CITADEL
AGENTS.md4 starsChanged 4 months agoMITView on GitHub
An agent orchestration system for Codex. This repo IS the harness. The harness is a collection of skills, agents, hooks, and state files that enable Codex to operate autonomously on any codebase. It provides: - Intent routing (/do) that dispatches tasks to the right tool - Campaign persistence across sessions - Parallel agent coordination with discovery sharing - Automated quality enforcement through lifecycle hooks Always run tests after modifying hooks or skills. Five levels: Adding a new skill: create skills/{name}/SKILL.md,…
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 repository50 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- architectSkill
- archonSkill
- ascii-diagramSkill
- autopilotSkill
- costSkill
- create-appSkill
- create-skillSkill
- daemonSkill
- dashboardSkill
- decision-mapSkill
- deploy-stewardSkill
- designSkill
- doc-genSkill
- doSkill
- evolveSkill
- experimentSkill
- fleetSkill
- grillSkill
- housecleanSkill
- improveSkill
- infra-auditSkill
- learnSkill
- live-previewSkill
- loopSkill
- mapSkill
- marshalSkill
- merge-reviewSkill
- organizeSkill
- postmortemSkill
- prdSkill
- pr-watchSkill
- qaSkill
- refactorSkill
- research-fleetSkill
- researchSkill
- reviewSkill
- scaffoldSkill
- scheduleSkill
- session-handoffSkill
- setupSkill
- systematic-debuggingSkill
- telemetrySkill
- test-genSkill
- triageSkill
- unharnessSkill
- verifySkill
- watchSkill
- wikiSkill
- workspaceSkill
The file
We split this AGENTS.md into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 30
Always run tests after modifying hooks or skills.
- Very specific: line 43
Adding a new skill: create skills/{name}/SKILL.md, then run node scripts/skill-lint.js {name} to verify structure.
- Very specific: line 45
Adding a benchmark scenario: create skills/{name}/__benchmarks__/{scenario}.md, then run node scripts/skill-bench.js --skill {name} to …
- Very specific: line 59
After completing any task, output a HANDOFF block:
≈ “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
Codex Harness
An agent orchestration system for Codex. This repo IS the harness.
What This Is
The harness is a collection of skills, agents, hooks, and state files that enable Codex to operate autonomously on any codebase. It provides:
- Intent routing (
/do) that dispatches tasks to the right tool - Campaign persistence across sessions
- Parallel agent coordination with discovery sharing
- Automated quality enforcement through lifecycle hooks
Key Conventions
- Skills live in
skills/. Each skill is a directoryskills/{name}/SKILL.md. - Agents live in
agents/. They spawn as sub-processes with their own context. - Hooks live in
hooks_src/. Installed per-project viascripts/install-hooks.js. - The
init-projecthook auto-scaffolds per-project state on session start:.planning/— campaign state, fleet sessions, intake, telemetry.citadel/scripts/— utility scripts (telemetry, coordination, compression).claude/agent-context/— rules injected into sub-agents
- Campaign state lives in
.planning/campaigns/(per-project). - Fleet session state lives in
.planning/fleet/(per-project). - Templates live in
.planning/_templates/(copied from plugin on init). - Configuration lives in
.claude/harness.json(generated by/do setup, per-project).
Testing
Always run tests after modifying hooks or skills. Five levels:
| Command | What it checks | When to run |
node scripts/test-all.js | Hooks + skill structure (fast, no LLM) | After any change |
node hooks_src/smoke-test.js | Hooks only | After hook changes |
node scripts/verify-hooks.js | Hook install + runtime (synthetic payloads, no LLM) | After hook changes |
node scripts/integration-test.js | Full Pre→tool→Post pipeline sequences (no LLM) | After hook changes |
node scripts/skill-lint.js | All SKILL.md structure | After skill changes |
node scripts/skill-bench.js | Scenario file validity | After adding benchmarks |
node scripts/skill-bench.js --execute | Live scenario execution | Before shipping |
node scripts/skill-bench.js --execute --verify-hooks | Live execution + confirms hooks fired | Before shipping |
Adding a new skill: create skills/{name}/SKILL.md, then run node scripts/skill-lint.js {name} to verify structure.
Adding a benchmark scenario: create skills/{name}/__benchmarks__/{scenario}.md, then run node scripts/skill-bench.js --skill {name} to validate.
CI: node scripts/test-all.js exits non-zero on any failure. Use --strict to treat WARNs as failures.
Review Guidelines
Codex GitHub review should focus on actionable P0/P1 issues:
- correctness regressions in hooks, skills, routing, MCP, generated config, or campaign state
- security boundaries around shell commands, protected files, permissions, and unattended automations
- missing verification when a change affects hooks, skills, runtime adapters, or generated Codex artifacts
- stale generated artifacts that would mislead Codex users
Handoff Summary (Required)
After completing any task, output a HANDOFF block:
---HANDOFF---
- What was built or changed
- Key decisions and tradeoffs
- Unresolved items or next steps
---3-5 bullets, under 150 words.
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.