session
Skill6 starsChanged 43 days agoMITView on GitHub
Close a work session — audit what was not saved, then record the session note and check the vault.
17 other projects use a version of this.
Adopted by
Most recently taken up by
- jskswamy/skillstheir own version, first seen Nov 8, 2026
- kimkon1011/orgiast-claude-rulestheir own version, first seen Nov 8, 2026
- mblauberg/provenanttheir own version, first seen Nov 8, 2026
- getCurrentThread/claude-session-kittheir own version, first seen Oct 29, 2026
- RainZZZYYY/claude-learn-skillstheir own version, first seen Oct 20, 2026
- charly-vibes/incitacionestheir own version, first seen Oct 14, 2026
Used with
Projects with it also use debug, release and 1 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.
AnthropicSkilldebug
Structured debugging session — reproduce, isolate, diagnose, and fix. Trigger with an error message or stack trace, "this works in staging but not prod", "something broke after the deploy", or when behavior diverges from expected and the cause isn't obvious.
Also in 3 projects that use this
MicrosoftSkillrelease
Use ONLY when the user explicitly asks to cut/publish a Huabu desktop release or tag a version. Performs an outward action — bumps the version in apps/desktop/package.json, commits + annotated-tags vX.Y.Z, and pushes the branch + tag. That tag push auto-builds the macOS/Windows installers and publishes a GitHub Release. Provides `status`, `bump`, `tag-push`, and a one-command `run` (bump → tag-push).
Also in 4 projects that use this
GoogleSkilltdd
Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.
Also in 3 projects that use this
Other versions17 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 17
- anthropics/claude-for-legal★ 9.5ksame name
- Tiendil/feeds.fun★ 397same name
- kagenti/kagenti★ 303same name
- Kanevry/session-orchestrator★ 53same name
- Tiendil/donna★ 29same name
- jayantak/vaultmem★ 3same name
- jskswamy/skills★ 3same name
- es6kr/skills★ 2same name
- Autocss-com/ai★ 1same name
- mblauberg/provenant★ 1same name
7 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.
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 4 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 14 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Use the conversation as evidence and git as the reality check.
- Very specific: line 21
Run python3 "${CLAUDE_PLUGIN_ROOT}/scripts/transcript.py".
- Very specific: line 57
Every thread must name what is unfinished and who finishes it.
- Distinctive: line 63
Never run or simulate tests to fill a gap.
- Distinctive: line 77
Write the session note (below).
- Very specific: line 78
Run health last — its commit is local to the vault, and asking for it separately turns a one-command close into a conversation.
- Very specific: line 80
Verify against the anchor: the notes exist in the vault and are indexed.
- Distinctive: line 120
Keep it factual and compact: commit ids, PR numbers, branches, failed approaches, and the constraints someone needs to pick the work back …
“Common” is measured: kept by at least 6 in 10 of the 14 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Close a session
Two things happen here, in this order: an audit of what the session leaves unsaved, and — only on the user's word — the writing. Showing is the default. An audit nobody acts on is a report; an audit that writes on its own fills the vault with paraphrase.
Use the conversation as evidence and git as the reality check. Never claim work git does not show, and never invent a candidate the conversation does not support.
Let a fresh reader do the audit
You lived through this session, so you are its worst auditor: your context is full, and whatever became routine along the way is now invisible to you.
Run python3 "${CLAUDE_PLUGIN_ROOT}/scripts/transcript.py". It prints this session's
transcript path and how many pieces it splits into. Start one sub-agent per piece, in
parallel, and give each the path it printed plus its own piece number —
transcript.py --path <path> --piece N — because a second session in the same project
would otherwise make a reader pick up someone else's conversation. Hand them the checklist
below; never paste the transcript into a prompt, since the whole point is that they read
the file rather than your recollection of it.
They return candidates and write nothing. You merge the lists, drop exact duplicates, and keep every disagreement visible — two readers of two pieces see different things, which is why it was split rather than duplicated. If no transcript is found, audit it yourself and say plainly that this is the weaker check.
A session that ended before the skill ran shows up in the startup digest and in
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/unclosed.py" --list. Audit it the same way: take
its path from there and run transcript.py --path <path> — that only marks it as closing;
it is recorded as closed when the session note is written.
