skills / codex
AGENTS.md1 starsChanged 13 days agoMITView on GitHub
The main session decides, writes briefs and reviews. Mechanical work goes to custom agents (installed in ~/.codex/agents) with fresh context on cheaper models, so raw output stays out of the main context. Details: skill esoul-model-routing. <!-- esoul-skills:end -->
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 31,439 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 repository15 other files it gives its agents.
- esoul-data-task-guardSkill
- esoul-freelo-task-authoringSkill
- esoul-maintain-application-documentationSkill
- esoul-mockup-authoringSkill
- esoul-model-routingSkill
- esoul-project-dataSkill
- esoul-project-lifecycleSkill
- esoul-taskingSkill
- esoul-uatSkill
- esoul-workflowSkill
- final-pr-reviewSkill
- freeloSkill
- gitmoji-commitSkill
- pre-pr-reviewSkill
- test-queueSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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 6
Search or reading across files, logs, exports: agent scout (prefer it over the built-in explorer).
- Distinctive: line 15
After two failed worker attempts, take over yourself.
- Very specific: line 18
Keep auto-compact at about 250K (model_auto_compact_token_limit).
≈ “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
<!-- esoul-skills:begin -->eSoul usage rules (esoul-skills)
The main session decides, writes briefs and reviews. Mechanical work goes to custom agents (installed in ~/.codex/agents) with fresh context on cheaper models, so raw output stays out of the main context.
- Search or reading across files, logs, exports: agent
scout(prefer it over the built-in explorer). - Tests, build, lint:
test-runner. - Implementation of size M and larger:
implementer. - Review lenses:
reviewer. Risky changes (authorization, money, migrations, personal data, concurrency):critical-reviewer. - UAT chapters:
uat-runner. - Commits, gh, Freelo chores:
ops. - Any other subagent: set a cheap model explicitly (
gpt-6-lunafor search and runs,gpt-6.1-terrafor implementation and review). Do not spawn agents that copy the whole conversation. - Main session:
gpt-6.1-solat medium effort. Raise effort per task (high for planning, xhigh for L or security work). Usegpt-6-astraonly by exception, it is very expensive. - Size S: do it directly.
- After two failed worker attempts, take over yourself.
- One task per session. Do not read whole files, logs or JSON dumps into the main context.
- Session title: a cheap model names the session from the first exchange, so when the session works on a GitHub issue or a Freelo task, start the first reply with the line
#<issue number> <issue title>(Freelo-only work:Freelo <task id> <task title>), shortened to about 60 characters. Look the title up first if the prompt gives only a number or a link. - Keep auto-compact at about 250K (
model_auto_compact_token_limit); a larger context only for one short session that must hold a large input at once (prefer a subagent for it). - Batch independent tool calls. Wait for CI in the background, never with sleep loops.
Details: skill esoul-model-routing.
<!-- esoul-skills:end -->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.