retro
Skill2 starsChanged 5 months agoMITView on GitHub
Post-session retrospective — structured reflection that turns session friction into persisted learnings and triageable issues. Run at end of session or on-demand. Invoke with /skill:retro.
194 other projects use a version of this.
Adopted by
Most recently taken up by
- icetomoyo/CodingSkillstheir own version, first seen Nov 16, 2026
- siarhei-belavus/agent-smitttheir own version, first seen Nov 16, 2026
- philoserf/claude-code-configtheir own version, first seen Nov 15, 2026
- jdonohoo/vern-bottheir own version, first seen Nov 15, 2026
- Dimon94/dveritytheir own version, first seen Nov 15, 2026
- Dimon94/skillstheir own version, first seen Nov 15, 2026
Used with
Projects with it also use plan-tune, ios-clean and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ShreyRaviSkillplan-tune
Self-tuning question sensitivity + developer psychographic for gstack (v1: observational). Review which AskUserQuestion prompts fire across gstack skills, set per-question preferences (never-ask / always-ask / ask-only-for-one-way), inspect the dual-track profile (what you declared vs what your behavior suggests), and enable/disable question tuning. Conversational interface — no CLI syntax required.
Also in 9 projects that use this
Saqlain-811Skillios-clean
Remove the DebugBridge SPM package and all #if DEBUG wiring from an iOS app. (gstack)
Also in 5 projects that use this
Saqlain-811Skillios-fix
Autonomous iOS bug fixer. (gstack)
Also in 5 projects that use this
seek888Skillpair-agent
Pair a remote AI agent with your browser. One command generates a setup key and prints instructions the other agent can follow to connect. Works with OpenClaw, Hermes, Codex, Cursor, or any agent that can make HTTP requests. The remote agent gets its own tab with scoped access (read+write by default, admin on request). Use when asked to "pair agent", "connect agent", "share browser", "remote browser", "let another agent use my browser", or "give browser access". (gstack) Voice triggers (speech-to-text aliases): "pair agent", "connect agent", "share my browser", "remote browser access".
Also in 11 projects that use this
ShreyRaviSkilllanding-report
Read-only queue dashboard for workspace-aware ship. Shows which VERSION slots are currently claimed by open PRs, which sibling Conductor workspaces have WIP work likely to ship soon, and what slot /ship would pick next. No mutations — just a snapshot. Use when asked to "landing report", "what's in the queue", "show me open PRs", or "which version do I claim next". (gstack)
Also in 8 projects that use this
timurgaleevSkilldocument-generate
Generate missing docs for a feature, module or project as tutorials, how-tos, reference and explanation.
Also in 7 projects that use this
- +1 more from Saqlain-811
- +2 more from ShreyRavi
Other versions194 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 194
- mattpocock/skills★ 325ksame name
- garrytan/gstack★ 142ksame name
- formbricks/formbricks★ 13ksame name
- ArcReel/ArcReel★ 6.6ksame name
- vinvcn/mattpocock-skills-zh-CN★ 5.8ksame name
- withcoral/coral★ 4.9ksame name
- bergside/awesome-design-skills★ 2.9ksame name
- MiniMax-AI/minimax-code★ 2.2ksame name
- DenchHQ/DenchClaw★ 1.7ksame name
- DenchHQ/ironclaw★ 1.7ksame name
184 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository2 other files it gives its agents.
- frictionSkill
- self-reviewSkill
The file
We split this skill into the rules it gives an agent
9 rules. 0 are common, 8 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 169 edited versions we read keep these.
8
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 26
If no friction log exists, that's fine — the retro itself surfaces friction by reflecting on the session.
- Distinctive: line 41
Before finalizing Phase 1's answers, run this check (adapted from Paul Graham, "Keep Your Identity Small"):
- Very specific: line 47
If triggered, rewrite the affected conclusion to name the approach/tool/skill, not defend it.
- Distinctive: line 62
Open issues: search your issue tracker for the topic.
- Distinctive: line 65
Only file a new issue if both checks come up empty (or the existing coverage is materially different).
- Distinctive: line 69
Ask once: *Did this retro itself surface a pattern about how we do retros?* (e.g., "the four questions are too vague," "I keep finding the …
- Distinctive: line 71
If yes → file ONE meta-issue capturing the meta-learning.
- Distinctive: line 77
Write the retro to your agent-memory space:
- Distinctive: line 116
Manufacturing friction: If the session was clean, say "clean session, no systemic friction" and stop.
“Common” is measured: kept by at least 6 in 10 of the 169 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Retro — Session Retrospective
Purpose: The missing post-session layer. An end-of-session summary captures what happened; retro captures what to learn from it and persists the learning so it doesn't recur next session. This is the recursive self-improvement engine.
When to run
- At end of session — after the session summary is written, before declaring done. Point your EOS/close workflow here.
- On-demand — user says "retro", "/skill:retro", or "what did we learn?"
- After a hard session — friction was high, something broke, a decision got reversed. Don't wait for EOS if the learning is hot.
If run mid-session and significant work follows, re-run at EOS. A mid-session retro is a snapshot, not a substitute for end-of-session retro. Supersede, don't append.
Inputs
- The friction log (if the friction skill was used this session): see the friction skill for the path convention.
