Agent Skill
2/7/2026retrospective
Generate a research retrospective document at the end of research sessions. Analyzes what went well/badly/ugly, reviews skill utilization, and generates actionable improvements. Use after completing research tasks, especially before final commits on feature branches.
T
tomlupo
0GitHub Stars
1Views
npx skills add tomlupo/ai-playground
SKILL.md
| Name | retrospective |
| Description | Generate a research retrospective document at the end of research sessions. Analyzes what went well/badly/ugly, reviews skill utilization, and generates actionable improvements. Use after completing research tasks, especially before final commits on feature branches. |
name: retrospective description: Generate a research retrospective document at the end of research sessions. Analyzes what went well/badly/ugly, reviews skill utilization, and generates actionable improvements. Use after completing research tasks, especially before final commits on feature branches.
Research Retrospective Skill
Generate structured retrospective documents to compound learnings from research sessions.
When to Use
- After completing a research task
- Before the final commit on a feature branch
- When prompted by the PostToolUse hook after commits
- When you want to document lessons learned from a session
Workflow
Phase 1: Context Gathering
Gather context about the current session:
- Identify the branch:
git rev-parse --abbrev-ref HEAD - Review recent commits:
git log --oneline -20 - Check changed files:
git diff --stat HEAD~5..HEAD(or since branch divergence) - Identify the research task: Infer from branch name, commit messages, and modified files
Phase 2: Self-Reflection
Answer these questions based on the session:
What Went Well?
- Which approaches worked on the first try?
- What tools/libraries were effective?
- What decisions saved time?
What Went Badly?
- What required multiple debugging iterations?
- What assumptions were wrong?
- What took longer than expected?
What Was Ugly?
- What code needs refactoring?
- What processes were inefficient?
- What hacks or workarounds were used?
Phase 3: Skills Analysis
Review skill utilization:
- List skills that were used and rate their effectiveness
- Identify skills that SHOULD have been used but weren't
- Note why skills weren't used (forgot, didn't know, seemed overkill)
Key skills to evaluate:
/context7- Was library documentation checked?/qrd- Was a structured specification created?/workflows:plan- Was planning done before coding?/workflows:review- Was code reviewed?Exploreagent - Was it used for codebase questions?data/samples/- Was cached data used?
Phase 4: Document Generation
Generate the retrospective document with this structure:
# [Task Name]: Process Retrospective
**Date:** YYYY-MM-DD
**Branch:** [branch-name]
**Task:** [Brief description]
**Outcome:** [Success/Partial/Failed - brief summary]
---
## Executive Summary
| Aspect | Rating | Notes |
|--------|--------|-------|
| Research Quality | /10 | [Brief note] |
| Code Quality | /10 | [Brief note] |
| Time Efficiency | /10 | [Brief note] |
| Skill Utilization | /10 | [Brief note] |
| Final Results | /10 | [Brief note] |
---
## What Went Well
### 1. [Category]
[Description with specific examples]
### 2. [Category]
[Description with specific examples]
---
## What Went Badly
### 1. [Issue]
[What happened, root cause, time wasted]
### 2. [Issue]
[What happened, root cause, time wasted]
---
## What Was Ugly
### 1. [Problem]
[Description and why it's ugly]
---
## Skills Analysis
### Skills Used
| Skill | Usage | Effectiveness |
|-------|-------|---------------|
| [Skill] | [What for] | [Rating + notes] |
### Skills That Should Have Been Used
| Skill | Why Not Used | What Was Missed |
|-------|--------------|-----------------|
| [Skill] | [Reason] | [Impact] |
---
## Code & Repo Metrics
| Metric | Value |
|--------|-------|
| Execution time (approx) | |
| Files changed | |
| Files added | |
| Files deleted | |
| LOC added | |
| LOC removed | |
| Test pass rate | |
| Lint issues | |
| Dependency changes | |
| Errors encountered | |
---
## Structural & Process Issues
Identify problems with:
- **Repo structure:** Misplaced files, unclear organization
- **Instructions:** Missing or unclear documentation
- **Automation:** Missing scripts or workflows
- **Redundant code:** Duplicated logic that should be shared
- **Missing abstractions:** Repeated patterns that should be modular
[List specific issues found]
---
## Recommended Skill Usage Next Run
Explicit checklist for next similar task:
- [ ] [Tool/script to prefer]
- [ ] [Cached data to use]
- [ ] [Shared module to leverage]
- [ ] [Workflow to follow]
- [ ] [Documentation to check]
---
## Actionable Improvements
### Immediate (Do Now)
1. [Specific action]
2. [Specific action]
### Short-Term (This Week)
1. [Specific action]
2. [Specific action]
### Long-Term (This Month)
1. [Specific action]
2. [Specific action]
---
## Lessons Learned
### For Future [Domain] Research
1. [Lesson]
2. [Lesson]
### For Claude Code Usage
1. [Lesson]
2. [Lesson]
---
## Summary
| Category | Key Takeaway |
|----------|--------------|
| **Research** | [Main insight] |
| **Code** | [Main insight] |
| **Skills** | [Main insight] |
| **Process** | [Main insight] |
---
## Final Self-Score
| Dimension | Score | Notes |
|-----------|-------|-------|
| Speed | /10 | |
| Code Quality | /10 | |
| Tool Usage | /10 | |
| Repo Alignment | /10 | |
| **Overall** | **/10** | |
**Justification (2-3 lines max):**
[Brief explanation of scores and main areas for improvement]
Phase 5: Save Document
Save to: output/{branch-name}/retrospective_{YYYYMMDD_HHMMSS}.md
# Get branch name (sanitized for filesystem)
branch=$(git rev-parse --abbrev-ref HEAD | tr '/' '-')
# Create output directory
mkdir -p "output/${branch}"
# Generate timestamp
timestamp=$(date +%Y%m%d_%H%M%S)
# Save file
# output/${branch}/retrospective_${timestamp}.md
Output Location
- Directory:
output/{branch-name}/ - Filename:
retrospective_{YYYYMMDD_HHMMSS}.md - Branch name sanitization: Replace
/with-
Behavioral Rules
When generating retrospectives:
- Be critical and honest - the goal is learning, not looking good
- Prioritize engineering improvements over process observations
- Prefer reuse over new code - identify what should have been reused
- Assume future runs should be faster and simpler
- Do not write motivational text or filler
- Do not repeat context already in the report
- Keep it technical and precise
- Quantify everything possible
Tips for Effective Retrospectives
- Be honest - The goal is learning, not looking good
- Be specific - "Debugging took long" is useless; "yfinance API change cost 30min" is actionable
- Quantify when possible - Time wasted, lines of code deleted, iterations required
- Focus on actionable items - Each improvement should be specific enough to implement
- Link to existing patterns - Reference
docs/memory/patterns.mdif applicable - Update memory files - If learnings are reusable, add to
docs/memory/
Integration with Other Skills
- After
/retrospective, consider running/workflows:compoundto formalize learnings - Update
.claude/rules/if new workflow patterns emerge - Add reusable code to
shared/if identified during retrospective
Skills Info
Original Name:retrospectiveAuthor:tomlupo
Download