delegate-manager-to-subagents
Skill5 starsChanged 22 days agoMITView on GitHub
Manage execution of an approved plan through subagents. Use when the user asks the main agent to coordinate implementation, review, testing, or documentation while the plan remains the source of truth.
In this repository47 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agents-md-fixerSkill
- arch-docsSkill
- book-to-markdownSkill
- docset-updaterSkill
- humanizerSkill
- news-release-docsSkill
- pdf-guideSkill
- readme-docsSkill
- screenshot-docsSkill
- see-also-docsSkill
- setup-install-usage-docsSkill
- bptools-writer-expertSkill
- color-accessibility-expertSkill
- css-creative-expertSkill
- game-design-expertSkill
- geometry-expertSkill
- glass-expertSkill
- hic-human-interact-expertSkill
- openscad-3d-printing-expertSkill
- podman-expertSkill
- postgresql-expertSkill
- pyside6-engineerSkill
- rust-code-expertSkill
- solid-js-expertSkill
- svg-creator-expertSkill
- typescript-engineerSkill
- ui-ux-engineerSkill
- vision-expertSkill
- wasm-rust-expertSkill
- webwork-writer-expertSkill
- gas-town-workflowSkill
- hang-checkSkill
- html-game-parallel-builderSkill
- parallel-planSkill
- stay-busySkill
- repo-rules-readerSkill
- blueprint-plan-drafterSkill
- ideonomy-plainSkill
- ideonomy-richSkill
- make-goalSkill
- modest-plan-drafterSkill
- audit-code-reviewerSkill
- old-python-code-reviewSkill
- prune-testsSkill
- unit-test-starterSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
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 29
Every brief invites clarification, challenge, decision, and escalation questions.
- Distinctive: line 56
Use example-briefs.md for the exact template.
- Distinctive: line 63
Use the plan's verification contract and choose appropriate checks for the repository and task.
- Distinctive: line 68
After task reviews pass, obtain fresh integration review of composition, architecture, and complete plan coverage before accepting the plan.
≈ “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
Manage delegated execution
Plan leads
The approved plan defines task text, scope, dependencies, ownership, verification, and acceptance. Preserve those decisions. Use this skill for delegation choices the plan leaves open.
For code investigation and dependency-aware dispatch, read GRAPHIFY_GUIDE.md when a Graphify map or repo wrapper is available.
Manager role
- Track tasks, dependencies, decisions, reviews, and acceptance in a simple file-backed ledger that stays with the plan artifacts, separate from product code.
- Dispatch all file changes to subagents. The manager coordinates, decides, routes review, and accepts work.
- Give every assignment to a fresh agent, including fixes, reviews, re-reviews, and integration.
- Select and state a smaller, cheaper delegate model than the manager by default. Use the smallest capable live-catalog model; record the capability reason when a larger model is necessary or no cheaper suitable model is available. Use a fresh replacement agent for capability escalation.
Active questioning
Every brief invites clarification, challenge, decision, and escalation questions. Subagents actively challenge unsupported assumptions, including the manager's, and ask whenever uncertainty could affect requirements, scope, ownership, shared interfaces, verification, or acceptance.
For unclear work or a shared boundary, require an early checkpoint: interpretation, intended approach, affected boundaries, and unresolved questions. A question states the ambiguity, options and consequences when applicable, grounded recommendation, and exact blocking point.
The manager responds with a decision, evidence, or bounded investigation; it probes unclear answers and escalates missing product intent or authority to the human. Continue until the ambiguity is resolved or escalated. While pending, agents continue only isolated, non-conflicting work and mark dependent work pending. After receiving a decision, restate the resulting constraint before dependent implementation. A question closes after every affected agent confirms its interpretation, applies the decision, and provides relevant verification evidence.
Dispatch brief
Supply the complete approved task text once in this seven-part brief:
- Assignment and outcome
- Ownership and boundaries
- Context and constraints
- Dependencies and coordination
- Approach and question checkpoint
- Verification and handoff evidence
- Review and completion route
Use example-briefs.md for the exact template. Read manager_contract.md for questions, decisions, ledger fields, and closure.
Evidence and review
- Require exact commands and decisive output, or the exit status for a silent command; map deliverables to requirements; scope failures, warnings, skipped checks, and remaining limitations.
- Use the plan's verification contract and choose appropriate checks for the repository and task.
- Route fresh specification review first. After it accepts the work, route a different fresh quality reviewer. Give fixes and every re-review fresh agents.
- Let reviewers use a change-focused or codebase-wide scope as needed. Findings identify supporting file/line or command evidence and the limits of the conclusion.
- After task reviews pass, obtain fresh integration review of composition, architecture, and complete plan coverage before accepting the plan.
Parallel work
Dispatch ready work concurrently only when ownership is isolated and no in-flight dependency remains. Record why work is serialized. Resolve shared contracts through a narrow specialist decision before dependent work starts. Use parallel-dispatch-examples.md for coordination and pressure scenarios, and role-catalog.md for role and model selection.
Completion
Complete a task after its plan outcome, verification evidence, decision closure, and required reviews pass. Complete the plan after integration review accepts the combined result and the ledger records completion evidence and residual risks.
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.

