orchestra-workflow
AGENTS.md1 starsChanged 18 days agoMITView on GitHub
You are running Orchestra 3.3.6. One contract, one conductor, open skills. Not a second toolbox. ORCHESTRA = CONTROL PLANE. SKILLS / MCPs / PLUGINS / LIBRARIES = CAPABILITIES. AGENTS = EXECUTORS. BRAIN = MEMORY. REGISTRY = RESOURCE KNOWLEDGE. Host rules are adapters. They do not invent a second plan. If two hosts could plan, the chat the human is talking to is the conductor. Print Load: and Skip: before writing. A skill the human names wins. A missing tool stops.…
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 repository67 other files it gives its agents.
- skills/humanizer/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agent-reachSkill
- animateSkill
- backend-architectureSkill
- ci-security-scanning-with-strixSkill
- diagram-generatorSkill
- eas-app-storesSkill
- emil-design-engSkill
- expo-data-fetchingSkill
- expo-dev-clientSkill
- expo-native-uiSkill
- expo-project-structureSkill
- expo-routerSkill
- expo-upgradeSkill
- final-showcaseSkill
- fix-security-vulnerabilities-with-strixSkill
- fresh-context-reviewSkill
- gsap-coreSkill
- gsap-performanceSkill
- gsap-reactSkill
- gsap-scrolltriggerSkill
- gsap-timelineSkill
- hallmarkSkill
- humanizerSkill
- impeccableSkill
- ml-experimentSkill
- obsidian-markdownSkill
- orchestra-conductorSkill
- orchestra-docsSkill
- orchestra-shipSkill
- orchestra-vaultSkill
- penetration-testing-with-strixSkill
- ponytail-auditSkill
- ponytail-reviewSkill
- ponytailSkill
- postgres-best-practicesSkill
- pretty-mermaidSkill
- prompt-forgeSkill
- r3f-shadersSkill
- react-best-practicesSkill
- reaSkill
- refero-designSkill
- research-paperSkill
- resource-scoutSkill
- review-animationsSkill
- semgrep-adapterSkill
- ship-safeSkill
- skill-security-checkSkill
- stitch::code-to-designSkill
- stitch::extract-design-mdSkill
- stitch::extract-static-htmlSkill
- stitch::generate-designSkill
- stitch::manage-design-systemSkill
- stitch::react-componentsSkill
- stitch::upload-to-stitchSkill
- store-releaseSkill
- supabaseSkill
- systematic-debuggingSkill
- taste-designSkill
- The first 60 shown.
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 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.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 43
Implement: only after the lab is cleared for visual work.
≈ “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 — Orchestra 3.3.6
You are running Orchestra 3.3.6. One contract, one conductor, open skills. Not a second toolbox.
ORCHESTRA = CONTROL PLANE. SKILLS / MCPs / PLUGINS / LIBRARIES = CAPABILITIES. AGENTS = EXECUTORS. BRAIN = MEMORY. REGISTRY = RESOURCE KNOWLEDGE.
Host rules are adapters. They do not invent a second plan. If two hosts could plan, the chat the human is talking to is the conductor.
Loop
- Forge.
prompt-forgewrites a short brief: intent, kind, route, acceptance, refs, constraints, out of scope. Then work. Ask only when a product decision is truly ambiguous. If it names a repo or a file, open the repo before trusting any brief. - Classify one route from
registries/design-resource-graph.json. Load that route, not every skill. The graph (registries/design-resource-graph.json) is capability routes. The registry is resource knowledge. - Kind sets the design bar.
skip the labstill bypasses it.
| kind | Bar | Lab |
| hiring-cv | FLAGSHIP | 23 cards, or 3 translations for a named reference, then contract, then stills |
| personal or client | PREMIUM | One human message approves the contract and the stills |
| college | STANDARD | No lab unless the human says PREMIUM |
| backend, fix, glue | STANDARD | No lab |
<!-- routing:start -->Routing
Print Load: and Skip: before writing. A skill the human names wins. A missing tool stops. It does not install itself.
| Job | Load | Skip |
| Award-style site | impeccable, taste-design, visual-golden-still, gsap-core, gsap-scrolltrigger, gsap-timeline | copying a reference site's files; a second motion stack |
| Site to echo | visual-forensics, impeccable, taste-design | copying assets, copy, or a logo |
| Scroll | gsap-core, gsap-scrolltrigger, gsap-timeline | gsap-react unless the motion lives in React components |
| One 3D scene | r3f-shaders | 3D as decoration; a second motion engine |
| SaaS or API | backend-architecture, test-driven-development | the design lab; supabase and postgres-best-practices unless that product uses them |
| Bug | systematic-debugging | the design lab |
| Paper | research-paper, humanizer | invented citations |
| Deck | orchestra-docs | host-provided pptx unless that host already has it |
| README and post | orchestra-ship, humanizer | posting for the human |
| Binary or APK | not ready | rea until Java is installed and the target is yours, a client's, a CTF, or open source |
| Hackathon or college | prompt-forge | the 23-card survey when a reference is named; the design lab for college unless the human says PREMIUM |
| Docs on disk | host-provided | docx, pptx, pdf, and xlsx stay on the host. They are not in skills/ |
<!-- routing:end -->- Implement only after the lab is cleared for visual work. A DESIGN.md is not visual evidence.
- Verify on the running app.
verification-before-completionplus the ledger. Thenfresh-context-review. - Remember what actually worked. Do not invent usage.
Engineering
| Job | Skill |
| Bug | systematic-debugging |
| Logic | test-driven-development |
| Before DONE | verification-before-completion |
| Diff | ponytail on engineering only, never on a Design Lab surface |
| Parallel hosts | using-git-worktrees and protocols/PARALLEL_WORK_PROTOCOL.md |
Research and reverse engineering
Agent-Reach reads the operator's own profiles and public pages. Read-only. Cookies stay local. No bulk scraping of other people. No post, like, DM, or apply. No ATS scrapers.
rea and APKLab run only on a target the task names as own, client, CTF, or open source.
Evidence
You may write DONE / FIXED / VERIFIED / PASSED / SHIPPED only when observed evidence is in the same message. Intention is not evidence.
Resources
Available is not loaded. A registry row is not proof the tool ran. New third-party skills pass skill-security-check before promotion. Never skills add --all.
Hosts
The same skills and the same core MCPs. Exceptions live in registries/host-stack.json. kit/orchestra-doctor.ps1 checks drift.
| Host | Role |
| Cursor | Usual conductor |
| Antigravity | Visual QA |
| Claude Code | Terminal and review. The hook reminds Forge. It does not rewrite the prompt |
| Paseo | Launches workers. Not a conductor |
| Codex, jcode | Same skills |
Overrides
- skip orchestra stands this down for the session.
- skip the lab bypasses the design gate for one task.
ORCHESTRA_CONTRACTcan pin an older contract. Rollback notes are inkit/ROLLBACK.md.
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.