claude-code-orchestration-template
AGENTS.md1 starsChanged 30 days agoMITView on GitHub
Orientation for AI coding agents (OpenAI Codex, Cursor, and similar) working on this repository. It does not replace CLAUDE.md; that file is the operating policy and wins on any conflict. This file just helps you get oriented fast. A documentation template for orchestrating Claude Code. The deliverable is a set of instruction files - a policy (CLAUDE.md), four subagents in .claude/agents/, and one skill in .claude/skills/ - that a developer copies into their own repo. There is no application code,…
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 31,700 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 repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- delegation-protocolSkill
- x-searchSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 25
Keep it declarative: No scripts, dependencies, package managers, binaries, hooks, CI, or telemetry.
- Very specific: line 28
Preserve the gates: Don't weaken or remove the spend gate (PREMIUM-APPROVED) or the author-is-not-the-reviewer rules.
- Distinctive: line 34
No tests exist. Sanity-check with git status --short, git diff --check, and confirm any Markdown links you add resolve to real files.
≈ “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
AGENTS.md
Orientation for AI coding agents (OpenAI Codex, Cursor, and similar) working on this repository. It does not replace CLAUDE.md; that file is the operating policy and wins on any conflict. This file just helps you get oriented fast.
What this repository is
A documentation template for orchestrating Claude Code. The deliverable is a set of instruction files - a policy (CLAUDE.md), four subagents in .claude/agents/, and one skill in .claude/skills/ - that a developer copies into their own repo. There is no application code, no build step, and no runtime. The Project commands placeholders in CLAUDE.md exist for downstream users to fill in their repositories; this template itself has nothing to build or test.
Map
| Path | Purpose |
README.md / README.es.md | Human-facing docs. English is canonical; Spanish mirrors it. |
CLAUDE.md | The orchestration policy - behavioral source of truth. |
.claude/agents/*.md | The four subagents: fast-worker, deep-reasoner, premium-reasoner, diff-reviewer. |
.claude/skills/delegation-protocol/SKILL.md | Delegation mechanics: task-spec template, routing table, review rules. |
.claude/skills/delegation-protocol/codex-review-transport.md | Skill §6, read on demand: Codex review transport, rollout verification, finding validation. |
.claude/skills/delegation-protocol/grok-fallback-review.md | Skill §7, read on demand: the optional Grok Build fallback review. |
.claude/skills/x-search/SKILL.md | X (Twitter) search through the Grok Build CLI - Claude's web tools cannot reach x.com. |
docs/example-workflow.md | Worked end-to-end example of the delegation protocol. Documentation only - not part of the install payload. |
llms.txt | Compact machine-readable map of the project. |
Working rules
- Keep it declarative. No scripts, dependencies, package managers, binaries, hooks, CI, or telemetry. Everything here is plain Markdown by design.
- One source of truth. Project-specific content (risk paths, validation commands, data boundaries) belongs only in
CLAUDE.md. Don't copy it into the agents or the skill. - Keep the READMEs in sync. A change to
README.mdthat affects meaning must be mirrored inREADME.es.md. English wins on any discrepancy. - Preserve the gates. Don't weaken or remove the spend gate (
PREMIUM-APPROVED) or the author-is-not-the-reviewer rules. If you rename the token or an agent, rungit grep -l PREMIUM-APPROVEDand update every hit in the same change. - Surgical diffs. Match the existing voice; no broad cosmetic rewrites.
- Don't strip provenance. Dated compatibility stamps (e.g., the Codex CLI version) are deliberate - update them when you re-verify, never delete them.
Validation
No tests exist. Sanity-check with git status --short, git diff --check, and confirm any Markdown links you add resolve to real files.
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.