developer-orchestrator
Skill1 starsChanged 5 months agoMITView on GitHub
Use when coordinating implementation of coding tasks with local work, subagents, or Codex background threads; assigning workstream ownership; handling multi-repo implementation; or deciding how scouts, workers, reviewers, and verifiers should collaborate under a Chief of Staff thread.
In this repository9 other files it gives its agents.
- accomplishment-journalistSkill
- automation-chiefSkill
- calendar-chiefSkill
- chief-of-staffSkill
- code-review-chiefSkill
- delivery-managerSkill
- inbox-triage-chiefSkill
- personal-chief-of-staffSkill
- relationship-managerSkill
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 9 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
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 12
Do not create a background thread merely because a task is non-trivial.
- Distinctive: line 13
Create a background thread when separation, persistence, independent steering, separate repo state, or a separate worktree is valuable.
- Distinctive: line 15
Create a new thread only for a new durable workstream or independent implementation attempt.
- Distinctive: line 16
Prefer subagents for disposable tasks that should return to the main thread: scouts, reviewers, verifiers, and narrow helper changes.
- Distinctive: line 17
Keep implementation local when the change is small, highly coupled, immediately blocking, or part of final integration.
- Very specific: line 21
Use codex_app.list_threads and codex_app.read_thread before creating new background threads when an existing workstream may already exist.
- Very specific: line 23
Use multi_agent_v1.spawn_agent for bounded parallel assistance that should return to the main thread.
- Distinctive: line 27
If Codex App thread tooling cannot target a requested VM or remote project, do not fabricate support.
- Distinctive: line 56
Start with scouts before implementation when repository boundaries, contracts, or conventions are unclear.
- Distinctive: line 58
Keep the canonical data model, API contract, vocabulary, and release order in the main thread.
- Distinctive: line 64
Before final delivery, inspect or reconcile every workstream result, verify tests or explain missing evidence, confirm scope against …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill 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
Developer Orchestrator
Coordinate engineering execution under a Chief of Staff model. The main thread owns intent, integration, verification, and final communication. Delegated agents or threads own bounded execution.
Decision Rules
- Do not create a background thread merely because a task is non-trivial.
- Create a background thread when separation, persistence, independent steering, separate repo state, or a separate worktree is valuable.
- Continue an existing thread for follow-up turns in the same workstream.
- Create a new thread only for a new durable workstream or independent implementation attempt.
- Prefer subagents for disposable tasks that should return to the main thread: scouts, reviewers, verifiers, and narrow helper changes.
- Keep implementation local when the change is small, highly coupled, immediately blocking, or part of final integration.
Coordination Primitives
Use codex_app.list_threads and codex_app.read_thread before creating new background threads when an existing workstream may already exist. Use codex_app.create_thread only for durable workstreams, and use codex_app.send_message_to_thread to steer those workstreams after creation.
Use multi_agent_v1.spawn_agent for bounded parallel assistance that should return to the main thread. Use multi_agent_v1.send_input for follow-ups, multi_agent_v1.wait_agent sparingly when blocked on the result, and multi_agent_v1.close_agent when the subagent is no longer needed. Use multi_agent_v1.resume_agent only when the previous subagent context materially helps.
Subagent handoffs should be narrow and disposable. Background-thread handoffs should be durable and include ownership, context, assumptions, expected output, verification, and blocker instructions.
If Codex App thread tooling cannot target a requested VM or remote project, do not fabricate support. State that the current tool surface cannot select that target, then choose the best available coordination path.
Delegation Modes
Scout: read-only investigation. Return patterns, relevant files, risks, and concrete recommendations. Do not edit.
Worker: bounded implementation. Own only assigned files/modules. Do not broaden scope or revert unrelated changes. Return changed files, commands run, deviations, blockers, and risks.
Reviewer: independent review. Prioritize bugs, regressions, missing tests, and architecture risks. Return findings first, ordered by severity, with file references.
Verifier: test/build/failure investigation. Return exact commands, pass/fail state, concise output summary, and likely cause. Edit only when explicitly assigned.
Thread Handoff Template
Role:
Goal:
Owned repo:
Owned files/modules:
Do not touch:
Context:
Accepted assumptions:
Expected output:
Verification:
Blockers:Multi-Repo Work
- Start with scouts before implementation when repository boundaries, contracts, or conventions are unclear.
- Map durable threads to ownership boundaries such as DB, backend, frontend, contract/integration, or separate implementation attempts.
- Keep the canonical data model, API contract, vocabulary, and release order in the main thread.
- Send precise deltas to affected threads when another workstream changes an assumption.
- Require each thread to report only deltas on follow-up turns unless a full recap is needed.
Completion Standard
Before final delivery, inspect or reconcile every workstream result, verify tests or explain missing evidence, confirm scope against acceptance criteria, and state residual risk.
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.

