alexleekt Session Retrospective
Review a completed session. Assess skill invocation, adherence, and propose improvements.
⚡ Quick Start
User says "review this session" or work feels complete? → Run the retrospective. It takes 30 seconds and produces actionable insight.
Process:
- Gather session evidence (git diff, commits, context-mode stats, memex)
- Score against each my-* skill
- Flag missed invocations and adherence failures
- Recommend skill updates
Activation Condition
This skill activates at session end or when the user explicitly requests a review.
Apply these rules when:
- The user says "review this session", "how did we do", or similar
- A significant session is ending (10+ turns, file mutations, multi-phase work)
- The user asks to evaluate skill effectiveness or automatic invocation
Scope
This skill covers:
- Skill invocation audit (what was used, what was missed)
- Workflow adherence check (did we follow my-workflow, my-vcs-hygiene, etc.)
- Automatic invocation assessment (did skills trigger when they should have?)
- Improvement recommendations (description updates, new sections, new skills)
Retrospective Process
1. Gather Evidence
# Git evidence
git log --oneline -10
git diff --stat HEAD~1 # or since session start
git diff HEAD~1 --name-only
# Context-mode evidence (if available)
ctx stats
# Session memory
cd <project> && grep -E "tool|skill|invoke" ~/.pi/session.log 2>/dev/null || echo "no session log"
2. Score Against my-* Skills
For each applicable skill, answer:
| Skill | Should Have Triggered? | Did Trigger? | Adherence | Notes |
|---|---|---|---|---|
| my-workflow | Direction assessment at start? | Yes/No | Good/Poor | |
| my-vcs-hygiene | Commits made? VCS commands given? | Yes/No | Good/Poor | |
| my-project-lifecycle | Plan/review/document phases? | Yes/No | Good/Poor | |
| my-code-review | Code reviewed before ship? | Yes/No | Good/Poor | |
| my-tech-stack | Tool decisions made? | Yes/No | Good/Poor | |
| my-context-strategy | Large outputs processed properly? | Yes/No | Good/Poor | |
| my-error-recovery | Errors handled gracefully? | Yes/No | Good/Poor | |
| my-communication-style | User feedback on verbosity? | Yes/No | Good/Poor | |
| my-decision-log | Decisions tracked? | Yes/No | Good/Poor | |
| my-agent-file-conventions | Agent files created? | Yes/No | Good/Poor |
3. Identify Patterns
Missed invocations — skills that should have triggered but didn't:
- Why? Description too narrow? Trigger phrases missing?
- Fix: Expand description, add activation examples
Adherence failures — skill triggered but rules not followed:
- Example: my-vcs-hygiene triggered but branch-from-main guard skipped
- Fix: Strengthen the rule, add to Common Mistakes
Over-invocation — skill triggered when it shouldn't:
- Why? Description too broad? Ambiguous trigger?
- Fix: Tighten description, add "DO NOT use for" cases
4. Recommend Improvements
Produce a structured report:
# Session Retrospective: <date>
## Session Summary
- Turns: ~N
- Files changed: N
- Commits: N
- Skills invoked: list
## Skill Invocation Audit
| Skill | Triggered | Should Have | Score |
|-------|-----------|-------------|-------|
| ... | ... | ... | ✅/❌ |
## Adherence Failures
1. **<Skill>** — <What went wrong>
- **Impact**: <Why it matters>
- **Fix**: <Specific change to skill>
## Improvement Recommendations
1. **<Skill>** — <What to change>
- **Rationale**: <Why this helps>
- **Priority**: High/Medium/Low
## Skill Update Candidates
- <skill>: <specific change>
Continuous Improvement Loop
Session completed
├── Retrospective runs (this skill)
│ ├── Produces adherence report
│ └── Recommends skill updates
└── User or agent applies updates
├── Edit skill description (better triggers)
├── Add new sections (new patterns observed)
├── Create new skill (new domain emerges)
└── Commit via @skills/my-vcs-hygiene
Goal: Each retrospective makes the my-* suite more accurate and more automatic.
Examples
Good: Retrospective after feature build
Session: Built async status widget for pi-event-horizon-provider.
Retrospective finds:
- ✅ my-project-lifecycle triggered — Plan, Build, Review, Document phases all present
- ✅ my-vcs-hygiene triggered — branch-from-main guard applied correctly
- ❌ my-context-strategy missed — processed 720-line diff with
git diffinstead ofctx_execute - ❌ my-error-recovery missed — no graceful handling for fetch timeout
Recommendations:
- my-context-strategy: Add trigger for "large diff" to ctx_execute instead of read
- my-error-recovery: Add section on HTTP timeout retry patterns
Good: Retrospective reveals under-triggering
Session: User gave visual feedback 5 times. No skill triggered for UI iteration.
Retrospective finds:
- ❌ my-vcs-hygiene under-triggered — description didn't include "move this left"
Fix: Expand description to include visual/UX feedback phrases.
Common Mistakes
❌ Wrong: Running retrospective without checking git evidence ✅ Right: Always check git log and diff — the ground truth of what happened
❌ Wrong: Retrospective that only lists what went well ✅ Right: Focus on misses and adherence failures — that's where improvement lives
❌ Wrong: Recommending skill updates without specific examples from the session ✅ Right: Every recommendation must cite a concrete session event
Related Skills
- @skills/my-workflow — What should have happened during the session
- @skills/my-vcs-hygiene — Commit discipline that was assessed
- @skills/my-project-lifecycle — Lifecycle phases that were assessed
- @skills/skill-creator — For implementing skill updates recommended by retrospective
Versioning
- Last updated: 2026-05-26
- Version: 1.0