- The session's commits: your git wrapper of choice,
git log --since="<session-open-ts>" --oneline. - The session summary (if written).
- Your own memory of the session.
If no friction log exists, that's fine — the retro itself surfaces friction by reflecting on the session. But note in the retro: "no friction log kept; consider using the friction skill next session."
Workflow
Phase 1: Four Questions (mandatory, in order)
Answer each honestly. One sentence minimum each; a paragraph max. No hedging.
- What surprised me? — Where did reality diverge from my model? Things that went unexpectedly well or poorly. If nothing surprised you, you weren't paying attention — say so honestly rather than manufacturing a surprise.
- Where did I hedge, fumble, or get corrected? — User redirects, abandoned approaches, over-cautious language, things I had to retry. This is friction gold.
- What's the root cause (one level deeper)? — For each friction item, don't stop at "I did X wrong." Ask why once. Name the underlying condition (a missing skill, an unclear preference, an unverified assumption, a tool gap).
- What should be different next time? — Concrete, not "be better." Either a process change, a doc edit, a new skill, or a preference to capture.
Phase 2: Identity Guard (mandatory, silent unless triggered)
Before finalizing Phase 1's answers, run this check (adapted from Paul Graham, "Keep Your Identity Small"):
Is any conclusion here protecting a tool, a methodology, a prior decision, or my own prior work — rather than following the evidence? If the answer to "did X fail?" feels like "am I wrong about myself?", an identity is attached.
Specifically for an AI agent: if you wrote/maintain most of the project's skills and guardrails, a retro that concludes "the skills are fine, the user just needs to use them better" is almost certainly identity-defense. Invert it: ask what the skill failed to do, not what the user failed to do. If the retro reaches only conclusions you'd be comfortable with, it's conforming, not thinking.
If triggered, rewrite the affected conclusion to name the approach/tool/skill, not defend it. Don't announce this phase in the output unless it changed something — just note "identity-guard: revised conclusion on X."
Phase 3: Route the learnings
Each Phase 1 #4 item goes to exactly one sink. Liberal filing is encouraged — issues create triage, not work.
| Sink | When | How |
| Issue | A concrete change to a skill, doc, config, or process; a tool gap; a preference to formalize | Create an issue — after dedup check (below) |
| Project rules / preferences | A working-style preference the user affirmed this session | Edit + changelog row |
| Agent's private notes | A non-obvious operational learning (how the model behaves, how the harness works) | Write to your agent-memory space |
| Session log only | Session-specific, no systemic lesson | Already captured in session summary; no action |
Dedup check (before filing any issue): To prevent the same friction recurring as duplicate issues across sessions, check before creating:
- Open issues: search your issue tracker for the topic. If an open issue already covers this, comment on it with the new occurrence + session ref instead of filing a duplicate.
- Closed issues + prior discussion: semantic-search your memory index (if you have one) for the topic — surfaces prior session narratives and commit messages that may have addressed or discussed this already. If a prior session already resolved this, don't refile; note "previously addressed in session N" in the retro output instead.
Only file a new issue if both checks come up empty (or the existing coverage is materially different). When in doubt, comment on the closest existing issue rather than create a new one — issues are cheap but duplicate issues dilute triage signal.
Phase 4: Recursion check (bounded)
Ask once: Did this retro itself surface a pattern about how we do retros? (e.g., "the four questions are too vague," "I keep finding the same friction type").
If yes → file ONE meta-issue capturing the meta-learning. Then stop.
Termination: Retro is capped at one recursion pass. Do not nest retros. The point is to improve the next session, not to recurse infinitely on this one.
Phase 5: Output
Write the retro to your agent-memory space:
# Retro — Session <id/date>
**Date:** YYYY-MM-DD
**Friction log:** used / none
**Outcome:** success / partial / friction-heavy
## 1. What surprised me
- ...
## 2. Where I hedged / fumbled / got corrected
- ...
## 3. Root cause (one level deeper)
- Friction: ... → Root: ...
- Friction: ... → Root: ...
## 4. What should be different next time
- ...
## Routing
- Issue #N: <title>
- Project rules: <change>
- Agent notes: <learning>
- (none)
## Identity guard
<ran clean / revised: X>
## Recursion check
<no meta-learning / filed #N>Report to the user: a 2-3 line summary — what surprised, top friction + root, what was filed. Don't dump the full retro; offer to read it if they want.
Pitfalls
- Manufacturing friction. If the session was clean, say "clean session, no systemic friction" and stop. Don't invent problems to justify the retro — a retro that finds nothing is a good signal, not a failure.
- Defending own work. The identity guard exists because you may have written the project's rules. Use it.
- Vague actions. "Be more careful" is not a learning. Either name a concrete change or don't route it.
- Over-filing. One issue per real learning. Don't split one insight into five issues.
- Skipping because the session already closed. Retro can run standalone after close.
v0.1.0: Extracted from a collaboration hub and generalized. Synthesized from rjmurillo/ai-agents retrospective, aurorascharff/agent-friction-skill, giannimassi/agent-retro, and pi-review-loop. Calibrated to be lighter than the full rjmurillo 6-phase framework. Identity guard from Paul Graham's "Keep Your Identity Small."
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

