agentleFS
darrenCWJdarrenCWJ/agent-skills-scaffold/.claude/skills/fan-out/SKILL.md

fan-out

Skill0 starsChanged 3 months agoMITView on GitHub

Delegate work to parallel subagents correctly — when to fan out, the briefing contract, and merge protocol. Use whenever a plan contains independent subtasks, broad multi-source searches, parallel verification, or the user says in parallel, agents, or delegate. Also read it to decide NOT to fan out.

9 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

5 rules. 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 5

    Distinctive

    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:

    When you'd spend more effort briefing than doing.

    line 16
  2. Distinctive:

    Every spawned agent gets ALL of:

    line 21
  3. Distinctive:

    Keep a fan-out ledger in the task list: one task per agent, marked with its scope.

    line 35
  4. Distinctive:

    Verify each result against its brief BEFORE integrating — spot-check claims, run its outputs, read its diffs.

    line 41
  5. Distinctive:

    Look for cross-agent conflicts: contradictory findings, incompatible edits, duplicated work.

    line 42

“Common” is measured: kept by at least 6 in 10 of the 7 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

Agent fan-out

When to fan out

  • Subtasks are independent (no shared files, no ordering dependency) AND each needs meaningful exploration or generation.
  • Broad research/search across many sources or files where you only need conclusions.
  • Independent verification: a fresh critic agent reviewing finished work.

When NOT to fan out

  • Single lookups you can do in 1–2 tool calls yourself.
  • Tightly coupled edits (agents stepping on each other's files).
  • When you'd spend more effort briefing than doing.
  • Deep sequential reasoning — parallelism doesn't help a chain.

Briefing an agent (the contract)

Every spawned agent gets ALL of:

  1. Scope — exactly what to do, with file paths / sources / boundaries.
  2. Deliverable — exact output format ("return a markdown list of findings, each with file:line evidence").
  3. Non-goals — what NOT to touch or do. Explicitly forbid writing outside its assigned paths.
  4. Context — the minimum background it needs (it starts with zero context; it cannot see this conversation).
  5. Quality bar — how it should verify its own work before reporting.

A brief the agent could misinterpret WILL be misinterpreted. Write briefs like API contracts.

Orchestration rules

  • Launch independent agents in parallel — one message, multiple Task calls.
  • Never let two agents write to the same file. Partition by directory or file list.
  • Writers get disjoint scopes; readers can overlap freely.
  • Keep a fan-out ledger in the task list: one task per agent, marked with its scope.
  • Cap parallelism at what you can meaningfully verify (typically 3–5 agents per wave).
  • For continuing work with an existing agent, message it rather than spawning a fresh one and re-briefing.

Merging results (fan-in)

  1. Verify each result against its brief BEFORE integrating — spot-check claims, run its outputs, read its diffs. Blind merging is a protocol violation.
  2. Look for cross-agent conflicts: contradictory findings, incompatible edits, duplicated work.
  3. Integrate sequentially, verifying after each merge, not after all merges.
  4. Record in the plan which results were accepted, rejected, or reworked — and why (design decision if significant).

Failure handling

An agent that returns garbage gets ONE re-brief with the misunderstanding corrected. If it fails again, do that subtask yourself — repeated re-delegation wastes more than it saves.

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.