subagent-workflow
Skill0 starsChanged 7 months agoMITView on GitHub
Use when Codex sub-agents should be applied in a disciplined way for analysis, implementation, or review. Trigger when a task is large enough to benefit from role-split work but still needs one main agent to own integration.
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- decision-recordingSkill
- harness-bootstrapSkill
- issue-driven-workflowSkill
- planning-workflowSkill
- review-workflowSkill
- spec-writingSkill
- technical-learning-noteSkill
- troubleshooting-workflowSkill
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
- Distinctive: line 8
Use sub-agents only when work naturally divides into roles and the main agent can still integrate the result safely.
- Distinctive: line 62
update docs, notes, playbooks, or run procedures
- Distinctive: line 66
Prefer the lightest execution mode that preserves quality: local first, then bounded delegation, then small team composition.
- Distinctive: line 68
Do not let two sub-agents edit the same files.
- Distinctive: line 72
Use sub-agents sparingly; speed matters only when coordination overhead stays low.
- Very specific: line 77
Use planning-workflow to decide the task breakdown first and review-workflow to define what a review should actually check.
≈ “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 Workflow
Use sub-agents only when work naturally divides into roles and the main agent can still integrate the result safely.
Stay local first until scope, affected paths, and validation are clear enough that delegation will reduce risk instead of creating guesswork.
Different files do not automatically make parallelism safe. Split work only when the lanes are behaviorally independent or joined by an agreed interface.
This skill is about staffing and ownership, not about replacing the spec, planning, issue, or review workflows themselves.
Good candidates
- analysis -> implementation -> independent review
- documentation research -> design proposal -> implementation
- large refactors with clearly separated file ownership
- implementation plus separate regression checks or docs updates
Avoid for
- one small fix
- overlapping write scopes
- unresolved product direction
- work that depends on continuous local reasoning across every step
Standard roles
Main agent
- owns the plan and routing
- decides whether delegation is justified
- assigns write ownership and review boundaries
- integrates results
- performs final verification
Analysis
- locate causes
- list affected files
- identify risks
- gather evidence for the main agent
Implementation
- change only approved files
- report exact edits
- report validation done
Review / verification
- stay read-only and independent
- look for regressions
- identify missing tests
- challenge completion claims with evidence
Writer / ops when relevant
- update docs, notes, playbooks, or run procedures
Guardrails
- Prefer the lightest execution mode that preserves quality: local first, then bounded delegation, then small team composition.
- The main agent decides the plan first.
- Do not let two sub-agents edit the same files.
- Different files do not automatically make parallel work safe; shared behavior and dependency chains still matter.
- Delegation does not widen authority or bypass protected paths.
- The main agent integrates and validates.
- Use sub-agents sparingly; speed matters only when coordination overhead stays low.
Resources
- See
references/subagent-patterns.mdfor recommended split patterns. - Use
planning-workflowto decide the task breakdown first andreview-workflowto define what a review should actually check.
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.