The checklist for each reader: which requests were completed and which were not; which
decisions were made and whether they were saved; commits, pull requests and anything
changed outside this repository; TODO/later/FIXME left in the conversation or the code;
errors that were never resolved; questions that were never answered; code-bound "never X"
/ "always Y" rules that belong in the repository rules. Then git log --oneline -15 and
git status --short, and no claim that outruns them.
Present the candidates
Show the whole list, uncut — the user is the only filter that works, and trimming it takes away their chance to catch what you read wrongly. Give each candidate a destination: the vault, the repository rules, a runbook, or nowhere; when you are unsure, nowhere.
Sort every loose end by whether doing something kills it. "Merge PR #41" dies when someone
merges it — a thread. "The RCON password went out in public chat" describes how the world
is and no action erases it — knowledge, and it goes through that. Threads fall out of
the digest by freshness, so a fact parked among them is a fact thrown away.
Every thread must name what is unfinished and who finishes it. The most repeated kind of
thread in this vault was some wording of e2e not verified this session — 56 of 629
threads — and it named neither,
so no later session could act on it. Write "the release checklist in PR #41 has not been
run against prod — user runs it" instead.
Never run or simulate tests to fill a gap. Say which checks actually ran.
Write, once they say so
--full is consent for the whole chain. Without it, stop here and offer.
- Anchor first: the session's opening request,
git rev-parse HEAD,git status --short. - Save accepted knowledge through
that, one candidate at a time. - Log every refusal: pipe
{"verdict": "rejected", "query": "<the candidate>", "notes": [], "note": "<why they said no>"}topython3 "${CLAUDE_PLUGIN_ROOT}/scripts/feedback-log.py". Each "no" is data; in a month it says what actually gets rejected, and the filter can be built on that instead of a guess. - Close the threads that are finished (below).
- Write the session note (below).
- Run
healthlast — its commit is local to the vault, and asking for it separately turns a one-command close into a conversation. - Verify against the anchor: the notes exist in the vault and are indexed. Do not expect
them in
git status—healthhas just committed them. If nothing changed at all, sayalready in order, nothing to redo.
Threads that are already finished
A digest offering work finished three weeks ago teaches the user to stop reading it. Pass
the threads it still shows to
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/threads.py" --check "<thread as the digest shows
it>", and read what comes back as evidence about the thing named, not a verdict on
the thread: "backport the merged PR #41" mentions something merged and is not itself done.
Close without asking only when the thread asked for exactly what the evidence shows —
--close "<thread>" --reason "<the evidence>". Otherwise propose it, carrying that
evidence, and let the user decide. Close nothing when the user is not there: an open
thread costs a line in a digest, a wrongly closed one costs the work it was holding. Say
which threads you closed and on what. The note that recorded them is never edited.
The session note
Threads live only in this note's pending section; the digest reads them from the index.
- Title
Session — YYYY-MM-DD <short topic>, distinct from others that day — prefer a PR number, ticket or branch. The filename stem is the title. - Frontmatter:
type: session,date,tagswithsessionfirst,projectif known. - Body:
# Session — …, then## Done,## Decisions,## Next steps / pending. - Unfinished threads go only in
## Next steps / pending, as- [ ], each carrying the state someone needs to resume: uncommitted work, the attempt that just failed, the plan half-executed. Say what is in flight right now, not only what was finished — the end of a context is the end of a session, and this note is the only handoff. ## Links: a MOC if a fitting one exists (find it withsearch.py), otherwise the notes this session actually touched. Do not invent a hub to fill the slot.- Write only through
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/vault-write.py"withfilenameandcontent. If the name exists, offeraction: "replace"withexpected_sha; onconflict, stop and tell the user. - A note with no
[[link]]comes backrejected— the writer holds the graph's shape now.neighboursbesidecreatedare the nearest notes; offer them as links, add only the ones the user accepts. - Report the note name, the thread count, what entered the index, what stayed unsaved.
Keep it factual and compact: commit ids, PR numbers, branches, failed approaches, and the constraints someone needs to pick the work back up.
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.

