agentleFS
Codagent-AICodagent-AI/agent-skills/skills/orchestrate-change/SKILL.md

orchestrate-change

Skill31 starsChanged 6 months agoMITView on GitHub

Implements a well-specified change (approved change artifacts, specs, or similar) as an orchestrator that delegates all code changes to sub-agents. Use by default when implementing any well-specified change, unless instructed otherwise.

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    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:

    If you have no way to spawn sub-agents, say so and stop.

    line 16
  2. Distinctive:

    Treat the approved change artifacts, specs, or similar as the source of truth.

    line 20
  3. Distinctive:

    If running autonomously, do not ask the user anything: resolve gaps with a ruling consistent with the artifacts and record it.

    line 23
  4. Distinctive:

    Plan: Read the artifacts and split the change into slices.

    line 28
  5. Distinctive:

    Verify: Do not accept "done" without evidence: require the sub-agent to report the automated tests it added and the test commands it ran …

    line 39
  6. Distinctive:

    Integrate: When all slices are done, have one sub-agent run every suite identified in planning and the checks and reviews the project …

    line 43

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

Orchestrate Change

You are the orchestrator for this change. Sub-agents write the code; you own the outcome.

Do not edit project files yourself, not even a one-line fix. Delegate every change to a sub-agent. Independent review only works if you never write the code you are judging. You may read files and run read-only and verification commands.

If you have no way to spawn sub-agents, say so and stop.

Inputs

Treat the approved change artifacts, specs, or similar as the source of truth. Treat any task files as hints, not the plan.

If running autonomously, do not ask the user anything: resolve gaps with a ruling consistent with the artifacts and record it.

Process

  1. Plan. Read the artifacts and split the change into slices. Every requirement and planned test must belong to a slice. Identify every automated suite, including ones outside the default command, and which already fail on the base commit. Keep the plan, your rulings, and progress in a file outside the repository so they survive context compaction.
  2. Delegate. Sub-agents do not see your conversation, so give each a self-contained brief: the goal, the artifact paths and the exact requirements it covers, relevant design decisions and earlier rulings, its boundaries, and how to report back. Tell each sub-agent to commit its own work. Slices that run concurrently must not share a working tree or other mutable resource (files, ports, databases, browsers); otherwise run them one at a time. If a slice runs in its own worktree or branch, name the commit it must start from in its brief, and tell the sub-agent to confirm its HEAD matches that commit before starting.
  3. Verify. Do not accept "done" without evidence: require the sub-agent to report the automated tests it added and the test commands it ran with their results, and confirm its work is committed. Rely on that evidence instead of rerunning the tests yourself. Send fixes back to a sub-agent.
  4. Integrate. When all slices are done, have one sub-agent run every suite identified in planning and the checks and reviews the project defines, once, and confirm every requirement and planned automated test is covered. Turn gaps into new slices.

Limit verification here to automated checks. Leave code review beyond the project's own checks, exploratory and acceptance testing, and archiving for after this skill.

Change the approved artifacts only if implementation proves one wrong, and record why.

Report

Summarize the slices and their commits, how the change was verified, and every ruling, assumption, and unresolved gap from you and your sub-agents. Leave the worktree clean.

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.