save-session
Skill1 starsChanged 53 days agoMITView on GitHub
Use when the user asks to save this unfinished session for a fresh one. Not for a plan another executor runs (spec) or finished work.
19 other projects use a version of this.
Adopted by
Most recently taken up by
- hydra-db/hydradb-claude-codetheir own version, first seen Nov 8, 2026
- jemshit/save-sessiontheir own version, first seen Nov 8, 2026
- kriscard/Skillstheir own version, first seen Nov 1, 2026
- abhayla/claude-best-practicestheir own version, first seen Nov 1, 2026
- artificialbeingz/abz-karpathy-llm-wikitheir own version, first seen Oct 20, 2026
- Sagargupta16/agent-braintheir own version, first seen Oct 13, 2026
Used with
Projects with it also use status.
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.
AnthropicSkillstatus
Case status summary by audience — client-facing (plain language), internal (for the professor), or court-ready (formal caption format per local rules). Same facts, different framing and depth. Use when a student needs to update the client, brief the professor, or prepare a court status report.
Also in 3 projects that use this
Other versions19 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 19
- faizkhairi/claude-code-blueprint★ 73same name
- jemshit/save-session★ 60same name
- jiten-singh-shahi/salesforce-claude-code★ 15same name
- kriscard/Skills★ 15same name
- abhayla/claude-best-practices★ 9same name
- ingpoc/SKILLS★ 9same name
- david-vrba/claude-power-skills★ 4same name
- hydra-db/hydradb-claude-code★ 4same name
- Erionis/Jarvis-Lite★ 3same name
- Sagargupta16/agent-brain★ 2same name
9 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 repository10 other files it gives its agents.
- .claude/CLAUDE.mdCLAUDE.md
- buildSkill
- configureSkill
- edit-skillsSkill
- find-causeSkill
- gateSkill
- shipSkill
- specSkill
- watchSkill
- write-docsSkill
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 2 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 18 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
If the tree is broken, say so in that commit's body.
- Very specific: line 30
Read the location instead of guessing it: node "${CLAUDE_SKILL_DIR}/scripts/handoff.mjs" path prints it, the same file …
- Distinctive: line 32
Write the file whole, replacing any handoff already at that path: the state it held is what this clear discards.
- Very specific: line 43
When a plan is running with <plan stem>-decisions.md beside it, carry its lines into ## Decisions unchanged.
- Very specific: line 44
Put a claim no command proved under ## Open questions, not ## Proven.
- Very specific: line 45
Run node "${CLAUDE_SKILL_DIR}/../../lib/ruled-out.mjs" show <n> for the task in progress and carry its lines into ## Open questions as …
- Very specific: line 46
Read ## Current state from git status --short, not memory.
“Common” is measured: kept by at least 6 in 10 of the 18 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
Handoff
Freeze what this session knows and the next one cannot rebuild. The enemy is a handoff that pastes the transcript back, so the fresh session pays again for the context the clear just bought. The overcorrection is a page of headlines naming no file, command or next step, leaving the reader to rediscover the work.
Before writing
- Take no irreversible action to pause: no push, no pull request.
- Pause between atomic steps, not inside one: complete or undo the step in progress and begin no new work.
- On the run branch of a plan (
exo:buildloop), commit nothing. Write partial work of the task in progress to a hint file such as.exo/carry/task<n>-partial.patch, and list it under## Current state. - On any other branch than the repository's default, commit every uncommitted edit first as one
chore: wip <topic>commit. - If the tree is broken, say so in that commit's body.
- On the default branch, leave edits uncommitted and list them under
## Current state. - Preserve verbatim any artefact the user explicitly asked to survive the clear, such as a question list or checklist.
Pointers, not content
- Each section names a path, command, issue number or line range, and stops there.
- Paste no diff, file body or log; the reader can open them, and pasting spends the context the clear freed.
- For an error, paste only its failing lines.
- Quote a value only when it exists nowhere on disk, such as a number from an unlogged run.
Where it goes
- Read the location instead of guessing it:
node "${CLAUDE_SKILL_DIR}/scripts/handoff.mjs" pathprints it, the same filehooks/session-start.mjspoints a resuming session at. - A branch name holding
/gives a nested path, so create parent directories first. - Write the file whole, replacing any handoff already at that path: the state it held is what this clear discards.
What it holds
Fill assets/handoff-template.md. Its first line answers in one sentence where the work stands, and ## Resume is always present while the other sections are left out when the session has nothing true for them. The file is read by the user, so write plain words in the user's language per ../../lib/references/user-read-format.md.
The last line of the file is Resume: <prompt>, one sentence in the user's language telling a fresh session to read this handoff by the path handoff.mjs path printed and continue. The sections carry every step, so this one line suffices.
In those sections:
- Record the user's decision as theirs, in their wording, not as shared.
- When a plan is running with
<plan stem>-decisions.mdbeside it, carry its lines into## Decisionsunchanged. - Put a claim no command proved under
## Open questions, not## Proven. - Run
node "${CLAUDE_SKILL_DIR}/../../lib/ruled-out.mjs" show <n>for the task in progress and carry its lines into## Open questionsas approaches the next session must not retry. - Read
## Current statefromgit status --short, not memory. - List which changes are committed, then every uncommitted path in full, because a next session that cannot see the difference treats working-tree edits as saved and discards them.
- For a session running a plan, open
## Current statewith the plan path and the number of the task in progress.
Resume
Copy this paragraph into ## Resume exactly, so the reading session inherits the rule without this skill:
Diff the work this handoff records as done against the work it still lists as next or open, name the resume point, and repeat no step it records as proven. Verify an inherited claim against the real artifact before building on it: a prior session's report is not the proof.References
| File | Read it when |
references/reconstructing-without-a-note.md | Save-session itself starts without a note and must rebuild context before saving. |
Judgment
- File read this session outranks memory of it: confirm path, branch and commit with a command before writing them down.
- A short true handoff outranks a full one that guesses. Put what the session does not know under
## Open questions. - Turn ends naming the written path, saying to run
/clear, and quoting theResume:prompt to paste after it. Nothing else written; no work continues after.
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.

