wrap
Skill2 starsChanged 6 days agoMITView on GitHub
Session-terminus driver — one /wrap runs the full closing ceremony by driving the existing organs, from coordination teardown through changelog and build progress. Auto-runs only the reversible spine; every canonical write gated accept/edit/reject. Use on /wrap or \"wrap up the session\". Mode A — never auto-fires.
45 other projects use a version of this.
Adopted by
Most recently taken up by
- rhasselle-oddball/claude-skillstheir own version, first seen Nov 8, 2026
- jolleyDesign/claude-skillstheir own version, first seen Nov 8, 2026
- fjwood69/moritheir own version, first seen Nov 8, 2026
- egregore-labs/egregoretheir own version, first seen Nov 6, 2026
- mag123c/toktracktheir own version, first seen Nov 3, 2026
- theyoungastronauts/polaristheir own version, first seen Nov 1, 2026
Used with
Projects with it also use start, update and 7 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.
AnthropicSkillstart
Set up your bio-research environment and explore available tools. Use when first getting oriented with the plugin, checking which literature, drug-discovery, or visualization MCP servers are connected, or surveying available analysis skills before starting a new project.
Also in 5 projects that use this
AnthropicSkillupdate
Sync tasks and refresh memory from your current activity. Use when pulling new assignments from your project tracker into TASKS.md, triaging stale or overdue tasks, filling memory gaps for unknown people or projects, or running a comprehensive scan to catch todos buried in chat and email.
Also in 3 projects that use this
LangChainSkillremember
Review the current conversation and capture valuable knowledge — best practices, coding conventions, architecture decisions, workflows, and user feedback — into persistent memory (AGENTS.md) or reusable skills. Use when the user says: (1) remember this, (2) save what we learned, (3) update memory, (4) capture learnings.
Also in 3 projects that use this
termdockSkillrecall
Retrieves cross-session memories about past decisions, lessons, and patterns. Use when user asks about prior work, past decisions, or mentions a feature. Triggers include check memory, what did we decide, how did we solve.
Also in 3 projects that use this
Microsoft AzureSkillbuild
Build the TypeSpec Rust emitter from source. Run this after cloning the repo, after pulling changes, or when TypeScript source files have been modified.
Also in 3 projects that use this
Microsoft AzureSkillreflect
Learning capture system that extracts HIGH/MED/LOW confidence patterns from conversations to prevent repeating mistakes. Use after user corrections ("no", "wrong"), praise ("perfect", "exactly"), or when discovering edge cases. Complements .squad/agents/{agent}/history.md and .squad/decisions.md.
Also in 3 projects that use this
- +1 more from anthropics
Other versions45 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 45
- UniClipboard/UniClipboard★ 2ksame name
- team-attention/stanford-cs146s-kr★ 294same name
- egregore-labs/egregore★ 293same name
- mag123c/toktrack★ 196same name
- neurofoo/agent-skills★ 111same name
- lsnackerman/raw-clauding-skills★ 45same name
- fjwood69/mori★ 32same name
- YahyaLendani/opus-tuner-claude-skill★ 14same name
- jackjin1997/ClawForge★ 12same name
- ucsandman/claude-skills★ 5same name
35 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 repository47 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .github/copilot-instructions.mdCopilot instructions
- autonomizeSkill
- blueprintSkill
- brainstormSkill
- captureSkill
- checkpointSkill
- close-outSkill
- coordinateSkill
- cycleSkill
- deliverSkill
- design-proposalSkill
- discoverSkill
- dispatchSkill
- draft-planSkill
- envisionSkill
- executeSkill
- extract-conventionsSkill
- glossarySkill
- humanizeSkill
- human-viewSkill
- improve-promptSkill
- memory-auditSkill
- new-agentSkill
- new-agents-mdSkill
- new-hookSkill
- new-mcpSkill
- orchestrateSkill
- pr-authorSkill
- prepSkill
- pre-pushSkill
- pr-triageSkill
- recapSkill
- retrospect-coreSkill
- reviewSkill
- rulesSkill
- shortcutsSkill
- specSkill
- systematic-debuggingSkill
- test-driven-developmentSkill
- update-changelogSkill
- update-codemapSkill
- using-git-worktreesSkill
- visualise-codemapSkill
- visualSkill
- work-stateSkill
- write-testsSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 4 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 40 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 42
Run the ceremony in this order.
