cognitive-update
Skill25 starsChanged 7 days agoMITView on GitHub
Update GOAL_STACK, BELIEF_STATE, SELF_MODEL, and PROCEDURES from explicit evidence. Use when updating cognitive state or runtime state.
In this repository71 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- docs/llms.txtllms.txt
- llms.txtllms.txt
- agent-fitnessSkill
- body-checkSkill
- briefSkill
- capsule-compactSkill
- closeSkill
- codex-adapterSkill
- comms-summarizeSkill
- compact-noteSkill
- context-capsuleSkill
- context-hygieneSkill
- critique-planSkill
- decideSkill
- deep-reconSkill
- DiagnoseSkill
- digestSkill
- dreamSkill
- EnvelopeSkill
- find-weaknessSkill
- First 10 MovesSkill
- first-winSkill
- frontend-slidesSkill
- Honesty CheckSkill
- humanize-docsSkill
- impeccableSkill
- learn-from-failureSkill
- LearnSkill
- memory-bridgeSkill
- meta-improveSkill
- meta-improve-vaultSkill
- nightly-close-paritySkill
- nightly-ledger-captureSkill
- nightlySkill
- openSkill
- operator-setupSkill
- orientSkill
- pauseSkill
- ponytailSkill
- port-specSkill
- PromoteSkill
- reconcileSkill
- research-deepSkill
- resumeSkill
- Sandbox RoutingSkill
- scout-vaultSkill
- Self-ReviewSkill
- semantic-searchSkill
- setupSkill
- ship-adapterSkill
- Skill AuditSkill
- skill-huntSkill
- skill-recallSkill
- solution-huntSkill
- sprite-specSkill
- startSkill
- statusSkill
- StuckSkill
- supersedeSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
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
- Very specific: line 63
Update only named capability, limitation, recurring-failure, reliability-risk, or achieved-learning-goal fields.
- Very specific: line 97
Only COGNITIVE_CONTRACT PASS is green.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
/cognitive-update
Nightly uses this shared protocol after /system-check. It never infers a
durable state change from one agent-authored claim.
Ownership fence
This protocol may update only:
vault/memory/runtime/GOAL_STACK.jsonvault/memory/runtime/BELIEF_STATE.jsonl(append-only)vault/memory/runtime/SELF_MODEL.jsonvault/memory/runtime/PROCEDURES.jsonl(append-only)
The documented memory/ fallback is valid for minimal installations. This
protocol never writes ACTIVE_STATE.json or STATE_EVENTS.jsonl; the runtime
state daemon owns those products. It never writes Dream or measurement ledgers.
Evidence admission
Direct updates require one of:
- the operator's explicit goal, belief, or procedure statement with a durable source reference;
- a repository-owned command or test invoked with explicit arguments and a literal exit code;
- an external artifact that directly decides the claim;
- two independent observed occurrences for a recurring capability or failure.
Capsules and Session Log entries are discovery sources. They cannot verify their own confident claims. If evidence is incomplete, skip the durable update and stage an observation through the nightly evidence flow.
File rules
GOAL_STACK.json
Move, append, or update only explicit goals, blockers, next actions, and completions. Completion requires success criteria plus an external artifact or operator confirmation. Preserve unknown fields and unrelated arrays.
BELIEF_STATE.jsonl
Append only. A revision reuses the original id and includes revision_note; a
new belief receives max numeric bNNN plus one. Do not use line count. Serialize
UTF-8 JSON and parse the appended line.
SELF_MODEL.json
Update only named capability, limitation, recurring-failure, reliability-risk,
or achieved-learning-goal fields. When alias arrays coexist, update each alias
pair in lockstep or fail SELF_MODEL_ALIAS. A one-off result is not a durable
capability.
PROCEDURES.jsonl
Append only when the sequence occurred at least twice, the operator explicitly
made it standing, or a non-obvious workflow was mechanically proven. Allocate
max numeric pNNN plus one and parse after append.
Write discipline
Hash every source immediately after read and compare again before write. A hash change aborts rather than clobbering newer state. Preserve JSON formatting where practical; always serialize JSONL.
A nightly fire uses the immutable snapshot created by nightly-pass begin.
For explicit use, create an operating-system temp snapshot:
node daemons/nightly-contracts.mjs snapshot --root <aigent-root> \
--out <os-temp>/cognitive-contract-before.jsonAfter any updates:
node daemons/nightly-contracts.mjs cognitive --root <aigent-root> \
--snapshot <os-temp>/cognitive-contract-before.json \
--new-belief-ids <csv> --new-procedure-ids <csv> \
--self-alias-additions <csv>Only COGNITIVE_CONTRACT PASS is green. The validator re-parses all four
products, proves append-only suffixes, derives allocated ids, validates new-row
schemas, and derives alias additions and removals.
Return:
COGNITIVE_UPDATE PASS goals=<N> beliefs=<N> self_model=<N> procedures=<N> skipped=<N>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.

