session-wrap-up

name	session-wrap-up
description	Wrap up a meaningful repo work session. Use when: folding durable learnings into repo-local skills, syncing AGENTS/README/handoff docs, or preparing the final close-out commit and push.

Session Wrap-Up

When to Use

• Closing a meaningful work session with new repo knowledge, guardrails, or file conventions
• A repeated workflow should become a repo-local skill
• AGENTS.md, README.md, or handoff.md drifted from the current repo state
• Preparing the final commit and push after operational or documentation changes

Canonical Files

• .agents/skills/
• AGENTS.md
• README.md
• handoff.md
• docs/INGESTION_OPERATING_MODEL.md
• .git/helpers/gh-app

Procedure

1. List the durable repo changes from the session: new workflow, guardrail, canonical path, or operator rule
2. If the learning belongs to an existing area, update that domain skill first; only add a new narrow skill under .agents/skills/ when the workflow is genuinely new
3. Sync .agents/skill-keeper-agent/AGENT.md when skill ownership, inventory rules, or the line-budget contract changed
4. Keep durable repo memory inside tracked repo artifacts such as .agents/skills/, AGENTS.md, docs/, and research/; keep handoff.md as a rolling snapshot only, not durable repo memory, and do not create off-repo memory files for repo workflow truth
5. If a workflow is approved but not yet implemented, write it into a decision doc and one short handoff next-step note instead of active repo-truth docs, a new skill, or an append-only issue diary
6. Sync AGENTS.md so working rules, key files, and the skills table match reality
7. Sync README.md for repo entry points, repo-local helpers, or operator-facing conventions that a fresh agent should see quickly
8. Sync handoff.md with current state, risks, next steps, and any repo-cleanliness notes future agents actually need; keep it as a rolling snapshot and short recent-closeout window, not a per-slice running log or second skill inventory
9. Keep docs/HANDOFF_2026-03-22.md as archive-only audit context; touch it only when correcting its archival framing, not to mirror current workflow truth
10. Update docs/INGESTION_OPERATING_MODEL.md when staging areas, legacy surfaces, or ingestion guardrails changed
11. If a repo-specific heartbeat automation was created for session work and the operator is closing or says it is too slow, delete or pause it before final handoff and record the current state in handoff.md
12. Audit git worktree list --porcelain, .git/worktrees/, and repo-linked ~/.codex/worktrees/ folders; remove only clean/stale linked worktrees and empty parent dirs
13. Audit the repo for stray temp files, forbidden wave-* / run-* / session-* artifacts in active swarm paths, and clearly disposable one-off untracked helper scripts; remove only safe junk and keep archives/history intact
14. Commit the wrap-up/docs changes separately from unrelated research or product edits when possible

Guardrails

• Do not create a new skill for one-off facts; only encode repeatable workflows
• Keep the skill narrow and under repo limits; use skill-standards when revising it
• Do not overwrite unrelated user changes while syncing docs
• Do not describe planned work as current repo behavior just because it is approved
• Do not delete ignored dependency or build caches just to make the folder look cleaner
• If nothing durable changed, update only the handoff or docs that materially drifted, and keep the handoff bounded instead of appending another dated closeout diary entry

Validation

• Re-check that AGENTS.md still matches the actual .agents/skills/ catalog, and do not turn handoff.md into a second skill inventory
• Re-check that every touched SKILL.md stays at or under 100 lines
• Re-check that handoff.md still reads as a rolling snapshot rather than an append-only issue-watch journal
• Re-check that no off-repo memory file or stray active-swarm temp artifact was introduced by the session
• Re-check git status --short --branch when the session goal includes a fully clean repo
• Before push, verify git log --format=fuller -1 matches the repo-local identity and use the repo-local auth flow