continuity
Skill5 starsChanged 57 days agoMITView on GitHub
Persist and resume long-running project state across AI sessions. Use at the start of work on an ongoing project to load prior decisions, constraints, and rejected approaches; and whenever the user makes a decision, sets a constraint, rejects an approach, freezes something, or reaches a milestone — to record it so future sessions don't lose it or contradict it.
7 other projects use a version of this.
Adopted by
Most recently taken up by
- hiroqt/overdrivetheir own version, first seen Nov 1, 2026
- wangsalin/ai-acting-systemtheir own version, first seen Oct 13, 2026
- inwrkai/continuitytheir own version, first seen Aug 6, 2026
- mychaelconnolly/continuity-skilltheir own version, first seen Jul 3, 2026
- tance-mang/chinese-webnovel-skillstheir own version, first seen Jun 12, 2026
- kevinmanase/continuity-agent-skillstheir own version, first seen May 22, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- tance-mang/chinese-webnovel-skills★ 97same name
- hiroqt/overdrive★ 2same name
- openmetaloom/skills★ 2same name
- wangsalin/ai-acting-system★ 2same name
- inwrkai/continuity★ 1same name
- kevinmanase/continuity-agent-skills★ 1same name
- mychaelconnolly/continuity-skill★ 0same name
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
4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
2
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
- Very specific: line 14
Call resume_context (pass project if the user names one).
- Very specific: line 18
If nothing is returned and the work is clearly a new ongoing project, offer to create_project.
- Distinctive: line 35
Do NOT record: your own explanations, restatements of existing claims, progress narration, or a rewrite of the next step unless the next …
- Distinctive: line 57
Never silently contradict a frozen claim: If the user's new direction conflicts with a frozen item, surface the conflict and confirm before …
“Common” is measured: kept by at least 6 in 10 of the 5 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
Continuity — keep long-running project state
A continuity MCP server stores durable project state (decisions, constraints, rejected alternatives, open questions, next steps) as versioned claims. Your job is to keep it current and to honor it — so a future session resumes correctly instead of re-litigating settled decisions.
At the start of a session on an ongoing project
Call resume_context (pass project if the user names one). Treat the result as authoritative:
- 🔒 FROZEN items must not be changed or reopened.
- 🚫 Do NOT revisit items are already rejected — do not re-propose them.
- Continue from the Resume at next step.
If nothing is returned and the work is clearly a new ongoing project, offer to
create_project.
While working — capture autonomously (do not ask permission)
Default to capturing NOTHING. Most turns warrant nothing. Record only what a future session could not re-derive, in one of these shapes:
- A decision the user settles on →
record_decision(include the reasoning in the body). - A constraint future work must respect →
record_constraint. - An alternative considered and rejected →
record_rejectionwith the reason. A rejection without a reason is refused: the reason is what stops it being re-proposed. - An open question, risk, milestone, or next action →
record_open. - FRAMING — a statement setting strategy or what matters ("X is the moat", "Y is the real bottleneck"). These do not look like decisions and are the most commonly missed. Record as a decision.
- A STANDING INSTRUCTION about how to operate or who decides ("never do X without asking", "you have full ownership"). Record as a constraint, never as a question — a question reads as an open topic rather than a rule to obey.
Do NOT record: your own explanations, restatements of existing claims, progress narration, or a rewrite of the next step unless the next step actually changed. Prefer superseding an existing claim over adding a near-duplicate, and keep bodies short — they are re-read in every future session. If in doubt, stop without capturing.
Discussion is not decision, and agreement between assistants is not approval. Only the user's commitment makes something durable project state.
For several claims at once prefer capture, which de-duplicates and parks anything contradicting
a frozen claim. The individual record_* tools write straight through and do neither.
Deliberate acts — only on explicit user request
- "freeze X" / "lock X" →
freeze_claim. - "why did we change X?" →
why. - Closing a claim →
resolve_claim, always with a reason. - Setting or changing the mission →
record_mission; replacing one needs a reason. - Finding a claim the resume context does not show →
search_claims. The projection is budgeted and never lists superseded, rejected or resolved claims.
Principles
- Record the REASON, not just the decision — a future session must know why a path was chosen or rejected, or it will be argued back into a reversal.
- Never silently contradict a frozen claim. If the user's new direction conflicts with a frozen item, surface the conflict and confirm before proceeding.
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.

