agentleFS
EnjoyoerEnjoyoer/loadout/skills/orchestration/opc/SKILL.md

opc

Skill0 starsChanged 25 days agoMITView on GitHub

Coding-focused multi-agent orchestration for exploration, implementation, review, and verified delivery.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    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
  1. Distinctive:

    Follow this state machine, reading linked references only when their branch applies.

    line 14
  2. Distinctive:

    All repository source and test edits, including small repairs, belong to Workers.

    line 38
  3. Distinctive:

    Verify returned agent/workspace identity, path, branch, commits, selected-base ancestry, and exact diff.

    line 46

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.

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

OPC

Each OPC turn, read version: first; if changed, reread this file and needed references.

The PM owns scope, topology, integration, verification, and publication. OPC implementation authorizes bounded Worker lanes, never subdelegation or route substitution. OPC ends with verified merge, required sync, and cleanup; earlier endpoints are ordinary scoped work. Local commits and PRs are internal stages.

Follow this state machine, reading linked references only when their branch applies.

1. Scope

PMs run on Pi fleet/claude-opus-5-5, thinking medium, unless the owner names another model or effort. If self differs at first PM ownership, update_agent only self's model and thinking.

Whenever you own a project with no custom icon that is not a scratch folder, use $project-icon.

Record outcome, assigned checkout, owner, committed base, permitted changes, and acceptance checks: observable behaviors from the owner's intent, preferably end-to-end or live. Consult root STATUS.html and LESSONS.md when resuming. Escalate scope or authority changes to the owner.

Enable browser review only when the initial request explicitly asks; the choice is immutable through repairs, resumes, and reinvocation. Initialize the execution mechanics before launching lanes.

2. Optional scouting

Choose minimal useful scouting topology, splitting only independent questions. Trivial, localized, or inherently sequential work skips scouting and planning. For useful independent questions, read exploration. Synthesize findings first. Only material residual ambiguity or unresolved cross-lane decisions justify the planner.

3. Worker

Resolve every route via routing; owner model, effort, and Fast win unadjusted.

Implement one feature (one user-visible outcome whose spec, diff and checks fit one context) in one implementer lane: its server, UI, docs and tests together. Never split it across parallel workers by type of work (frontend, backend, tests) or by file. Use parallel lanes only for separable outcomes, or for work too large for one context (a port or rewrite) (provisional), cut by layer or crate with one owner per file; the PM writes any shared contract (flags, schemas, shared docs) before those lanes start. Use fresh agents only where a fresh context helps: a reviewer that gets only the spec and the diff, a verifier that gets only the user flows to check, and research questions that need no project context. One Worker lineage per lane; shared live resources serialize only live checks.

Visual design (new screen, layout, component) gets one read-only Fable high ui-design lane and optional ui-check; code builds it. Copy-only UI changes stay code.

All repository source and test edits, including small repairs, belong to Workers. The PM stays in its checkout as sole integrator and publisher. For hosted Sites work only, read Sites. For cancellation or uncertain execution, read recovery before acting.

4. Yield

Launch every OPC agent and background follow-up through agent-scoped Paseo create_agent or send_agent_prompt MCP with notifyOnFinish: true; finish independent work, then end the PM turn. On completion, error, or permission notification, read get_agent_status and get_agent_activity once each. No polling, babysitting, sleeps, CLI waits, terminal inspection, status narration, retries, or replacements while running. Report terminal failures.

5. Integrate

Verify returned agent/workspace identity, path, branch, commits, selected-base ancestry, and exact diff. Integrate by repository merge/cherry-pick policy, never merely on completion/idle status. Send repairs to the same Worker; repeat yield and integration.

6. Verify

Independently check required behavior and final committed deliverable bytes using the execution mechanics. Reuse a receipt only while head, bytes, command, and evidence remain unchanged; otherwise rerun. Review rules bind every review; initially enabled review reads it at the published candidate head. Security-sensitive diffs review on review-security. Review never replaces PM correctness checks. Platform or browser checks not run remain unverified.

7. Merge and cleanup

PRs with visible changes carry before and after evidence; other PRs say "No visible change". Use the execution mechanics to verify delivery and merge, then follow workspace cleanup. Confirm live GitHub merge state and reviewed-head ancestry. Follow repository sync requirements, distinguishing source publication from installation. Report test, integration, and PR evidence, review verdict/head or unavailability, planner outcome and Worker usage, remaining issues, and cleanup gaps separately.

For package maintenance only, use maintenance.

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.