}For-each phasesStep description from SPRINT.yaml
{{phase.id}}All phasesCurrent phase identifier

Adding Learning Extraction to a Workflow

Add as a final phase in any sprint workflow:

yaml
phases:
  # ... your existing phases ...

  - id: learning-extraction
    prompt: |
      Extract learnings from this sprint's development sessions.

      Use /m42-signs:extract to process transcripts at:
      .claude/sprints/{{sprint.id}}/transcripts/

      Use --confidence-min medium to filter low-confidence patterns.

Using the Workflow Templates

Two templates are provided in assets/:

  1. learning-extraction-workflow.yaml - Standalone learning extraction workflow

    yaml
    # Reference as a workflow
    - id: learning
      workflow: learning-extraction-workflow
    
  2. sprint-with-learning.yaml - Complete sprint workflow with learning extraction

    yaml
    # Use as your sprint workflow
    workflow: sprint-with-learning
    

Example Configuration

Full sprint configuration with learning extraction:

yaml
# SPRINT.yaml
name: Feature Implementation Sprint
workflow: sprint-with-learning

steps:
  - id: step-1
    prompt: Implement user authentication

  - id: step-2
    prompt: Add authorization middleware

# Learning extraction runs automatically after all steps

Manual Extraction

To manually extract learnings from a specific transcript:

bash
/m42-signs:extract .claude/sprints/<sprint-id>/transcripts/<file>.jsonl --confidence-min medium

Post-Extraction Review

After extraction completes:

  1. Run /m42-signs:status to see backlog summary
  2. Run /m42-signs:review to approve/reject learnings
  3. Run /m42-signs:apply to apply approved signs to CLAUDE.md files