delegated-execution

Coordinates subagent delegation with briefs, model policies, review, and recovery protocols.

2|Updated Sep 14, 2026
One-click install
npx skills add https://github.com/kwanpham2195/agent-stuff --skill delegated-execution-kwanpham2195
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: delegated-execution
Source: https://github.com/kwanpham2195/agent-stuff/tree/main/skills/delegated-execution
Command: npx skills add https://github.com/kwanpham2195/agent-stuff --skill delegated-execution-kwanpham2195

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Delegating work to subagents often fails because workers receive vague briefs, wrong models, unclear acceptance criteria, or unverified results. This Skill defines a complete contract for dispatching, monitoring, reviewing, and recovering delegated work so outcomes stay accountable. ## Core Features & Use Cases - Structured briefs and reports: Dispatch workers with exact scope, acceptance criteria, verification commands, and size ceilings, and require evidence-based status reports in return. - Role-based delegation: Choose between worker, reviewer, and advisor roles, with a dedicated advisor dialogue protocol for consequential decisions and a model policy mapping roles to harness-specific models and effort levels. - Bounded exploration and recovery: Run up to three read-only explorers concurrently, verify diffs against acceptance criteria before accepting work, and follow checkpoint and resume protocols when corrections or timeouts occur. - Use Case: When implementing a multi-file feature, dispatch a worker with a bounded brief and required skill paths, then inspect the actual diff and verification output before accepting the result. ## Quick Start Delegate the bounded task of fixing the filter reset bug to a worker with exact paths, acceptance criteria, and verification commands, then review its evidence before accepting.

Frequently Asked Questions about delegated-execution

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I delegate a coding task to a subagent safely?▼

Give the worker one bounded task with exact paths, acceptance criteria, verification commands, out-of-scope items, and a numeric size ceiling. Resolve required skill paths before dispatch and include their absolute paths in the brief, since the child does not inherit your context.

When should I use an advisor subagent instead of a worker?▼

Use an advisor at a decision boundary when several plausible approaches have consequential tradeoffs or new evidence undermines an accepted plan. Advisors are read-only thinking partners; factual lookups belong to research and checking completed work belongs to review.

How many subagents can run concurrently during delegation?▼

At most three concurrent children are allowed. Parallel writers need isolated worktrees with one writer per checkout, parallel reviews need disjoint surfaces, and children must not delegate further.

Which model should a delegated worker or reviewer use?▼

The model policy maps roles to harness-specific defaults, such as Opus for implementation and review on Claude Code, or Terra and Sol on Pi and Codex. Explicit user choices take precedence, and unavailable models require stopping to ask before substituting.

What should I check before accepting delegated work?▼

Inspect the actual diff and verification output, compare results against each acceptance criterion, and check changed assertions, suppressions, and call sites. A worker's completion report alone is not acceptance; pending or unsupported criteria block it.

What tasks should not be delegated to subagents?▼

Do not delegate intake, quick fixes, or small factual gaps; handle small tasks inline. Delegation is justified only when independent work or context isolation earns the coordination cost.