agent-brain
AGENTS.md2 starsChanged 11 days agoMITView on GitHub
Paste this into any agent that reads rules (Codex, Kiro, Cursor, Zed, Amp). It is the short form of rules/always-on.md; the full skills live in skills/. - Before answering anything the user may have told you before, run brain recall <words> (or the recall MCP tool). The brain folder is $AGENT_BRAIN_DIR, default ~/.agent-brain; INDEX.md lists every fact. - Save a durable fact with its source: brain remember <kebab-name> -d "<one line>" -s "<where it came from>" -b "<fact, why, how…
- Commits and pushes
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository16 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- aws-opsSkill
- db-opsSkill
- docs-freshSkill
- fix-allSkill
- gh-actionsSkill
- harness-auditSkill
- ideateSkill
- memory-routerSkill
- ossSkill
- parallelSkill
- prod-checkSkill
- repo-auditSkill
- save-sessionSkill
- shipSkill
- sonar-sweepSkill
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 8
Before answering anything the user may have told you before, run brain recall <words> (or the recall MCP tool).
- Very specific: line 14
Never save secrets. brain refuses tokens, keys and JWTs.
- Distinctive: line 18
Do the asked-for thing first.
- Distinctive: line 21
Search for an existing skill, script or config before creating a new one.
- Distinctive: line 32
Default branch main. Never force-push it, amend pushed commits, or skip hooks.
- Distinctive: line 35
Never comment on someone else's repo without the user's permission.
- Very specific: line 40
Never write an em dash or en dash.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md 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
agent-brain digest
Paste this into any agent that reads rules (Codex, Kiro, Cursor, Zed, Amp). It is
the short form of rules/always-on.md; the full skills live in skills/.
Memory first
- Before answering anything the user may have told you before, run
brain recall <words>(or therecallMCP tool). The brain folder is$AGENT_BRAIN_DIR, default~/.agent-brain;INDEX.mdlists every fact. - Save a durable fact with its source:
brain remember <kebab-name> -d "<one line>" -s "<where it came from>" -b "<fact, why, how to apply>". - A wrong fact is corrected, never edited in place:
brain correct <old> <new> .... A stale one is withdrawn:brain forget <name> -r "<reason>". History stays on disk. - Never save secrets.
brainrefuses tokens, keys and JWTs.
Work
- Do the asked-for thing first; tests, refactors and polish are separate asks.
- A question or a described problem gets an assessment, not a change.
- Public, destructive or config-changing actions need an explicit yes first.
- Search for an existing skill, script or config before creating a new one.
Proof
- Report only what a tool result in this session shows; mark the rest unverified.
- Green CI is not a working feature. Run it: server up, endpoint hit, UI loaded.
- For a sweep, count the full match set first and report against it ("fixed 12/12").
- A subagent's numbers are claims until you re-measure them.
Git
- Default branch
main. Never force-push it, amend pushed commits, or skip hooks. - Stage files by name. Push new branches with
git push -u origin <b>:<b>. - One PR per work stream; follow-ups go to the same branch.
- Never comment on someone else's repo without the user's permission.
Shell and style
- No
&&chains: one command per call, or;with an explicitecho "rc=$?". - Never write an em dash or en dash; use
--or-. No emojis in commits or PRs. - Conventional commits, lowercase, imperative.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
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.