- Distinctive: line 56
Read + reconcile (auto): Read this session's scratch and apply contradiction reconciliation — a later observation superseding an earlier …
- Very specific: line 71
If the session shipped a slug (a .claude/specs/<slug> + .claude/plans/<slug> whose done-criteria are met), invoke close-out <slug> — …
- Very specific: line 94
Not silent-write: Every canonical write (recap file, capture, scoreboard, spec reconcile, artefact move, changelog line, BACKLOG-row …
- Distinctive: line 97
Not mid-session supervision: No background watcher; the auto-fire judgment runs once, at the next SessionStart.
- Distinctive: line 98
Not a double-offerer: When close-out runs (step 4), the changelog (step 5) and its re-offer are suppressed — close-out contains the …
“Common” is measured: kept by at least 6 in 10 of the 40 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
wrap
The session-terminus driver. The closing organs — recap, close-out, update-changelog, capture, plus coordinate teardown — each exist and are individually good, but nothing sequences them and nothing fires them on its own, so the closing ceremony is several manual invocations the user must remember in the right order with the right dedup. /wrap is that sequencer.
It is a driver, not a new organ: it calls the existing organs in order and adds only the missing connective tissue (the sequence, the dedup, the autonomy-posture read, the BACKLOG-shipped-row sweep, the ship-narrative-record offer). If it reimplements recap/close-out/changelog logic, it is wrong.
Governed by the autonomy write-class line: automate the reversible read/draft/detect spine, keep the accept/edit/reject gate on every canonical write. Same pattern as capture's auto-propose-never-write and prep's auto-fire-read-only.
Where it sits. The forward chain (envision → spec → draft-plan → blueprint → execute) is complete; /close-out closes a shipped slug; /update-changelog is the one changelog writer; recap records the session. /wrap sits above all of them as the session-close orchestrator — it decides which of them the session earned and runs them in one pass. checkpoint --scope corpus audits coherence; /wrap closes the session.
When to run
- User types
/wrap(explicit trigger). - User says "wrap up the session", "close this out", "run the closing steps", "end-of-session".
- The optional auto-fire (opt-in,
enabled:falseby default): aSessionStart[clear|compact]recovery hook (SessionEnd/PreCompactcannot surface — their stdout is debug-only) reads the prior session's transcript at the next start and, on an in-context judgment that the prior session shipped or closed work whose terminus was never run, proposes/wrap. It never auto-runs the ceremony — it surfaces the proposal, the user invokes. - Otherwise never auto-fires. Mode A. Every canonical write inside the ceremony passes its own accept/edit/reject gate.
The interrupt posture — read the autonomy contract once
At entry, resolve the effective autonomy contract once (per autonomize: live > kickoff > inferred > default > implicit-default) and read its ask field. Pass the resolved posture down to the sub-steps — the sub-organs do not each re-read it, so the whole ceremony runs at one consistent posture:
forks-only(the default) — run the reversible spine (coordinate teardown, gather/draft/detect) without per-step confirmation; stop only at the canonical-write gates and real forks.every-step— confirm each consequential step; the whole ceremony reverts to per-step propose-confirm (the posture for a user who wants it tightly supervised).until-blocked— run the reversible spine and surface the canonical-write proposals without pausing between steps, stopping only when genuinely blocked.
The contract sets the interrupt cadence; it never converts a canonical write into an auto-write (§ the gate below). If no contract resolves, fall back to forks-only.
