do-delegation

Orchestrates parallel agent dispatch, background execution, and worktree-isolated edits with coordinator verification.

2|Updated May 25, 2026
One-click install
npx skills add https://github.com/edheltzel/Do-Skills --skill do-delegation-edheltzel
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: do-delegation
Source: https://github.com/edheltzel/Do-Skills/tree/main/skills/operations/do-delegation
Command: npx skills add https://github.com/edheltzel/Do-Skills --skill do-delegation-edheltzel

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Deciding when and how to delegate work to subagents is error-prone: over-delegation wastes effort, under-specified prompts produce unusable results, and parallel writers collide in a shared checkout. This Skill provides a decision gate and dispatch patterns for routing independent work through the Agent tool in Claude Code or Pi. ## Core Features & Use Cases - Delegation Gate: A six-step check (direct-work, independence, context, write-conflict, verification, fan-out) that decides between direct work and N agents before any dispatch. - Dispatch Patterns: Foreground agents, parallel background agents, worktree-isolated editing agents, custom role briefs, and coordinator-managed role cohorts for multi-specialist deliverables. - Verification Contract: Treats agent completion messages as claims, requiring diff inspection, test runs, and evidence checks before synthesis. - Use Case: When implementing a cross-layer feature, dispatch API, UI, and test work to separate worktree-isolated agents after interfaces are defined, then verify each diff and synthesize the results. ## Quick Start Ask the agent to evaluate whether the current task should be handled directly or fanned out to parallel background agents, then dispatch and verify the results.

Frequently Asked Questions about do-delegation

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

FAQPage Schema
How do I run parallel agents in Claude Code?▼

Issue one Agent tool call per independent unit in the same assistant message with run_in_background set to true. Trust completion notifications, continue non-dependent work, and verify each result before synthesis.

When should I delegate a task to a subagent?▼

Delegate only when specialization, context isolation, or parallel execution saves more time than dispatch and verification cost. If a few direct reads can settle the question, use zero agents.

How do I prevent parallel agents from conflicting on file edits?▼

Use worktree isolation when the Agent schema supports it, giving each worker a non-overlapping objective. If isolation is unavailable, serialize editing work so writers never share one checkout.

Do subagents share context or memory with the parent conversation?▼

No. Independent agents do not share context, memory, task state, or results. Every agent prompt must be complete, and follow-up work requires a fresh agent with prior results embedded in its prompt.

Why should agent completion messages not be trusted directly?▼

A completion message is a claim, not proof. For writing agents, confirm files exist, inspect the actual diff, and run relevant checks; for research agents, verify scope coverage and cited evidence.