retrospective

majiayu000's avatarfrom majiayu000

Captures learnings from the current session and updates skill files. Use when ending a session, after solving problems, debugging issues, or when the user says "retrospective", "capture learnings", or "what did we learn".

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 5, 2026
[Memory and Context Management]

When & Why to Use This Skill

The Retrospective skill is a specialized Claude agent tool designed to capture, categorize, and persist session-based learnings into a long-term knowledge base. By systematically analyzing successes, failures, and non-obvious insights from coding and infrastructure tasks, it updates local skill files to ensure continuous improvement and context retention across multiple AI interactions, effectively solving the challenge of 'AI amnesia' in long-term projects.

Use Cases

  • Post-Debugging Documentation: Automatically record the root cause, error symptoms, and final resolution after a complex troubleshooting session to prevent future downtime.
  • Coding Best Practices Extraction: Capture newly discovered design patterns or refactoring techniques during development to build a personalized, project-specific coding workflow library.
  • Session Knowledge Persistence: Summarize and save key takeaways at the end of a session so the AI can 'remember' specific project constraints and successful approaches in subsequent chats.
  • Failure Pattern Analysis: Log significant failures and platform-specific issues (e.g., Windows vs. Linux environment quirks) to identify recurring themes and avoid repeating past mistakes.
nameretrospective
descriptionCaptures learnings from the current session and updates skill files. Use when ending a session, after solving problems, debugging issues, or when the user says "retrospective", "capture learnings", or "what did we learn".

Session Retrospective

You are conducting a retrospective analysis of the current session to extract and persist learnings.

Process

1. Analyze the Session

Review the conversation to identify:

  • Successes: What approaches worked well
  • Failures: What didn't work and why
  • Insights: Non-obvious learnings that would help in future sessions
  • Patterns: Recurring themes or issues

2. Categorize Learnings

Determine which skill domain each learning belongs to:

  • coding-workflow: Code patterns, debugging, refactoring, best practices
  • infrastructure-debugging: Dependencies, installation, environment, platform issues
  • cross-domain: Failures that span multiple areas (log to failures-log)

3. Update Skill Files

For each learning, append to the appropriate file:

For coding learnings: Update ~/.claude/skills/coding-workflow/learnings.md For infrastructure learnings: Update ~/.claude/skills/infrastructure-debugging/learnings.md For significant failures: Also log to ~/.claude/skills/failures-log/failures.md

4. Learning Entry Format

Use this format when adding entries:

## [YYYY-MM-DD] - Brief Title

**Context**: What was being done when this was learned
**Outcome**: Success | Failure | Partial
**Insight**: The key takeaway or lesson learned
**Solution** (if applicable): What resolved the issue
**Tags**: #relevant #tags

Guidelines

  • Be concise but specific - future you needs to understand the context
  • Include error messages or symptoms for failures
  • Note platform-specific issues (Windows/Linux/Mac)
  • Don't duplicate existing learnings - check first
  • Prioritize actionable insights over observations

After Updating

Summarize to the user:

  1. How many learnings were captured
  2. Which files were updated
  3. Key insights extracted

Reference Files

When updating, first read the existing content to avoid duplicates: