dispatch-criteria
Skill2 starsChanged 5 months agoMITView on GitHub
Six gates for deciding whether to delegate to a subagent or do it in-session. From ChristopherKahler/paul. The "is this work parallelizable" decision aid.
In this repository60 other files it gives its agents.
- .claude/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- analyzeSkill
- animation-disciplineSkill
- api-versioningSkill
- autopilotSkill
- browser-e2eSkill
- characterizeSkill
- clarifySkill
- codemodSkill
- constitution-compactSkill
- constitutionSkill
- context-budgetSkill
- creative-webSkill
- db-migrationSkill
- debugSkill
- detect-stackSkill
- dreamSkill
- experimentSkill
- fastapiSkill
- flag-cleanupSkill
- flag-rolloutSkill
- flask-realtimeSkill
- framework-detectorSkill
- front-end-designSkill
- grill-meSkill
- handoffSkill
- image-generationSkill
- implementSkill
- initiativeSkill
- instinctSkill
- loopSkill
- nextjsSkill
- observabilitySkill
- okr-alignSkill
- parallel-swarmSkill
- pivotSkill
- planSkill
- pr-bodySkill
- prime-disciplineSkill
- prod-observabilitySkill
- react-nodeSkill
- react-viteSkill
- release-trainSkill
- researchSkill
- reviewSkill
- security-guardSkill
- self-healSkill
- shipSkill
- skill-creatorSkill
- specifySkill
- tailwind-disciplineSkill
- tasksSkill
- tdd-loopSkill
- token-budgetSkill
- typographySkill
- verify-loopSkill
- verifySkill
- wip-checkpointSkill
- <kebab-case-skill-name>Skill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
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
- could be automatic: a setting could enforce it
- Distinctive: line 17
If yes to either → don't delegate (sequential in main session).
- Distinctive: line 22
If no → sharpen the brief first, then decide.
- Distinctive: line 27
If both no → don't delegate (overhead exceeds value).
- Distinctive: line 37
If both yes → delegate (this is the strongest signal).
- Distinctive: line 42
If half-edits → don't delegate (no merge mechanism).
- Very specific: line 54
Default to in-session for code edits, delegate for read-heavy investigation/research/review.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Dispatch Criteria — Six Gates
Subagents are not free. Spawning one costs startup tokens, adds latency, and burns a context window. Use this checklist before every Task/Agent invocation.
The six gates
- Independence
- Does the subtask share mutable state with the main task?
- Does it need to wait for in-flight edits to settle?
- If yes to either → don't delegate (sequential in main session).
- Clear scope
- Can you write the brief in one paragraph?
- Does the subagent know exactly which files / endpoints / questions are in scope?
- If no → sharpen the brief first, then decide.
- Parallel value
- Will running it in parallel save wall-clock time (>30s)?
- Will it save tokens by isolating heavy reads?
- If both no → don't delegate (overhead exceeds value).
- Complexity sweet spot
- Too simple (a regex or one
grepwould do) → don't delegate. - Too complex (would need its own multi-day session) → don't delegate, decompose first.
- Sweet spot: 5-30 min of subagent work, single-purpose outcome.
- Too simple (a regex or one
- Token efficiency
- Would the work in main context cost >20K tokens of reads/searches/fetches?
- Is the output a clean summary that fits in <2K tokens?
- If both yes → delegate (this is the strongest signal).
- State compatibility
- Is the subagent's output a markdown summary, JSON, or list?
- Or does it need to leave in-flight, half-applied edits in the working tree?
- If half-edits → don't delegate (no merge mechanism).
Decision matrix
Count gates passed (out of 6):
- 6/6 → delegate
- 5/6 → delegate, but tighten the brief on the failing gate
- 4/6 → marginal — try if you're context-pressured, otherwise in-session
- ≤3/6 → don't delegate. Do it yourself, or decompose first.
When in doubt
Default to in-session for code edits, delegate for read-heavy investigation/research/review.
References
- ChristopherKahler/paul/src/references/subagent-criteria.md (source)
- obra/superpowers/skills/dispatching-parallel-agents (parallel = N independent investigations, never N parallel edits)
- obra/superpowers/skills/subagent-driven-development (single sequential subagent per task with two-stage review)
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.

