subagent-supervision
Skill4 starsChanged 4 months agoMITView on GitHub
Supervise delegated work by giving subagents bounded, outcome-focused instructions and critically evaluating their reports before acting. Use whenever work is delegated to one or more subagents.
In this repository14 other files it gives its agents.
- AGENTS.mdAGENTS.md
- build-incrementalSkill
- code-simplifySkill
- create-or-clone-repositorySkill
- design-firstSkill
- implementation-workflowSkill
- manage-git-worktreesSkill
- plan-driven-developmentSkill
- setup-wiki-backed-project-skillsSkill
- task-briefSkill
- test-driven-debugSkill
- tidy-firstSkill
- write-personal-wikiSkill
- write-project-wikiSkill
The file
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
- Distinctive: line 12
Keep each assignment bounded and outcome-focused.
- Distinctive: line 20
Stop and blocker conditions: When to stop, ask for help, or report a blocker.
- Distinctive: line 41
Treat the report as evidence, not authority.
- Distinctive: line 42
Look for unsupported claims, missing evidence, scope deviations, unresolved assumptions, and inconsistencies.
- Distinctive: line 43
Use the report to target further checks.
- Distinctive: line 44
Use focused follow-ups when the report is incomplete, ambiguous, unsupported, or conflicting.
≈ “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 Supervision
Delegate bounded work to subagents while retaining responsibility for decisions and completion.
How To Instruct Subagents
Keep each assignment bounded and outcome-focused. Include only the items that affect the work:
- Goal: The concrete outcome to produce.
- Success criteria: What must be true for the assignment to be complete.
- Scope: What is included and excluded.
- Context and evidence: The information and artifacts needed to perform the work.
- Constraints and authority: Applicable rules, allowed actions, and approval boundaries.
- Expected output: The artifact to produce and the required report.
- Stop and blocker conditions: When to stop, ask for help, or report a blocker.
Describe the desired outcome without prescribing unnecessary steps. State each instruction once and omit unrelated context.
For independent evaluation, provide source artifacts and evaluation criteria without revealing the intended conclusion or another agent's findings.
Parallelize only independent assignments. Keep dependent work sequential.
How Subagents Should Report
Require a concise report containing:
- Outcome: What was completed, partially completed, or blocked.
- Evidence and artifacts: The sources, files, changes, or other evidence supporting the result.
- Verification: The checks performed and their results.
- Deviations and remaining issues: Scope changes, blockers, unresolved questions, or material risks.
Include the evidence needed to evaluate the result, but omit raw logs and unnecessary process narration.
How To Evaluate Subagent Results
- Treat the report as evidence, not authority. Compare it with the goal, scope, constraints, and success criteria.
- Look for unsupported claims, missing evidence, scope deviations, unresolved assumptions, and inconsistencies.
- Use the report to target further checks. Inspect artifacts or rerun verification only where a claim is material or uncertain; do not repeat the entire task or every check by default.
- Use focused follow-ups when the report is incomplete, ambiguous, unsupported, or conflicting.
- Main agent decides the next action and whether the work is complete.
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.

