agentleFS
hyperb1isshyperb1iss/hyperskills/skills/orchestrate/SKILL.md

orchestrate

Skill36 starsChanged 21 days agoMITView on GitHub

Use this skill when coordinating independent agents with explicit ownership, dependencies, integration, and verification. Activates on mentions of swarm, parallel agents, multi-agent, orchestrate, fan-out, wave dispatch, research army, unleash, dispatch agents, or parallel work.

161 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use define-requirements, orchestrate-plan and 9 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • sson0-erSkill

    define-requirements

    Work with the user, one question at a time, to complete the requirements document (docs/claudex/<slug>/01-requirements.md). Run by the orchestrator (parent). Stays out of design and implementation.

    Also in 3 projects that use this

  • EternallLightSkill

    orchestrate-plan

    Use when executing a locked, reviewed PLAN.md by delegating the implementation to visible worker agents in Orca, herdr, or cmux ONE PHASE at a time — splitting the plan into context-window-sized phases, giving every phase a pristine worker context, and (when the plan touches INDEPENDENT repos) running one worker per repo IN PARALLEL. Triggers on "run the plan with a worker", "execute the plan in the other pane", "split the plan into phases and build it with workers", "build the backend and frontend in parallel panes". Not for edits you can finish inline; outside Orca, herdr, or cmux use tgc-skills:orchestrate.

    Also in 3 projects that use this

  • sson0-erSkill

    record-evidence

    How to record external facts (docs, command output, codebase findings) as evidence and design choices as decisions. Used by the researcher, designer and reviewer agents.

    Also in 3 projects that use this

  • PinkneyJiangSkill

    improve-architecture

    扫描代码库中的架构摩擦点,识别浅模块并挖掘深模块演进机会。生成可视化交互式 HTML 诊断报告,并在用户确认后闭环实施深度重构。

    Also in 4 projects that use this

  • kjnezSkill

    ticket

    Work on a JIRA/Linear ticket end-to-end: read ticket details, explore codebase, create a branch, implement with TDD, run quality checks, update ticket status, and create a PR. Use when the user provides a ticket ID to implement. Triggers on requests like "/ticket PROJ-123", "work on this ticket", "implement PROJ-123".

    Also in 4 projects that use this

  • xfstudioSkill

    reverse-engineer

    Expert reverse engineer specializing in binary analysis,

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 144 edited versions we read keep these.

  • 1

    Distinctive

    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
  1. Distinctive:

    Before dispatch, define how partial progress becomes integrated progress: the artifact each consumer expects and when it is ready.

    line 41
  2. Very specific:

    Read references/dispatch-briefs.md when composing worker, research, verification, or watcher instructions.

    line 99

“Common” is measured: kept by at least 6 in 10 of the 144 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Multi-Agent Orchestration

The coordinator owns integration and the final claim of completion.

The user's instructions take precedence over this skill; delegation, tools, and external actions remain subject to current host instructions and authorization, and this skill grants a worker no authority the coordinator lacks.

Check Capabilities Before Dispatch

Inspect the actual tools and current schemas instead of inferring capabilities from a product name. An external CLI reviewer is still delegation, not an escape hatch.

Choose the Shape by Dependency

Work shapeCoordination pattern
Independent research questionsParallel investigators, evidence-based synthesis
Disjoint features, stable interfacesParallel builders, designated integration owner
One change repeated across modulesScoped transform or partitioned workers, combined checks
Dependent edits or shared mutable stateOne owner for the dependency, parallel work around it
Consequential reviewIndependent read-only reviewers with distinct risk lenses
Small, tightly coupled changeDirect execution

Fan out by default: anything that splits across independent surfaces runs in parallel, and the coordinator keeps working while workers run. The limits are resources, not agent count: one owner per shared mutable surface, one committer per index, and a sequential chain stays sequential. Check the critical path, since a branch waiting on one shared interface gains nothing from more agents. Do not cap a fleet to look frugal or serialize work to look careful; name the contended resource or don't restrict. A worker that double-checks a small change rarely pays; the verification triggers below say when it does.