The gate — what auto-runs vs what stops
Per the autonomy write-class line:
- Routine-reversible → auto (no per-step confirm under
forks-only): the coordinatedoneteardown (ephemeral board), reading the session's git/state, drafting the recap body, detecting a shipped slug, detecting shipped BACKLOG rows, reading the autonomy contract, the scratch read/reconcile/dedup pass and its post-promotion clear (session-scoped, disposable, already TTL-bound). - Canonical-write → gated (accept/edit/reject, always, regardless of contract —
every-steponly adds confirmation, never removes it): the recap file write, the promotion→capture, the scratch promotion→capture, the MEMORY.md scoreboard write, the spec reconcile, the artefact move, the changelog line, the BACKLOG-row removal, the ship-narrative-record edit.
Procedure
Run the ceremony in this order. Each step is conditional — skip it when it doesn't apply, silently.
1. Coordinate teardown (routine-reversible — auto)
Invoke coordinate done to tear down this session's board row (.claude/sessions-active.md, gitignored/ephemeral). This is a routine-reversible board write — under forks-only it runs without a confirm. Skip silently if no board / no row for this session. (Composes with coordinate's auto mode, which auto-registers the row at session start.)
2. Checkpoint (conditional; analysis auto, findings gated)
If the session built a module- or corpus-scope change (not a diff-scope tweak), offer checkpoint — "does what I built still fit + what did I learn?" Reuse checkpoint's own auto-scope detection to decide whether it's warranted; skip for a diff-scope tweak. Its analysis is read-only (auto); any findings route through checkpoint's own gate. Do not run a reconcile pass here that step 4's close-out will re-run on the same slug — checkpoint detects, close-out acts; running both reconciles on one slug is duplicate work.
2.5 Scratch reflection pass (read/dedup auto; promotion gated)
The session's scratch tier (written silently during work by execute (Record the outcome to scratch) and checkpoint step 7) holds in-flight observations that are deleted un-promoted at session end. This is the one pass that can turn any of them into durable memory. It runs before recap so recap narrates a session whose promotions are already settled.
- Read + reconcile (auto). Read this session's scratch and apply contradiction reconciliation — a later observation superseding an earlier one wins, so a corrected reading is never offered beside the stale one it replaced.
- Filter by salience (auto). Drop
low-salience entries from candidacy. Those are the passing-verification notes: real value during the session for re-orientation, near-zero value as durable memory. They are filtered from the candidate list, not deleted early. - Dedup + merge (auto). Collapse near-duplicates into one candidate. Several steps hitting the same seam should promote as one memory, not four.
- Present as ONE batch (gated). Show the survivors as a single accept/edit/reject list — "promote 1,3 / all / none / show <id>" — the same select-don't-loop posture as checkpoint's findings flow. Not a per-candidate confirm sequence: per-fact prompting is the cost this tier exists to remove.
- Promote via capture (gated). Accepted candidates go to
capture, which owns durable authoring — frontmatter shape, secret redaction, MEMORY.md indexing./wrapnever writes a canonical memory itself. Rejected candidates are simply not promoted; they need no further action, since step 6 discards everything anyway. - Clear the scratch (auto, after promotion resolves). Un-promoted scratch is deleted — a wrong observation must not outlive the session that produced it. The clear reports how many entries were discarded, so a session that promoted nothing still says so out loud rather than silently dropping its accrual.
Skip the whole step silently when the session wrote no scratch. Ordering is load-bearing: the clear runs after promotion resolves, never before — an aborted or declined promotion must leave the entries readable, not consume them.
3. Recap (gather/draft auto; writes gated) — invoked with the under-wrap signal
Invoke recap with the --under-wrap signal so it suppresses its own §7.4 close-out/changelog doorways — /wrap owns those offers (steps 4–5), and a double-offer is the parallel-surfaces-drift landmine (two surfaces of one contract diverge and mask each other). recap still runs its gather + draft (auto) and its §9 bloat nudge; its file write, promotion→capture, and scoreboard write stay gated.
