vault-checkpoint
Skill2 starsChanged 55 days agoMITView on GitHub
Close a working session, or checkpoint after a single task, so the next session can pick up cold without losing a decision. Use when the user says "close the session", "checkpoint", "wrap up", "save state", "I'm done for today", when a task finishes, or when context is about to run out. Runs the seven-step routine: Current-State, LOG, index, lessons, private journal, clean working tree, and evidence. Do NOT use it to audit vault health (that is vault-lint) or to initialise a vault (that is vault-bootstrap).
In this repository2 other files it gives its agents.
- vault-bootstrapSkill
- vault-lintSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
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 20
Run the full routine at the end of a session, and a reduced version after
- Very specific: line 25
Never take the date from the prompt, a filename, or memory: Run the system date command in the first turn that is going to write anything …
- Very specific: line 65
If the LOG passed its ceiling, rotate it (see vault-lint).
- Distinctive: line 69
Update it if pages were created, moved, or re-described.
- Distinctive: line 82
If the project uses one, add a raw entry.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Checkpoint a session
Why after every task, not just at the end
Long sessions die without warning: compaction, token limits, a model change. A checkpoint that only runs "at the end" is a checkpoint that often never runs.
Run the full routine at the end of a session, and a reduced version after each task.
Before anything: get today's date
Never take the date from the prompt, a filename, or memory. Run the system
date command in the first turn that is going to write anything dated: a file in
90-Raw/, a LOG entry, an updated: field, a note in an external system.
A prompt written on Friday and executed on Monday is a stale clock. A long run with retries can cross midnight mid-execution, so on a long session run it again before stamping any new snapshot.
date +%Y-%m-%dThe seven steps
1. Current-State.md reflects reality
Present tense only: milestone, what closed, what is in flight, what is blocked and with what diagnostic suspicion, visible debt, risks.
It must end with a "NEXT SESSION" section: what comes next, in what order,
and which decisions are waiting on the human, listed explicitly. An agent
starting cold, possibly a different model, has to continue from
CLAUDE.md → Current-State.md alone without losing a decision.
Stay under the 2,500 token ceiling. If you catch yourself narrating how the project got here, that belongs in the LOG.
Record blockers with their suspicion, not just their existence. "Hangs; suspicion: resource contention; next step: X" means the next session starts investigating instead of rediscovering.
2. LOG.md
One entry per operation this session, newest at the top:
## [YYYY-MM-DD] op | short title
What happened, decisions made, pointers.If the LOG passed its ceiling, rotate it (see vault-lint).
3. 00-Index.md
Update it if pages were created, moved, or re-described.
4. Lecciones.md
If the session paid for a lesson, write it before closing. A lesson that does not get written gets paid for again.
Each entry is an actionable rule with its reason: never X, on this date it caused Y, do Z instead. This file is overwritten and refined, not appended to: a lesson that turns out to be incomplete gets corrected on the spot.
5. Private journal
If the project uses one, add a raw entry. It never goes to the repo. The curated LOG entry can be distilled from it later.
6. Clean working tree
Descriptive commit and push. Nothing stays uncommitted unless the human
explicitly decides so, and then it is noted in Current-State.md as WIP.
In a multi-axis repo, stage only the axis you worked on. Files another axis left modified are not yours to commit.
7. Nothing is reported as finished without evidence
Green gates, or reviewable artefacts. Anything unverified is marked "pending sign-off", without embarrassment.
And evidence only counts if the instrument is validated. A green result from a suite with no control case proves nothing (VAULT-STARTER §8.5). Before reporting a measurement, ask whether the instrument would have detected the failure had it occurred.
The reduced version, after each task
Steps 1, 2 and 6. Thirty seconds. It is what makes the difference between losing one task and losing a session.
Anti-goals
- Do not write history into
Current-State.md. It is the present. The LOG is the past. - Do not summarise when archiving. Rotating a LOG preserves it verbatim; summarising destroys exactly the detail it exists for.
- Do not mark something ratified because the human liked it in chat. Ratification is explicit or it is not ratification. Ask, and record it with a date.
- Do not commit another axis's changes to tidy the working tree.
- Do not close with a lesson unwritten. That is the one step whose cost is invisible today and certain later.
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.