Establish Ownership and Integration

Map shared resources as well as files: generated artifacts, schemas, lockfiles, fixtures, ports, build outputs, and caches collide across otherwise disjoint modules.

IsolationAppropriate useRequired coordination
Shared tree, disjoint pathsSmall changes with clear ownershipOne owner for contended files and the Git index
Separate worktreesIndependent branches or build stateKnown base, branch, merge order, and integration owner
Read-only revisionReview or researchFixed revision or diff plus permitted scratch outputs

Worktrees isolate tracked files and the index, not ports or databases. Never overwrite a human's or sibling's edits to clear a conflict. In a shared tree, workers report patches while one coordinator stages and commits. Investigate a lock's actual owner before removing it; an inconclusive process listing does not establish that a lock is safe to remove.

Before dispatch, define how partial progress becomes integrated progress: the artifact each consumer expects and when it is ready.

Write Briefs

Give each worker the original request plus facts absent from their checkout, such as user corrections and untracked decisions.

Brief elementPurpose
Outcome and scopeResult needed, owned surfaces, exclusions
Current evidenceCode, versions, interfaces, verified premises
DependenciesWhat is ready, what is promised, who owns it
AuthorityPermitted mutations, commit rights, external actions
VerificationChecks and acceptance conditions
ReturnChanged artifacts or findings, evidence, limits, justified deviations, calls made where the brief was silent

Let workers challenge a premise. A first independent review gets the raw change without coaching; a follow-up verification gets the finding and the fix claim so it can test closure. Templates are in references/dispatch-briefs.md.

Supervise

Harvest dependency-ready results as they arrive. Silence alone does not prove an agent is stuck; check its artifacts before interrupting. Polling must respect the host's limits. When requirements change, tell the affected workers.

A watcher needs a completion condition, failure detection, an expiry, and a stale-state check, with its signal tested on a known state. Cancel only processes this run owns. A watch loop checks for new user messages between polls and yields to them.

Harvest, Integrate, Verify

A worker's summary is a claim; inspect the actual diffs. Integrate at a stable checkpoint and run checks covering the combined behavior.

Independent verification is required when consequence or run length justifies it (shared or production state, an unattended run past a checkpoint, an acceptance check the implementer cannot alter) or when the project contract requires it. The implementer cannot certify an independent PASS. Tie each verdict to a revision or recorded snapshot.

VerdictMeaning
PASSDeclared scope completed, no unresolved blocking findings
FAILA supported defect violates the requested outcome or contract
INCOMPLETEReview or a required check did not finish; list uncovered scope

An interrupted review cannot grant a full PASS. Agreement does not establish severity; impact does. Do not reuse a PASS for changed runtime code.

Finish With an Honest State

Report integrated outcomes, the checks run and their revision, unresolved limits, and any blocked gate with the action it needs. Capture durable coordination failures in project memory: the contended resource, mechanism, fix, and conditions.

Evidence

Reviewed 2026-09-04 and 2026-10-27; sources and their limits are in references/evidence.md.

Anti-Patterns

Anti-patternBetter move
Choose a fleet size before finding independent workName useful lanes and their integration path
Serialize parallelizable work to look carefulFan out; limit by shared resources, not agent count
Assume different files imply isolationInspect shared runtime resources and generated outputs
Let all shared-tree workers commit concurrentlyAssign ownership of the index and commit operation
Vote-count findingsVerify mechanism and impact
Reduce review automatically over timeMatch checks to current risk and changed evidence

References

Read references/dispatch-briefs.md when composing worker, research, verification, or watcher instructions.

What This Skill is NOT

  • Permission to ignore host delegation rules or user ownership.
  • A fixed pipeline, agent quota, or requirement to use multiple agents.
  • A substitute for integrated verification or the coordinator's judgment.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.