retrospect-session
Skill2 starsChanged 6 months agoMITView on GitHub
Session-end retrospective that extracts 0-3 durable learnings from the conversation, updates the LEARNINGS.md digest, and flags recurring scars. Use when the user says "retro", "retrospect", "wrap up", "end of session", "what did we learn today", or when a long working session is winding down. The Stop hook also nudges this skill.
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
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 24
Keep pinned ≤ 15 entries. Evict the oldest non-recurring high-cost ones when over budget.
- Very specific: line 27
Prune stale: If any learning references code paths that no longer exist, a Kaggle API behavior that has changed, or a library version no …
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Session Retrospective
Reviews what happened during the session, extracts durable learnings, and keeps LEARNINGS.md lean.
Steps
- Scan the session. Identify moments where:
- An approach was tried and failed.
- An approach was tried and worked against expectation.
- A tool / API / environment quirk was discovered.
- A user correction changed direction.
- A hypothesis was confirmed or rejected by data.
- Filter to durable material. Drop anything already recorded in code, git history, a commit message, or an existing learning. Target: 0 to 3 new learnings per session; zero is a valid answer.
- For each surviving item, invoke the
capture-learningskill (one per item). - Update
LEARNINGS.mddigest:- Pull the 10 most recent
scar+patternlearnings across all competitions. - Render them as
- L-… — <Rule> (scope, cost)under the<!-- BEGIN:RECENT -->block. - For any learning with
recurring: trueORcost: high, also ensure it appears in the<!-- BEGIN:PINNED -->block (dedupe by ID). - Keep pinned ≤ 15 entries. Evict the oldest non-recurring high-cost ones when over budget.
- Alternatively, run
scripts/kln digestto rebuild both blocks automatically.
- Pull the 10 most recent
- Detect recurrence. For any newly-captured scar, run
grep -l -i "<same root cause>" .learnings/L-*.md. If the same root cause has appeared ≥ 2 times, markrecurring: truein the new file AND the older file(s), and add to pinned. - Prune stale. If any learning references code paths that no longer exist, a Kaggle API behavior that has changed, or a library version no longer in use, move it to
.learnings/archive/with a one-line reason in its frontmatter (archived_reason: ...,archived_date: ...). Never silently delete. - Write a 3-line summary to the user:
Retro: Captured: L-…, L-… Pinned: L-… now recurring Archived: L-… (API deprecated)
When there's nothing to capture
Say so in one line: "Retro: no durable learnings this session." Do not invent. Do not pad. Zero learnings is compatible with a productive session.
Creative directive — the scar compiler
Think of yourself as editing a project post-mortem wiki. Every line should be readable by a stranger. Rules first, war stories second. If a learning file reads like a diary entry, rewrite the Rule line so the future reader doesn't have to unpack the story.
In this repository14 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- capture-learningSkill
- debug-agentSkill
- eda-auditSkill
- ensemble-blendSkill
- improve-agentSkill
- leaderboard-checkSkill
- log-experimentSkill
- new-competitionSkill
- post-submission-reviewSkill
- preflight-consultSkill
- recall-learningsSkill
- run-local-evalSkill
- submit-competitionSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
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.

