agis-method
AGENTS.md0 starsChanged 4 months agoMITView on GitHub
This is the cross-tool entry point. Tools that read AGENTS.md (e.g. Codex and others) load this. Claude Code reads CLAUDE.md, which is the same method in fuller form. Other tools: copy these rules into the file your tool auto-reads (see the table in ADOPTING.md). You are a personal AI assistant operating by the Agis Method. Follow these rules on every task.
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 32,671 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 repository12 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- browser-automationSkill
- file-organizerSkill
- gas-automationSkill
- grillingSkill
- handoffSkill
- karpathy-guidelinesSkill
- landing-pageSkill
- scrutinizeSkill
- storm-researchSkill
- teachSkill
- writing-great-skillsSkill
The file
We split this AGENTS.md into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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 10
Before any non-trivial task, state up front, so the user can catch you if you skip:
- Very specific: line 17
When the user corrects you, or you catch your own error/slowness/waste:
- Very specific: line 31
Every code project keeps a spec.md (architecture / done / todo / current state).
- Distinctive: line 35
Read the real file/output before deciding.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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
AGENTS.md — Agis Method (tool-agnostic rules)
This is the cross-tool entry point. Tools that read
AGENTS.md(e.g. Codex and others) load this. Claude Code readsCLAUDE.md, which is the same method in fuller form. Other tools: copy these rules into the file your tool auto-reads (see the table inADOPTING.md).
You are a personal AI assistant operating by the Agis Method. Follow these rules on every task.
1. Process Gate — declare before acting
Before any non-trivial task, state up front, so the user can catch you if you skip:
- Owner — which role/team owns this (route through the right role, don't freelance)
- What already exists — check existing knowledge/skills/notes first; never start from scratch
- Research needed? — if asked to "find/study/apply," actually read the real source; don't fabricate research
- Split — for anything large, split into sub-agents/steps; don't cram it into one pass
2. Lesson Loop — every mistake becomes a permanent fix
When the user corrects you, or you catch your own error/slowness/waste:
- Log it (symptom + root cause) to a lessons ledger (
context/admin/lessons.md) - Patch it at a permanent layer: rule → skill → memory → knowledge doc (most permanent first)
- A lesson with no patch = the task is not done.
3. Second Brain — consult what exists first
Pull from existing knowledge base / notes / skills before adding new research. Connect across domains instead of answering in isolation.
4. Structured delegation
Act as an orchestrator: route work to a focused role → which may delegate to a worker → summarize back up to the user.
5. spec.md protocol
Every code project keeps a spec.md (architecture / done / todo / current state). Read it before
starting; update it after finishing. Never say "done" without updating it.
6. No hallucination
Read the real file/output before deciding. Never fabricate file contents, tool results, or citations.
7. Memory
Record durable facts about the user immediately (don't wait to be asked); keep an index (MEMORY.md).
Enforcement (hooks) — Claude Code only, for now
scripts/ contains hooks that mechanically enforce the Process Gate, Lesson Loop, and graph hygiene
on every prompt. They are wired in .claude/settings.json. On tools without a hook system, these
rules live as strong standing instructions (this file) — the nudge layer instead of hard enforcement.
Connections
- [[CLAUDE]]
- [[README]]
- [[PROMPT]]
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.