4. Close-out (conditional — only if a slug shipped; fully gated)
If the session shipped a slug (a .claude/specs/<slug> + .claude/plans/<slug> whose done-criteria are met), invoke close-out <slug> — reconcile spec + retire artefacts + record the ship line. Fully gated (it moves files + edits the canonical spec).
Dedup (load-bearing): close-out's record step already invokes update-changelog. So when /wrap runs close-out, it does not separately run step 5 for that slug — close-out is the superset, not a second changelog offer.
5. Update-changelog (conditional — only if a non-slug ship AND close-out did not run)
If the session shipped a non-slug unit of work (a multi-organ change, a fix — something with no spec/plan to close out) and step 4 did not run for it, invoke update-changelog directly — compose the ship line (draft, auto), prepend it (gated). Skip silently if nothing shipped.
6. BACKLOG-shipped-row sweep (detect auto; removal gated)
Sweep BACKLOG.md for rows whose unit already shipped — cross-reference each row against CHANGELOG.md and the .claude/archive/<date>-<slug>/ close-out folders. This closes the currently-unowned gap: a unit shipped without a changelog line orphans its BACKLOG row forever, and no other organ sweeps for that independent of a ship event. Detection is auto (a slug/unit-id match; model judgment only for ambiguous rows — never guess a row shipped); the row removal is gated (propose the retirement, accept/edit/reject). See the sweep detail in backlog-sweep.mjs when the detection warrants a script. Skip silently if there's no BACKLOG.md in the project.
7. Ship-narrative-record offer (conditional — on a real ship; gated)
On a real ship, offer a gated update to the project's ship-narrative record (a build-progress memory or equivalent), if it keeps one — its narrative shipped-list: "Log this ship to the ship-narrative record? (y/N)". It is a memory (canonical write) → propose-confirm, never auto-write. Skip silently when nothing shipped or the project keeps no such record.
8. Close
Report what ran: which steps fired, which were skipped, what's staged for the user's gates. Do not auto-commit anything. Do not re-offer close-out/changelog if step 4 already handled the slug.
What wrap does NOT do
- Not a new closing organ. A driver over recap / close-out / update-changelog / capture / coordinate. It sequences and dedups them; it never reimplements their logic.
- Not silent-write. Every canonical write (recap file, capture, scoreboard, spec reconcile, artefact move, changelog line, BACKLOG-row removal, ship-narrative-record edit) passes accept/edit/reject. Auto = auto-propose.
- Not on by default. The auto-fire hook ships
enabled: false; the manual/wrapworks with no hook at all. - Not a Stop/SessionEnd hook.
SessionEnd/PreCompactstdout is debug-only and exit-2 blocks, so a non-blocking guard can't surface there; auto-fire is SessionStart-recovery only. - Not mid-session supervision. No background watcher; the auto-fire judgment runs once, at the next SessionStart.
- Not a double-offerer. When close-out runs (step 4), the changelog (step 5) and its re-offer are suppressed — close-out contains the changelog write.
Relationship to other organs
- recap — invoked at step 3 with the
--under-wrapsignal so it suppresses its §7.4 doorways (which/wrapowns). recap standalone keeps §7.4 unchanged. - close-out — step 4 (only on a shipped slug); it contains the changelog write, so it dedups step 5.
- update-changelog — step 5 (only on a non-slug ship when close-out didn't run); the one changelog writer.
- checkpoint — optional step 2; it also reads the autonomy contract (the pattern
/wrapextends to the whole ceremony). Its reconcile must not double-run against close-out's. - coordinate — step 1 teardown; coordinate's auto mode registers the row at session start and shares
/wrap's SessionStart hook home. - capture — recap's promotion pass routes here (gated); same auto-propose-never-write shape as the auto-fire.
- autonomize — the contract
/wrapreads once at entry to set the interrupt posture. - clear-context guard — the SessionStart-recovery hook pattern the auto-fire inherits (and the SessionEnd-can't-surface constraint).
- prep — the same all-read-only auto-fire-on-SessionStart pattern.
- Governed by the autonomy write-class line — automate the reversible spine, gate every canonical write.
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.

