subagent-fanout
Skill98 starsChanged 2 months agoMITView on GitHub
Delegate breadth — multi-file searches, independent investigations, noisy command output — to subagents with precise briefs, then verify their claims before acting on them. Use when locating an answer means reading more than ~4 files, when several unrelated investigations could run in parallel, or when raw output would flood the context needed for synthesis and edits. Not for a single known file or symbol lookup.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- spence-grows/claude-skillstheir own version, first seen Oct 11, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- spence-grows/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository38 other files it gives its agents.
- ambiguity-commitSkill
- api-surface-careSkill
- calibrated-recommendationSkill
- codebase-orientationSkill
- concurrency-reasoningSkill
- config-and-flagsSkill
- context-checkpointSkill
- cross-platform-careSkill
- data-loss-guardSkill
- data-migration-safetySkill
- dependency-changesSkill
- docs-as-contractSkill
- edge-case-sweepSkill
- environment-firstSkill
- error-message-qualitySkill
- escalate-vs-decideSkill
- estimation-and-scopingSkill
- explain-at-the-right-levelSkill
- generalize-the-correctionSkill
- git-hygieneSkill
- incident-diagnosisSkill
- leave-no-messSkill
- llm-app-codeSkill
- numerical-careSkill
- performance-investigationSkill
- perf-sanitySkill
- question-vs-taskSkill
- resource-lifecycleSkill
- reviewing-a-changeSkill
- root-cause-debuggingSkill
- security-reflexesSkill
- stop-thrashingSkill
- surgical-refactoringSkill
- task-decompositionSkill
- test-designSkill
- untrusted-content-guardSkill
- verify-ui-visuallySkill
- workmanshipSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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
- Distinctive: line 18
The single lookup: If you know the file or symbol, one Grep/Read beats an agent round-trip.
- Distinctive: line 32
If you can't write the return format, you don't yet know what you're asking.
≈ “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.
Files
subagent-fanout
Weak delegation looks like: reading twelve files inline to answer one "where is X handled" question, or sending an agent "look into auth" and getting back an essay you can't act on. Strong delegation spends subagent context on search and keeps main context for synthesis and edits.
When to fan out
- Breadth, not depth: "where is auth handled", "find every caller of X", "which configs mention Y". If locating the answer would mean reading more than ~4 files, delegate.
- Independent investigations in parallel: three unrelated questions = three agents launched in one message, not a sequence.
- Noisy output: a test run, log grep, or build whose 2,000 lines you'd skim for three — have the agent run it and return the three.
When not to
- The single lookup. If you know the file or symbol, one Grep/Read beats an agent round-trip.
- Work that needs your context: judgment calls that depend on what the user said, or edits to code you've already reasoned about. Delegate finding; keep deciding.
- Already delegated. Once an agent is searching, don't run the same search inline while waiting.
The brief
A subagent has none of your context. Every brief states:
- Goal — the one question, phrased so it has a definite answer.
- Scope — directories or globs to search, and what to skip (generated code, vendored deps, tests unless asked).
- Constraints — read-only unless told otherwise: no edits, commits, or installs.
- Return format — e.g. "list of
file:line+ one-line role each, then a two-sentence conclusion; say 'not found' rather than guessing". Cap the length. - What you already know, so it doesn't rediscover it.
If you can't write the return format, you don't yet know what you're asking.
After it returns
- Verify at the point of use. Before editing based on an agent's claim about a signature, call site, or behavior, Read the cited lines yourself (see workmanship). Agents can invent paths and line numbers; a cited
file:linethat doesn't exist invalidates that finding, not the whole report. - Keep the conclusion, drop the transcript. Record what you carry forward in one or two lines; don't paste the agent's full output into your reasoning.
- Attribute in the report. A finding that came from a subagent and wasn't independently confirmed is labelled as such.
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.