delegation

Delegate sub-tasks to agents, specialist models, or background jobs.

33|1|Updated Mar 4, 2026
One-click install
npx skills add https://github.com/infinition/LaRuche --skill delegation-infinition
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: delegation
Source: https://github.com/infinition/LaRuche/tree/main/laruche/skills/delegation
Command: npx skills add https://github.com/infinition/LaRuche --skill delegation-infinition

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Complex tasks often exceed what a single agent pass can handle well, and choosing the wrong offloading method wastes model calls or loses results. This Skill guides an agent in picking the right delegation mechanism and briefing it correctly so the work comes back verifiable. ## Core Features & Use Cases - Four delegation modes: delegate for self-contained sub-tasks, spawn_specialist for tasks needing a different model or role, mixture_of_agents for parallel merged answers, and submit_job for long-running background work. - Cold-start briefing rules: Enforces complete task briefs with absolute paths, explicit acceptance criteria, and constraints, since delegates cannot see the parent conversation. - Result verification discipline: Requires polling check_job_status before reporting background results and verifying delegate claims before trusting them. - Use Case: An agent facing a large codebase audit delegates the security review to a specialist model with a full brief, continues its own analysis, then polls the job and reads the produced files before summarizing findings to the user. ## Quick Start Ask the agent to delegate the self-contained sub-task to a specialist model with a complete brief including absolute paths and acceptance criteria.

Frequently Asked Questions about delegation

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

FAQPage Schema
How do I delegate a sub-task to another AI agent?▼

Use the delegate tool with a task parameter written as a complete brief a stranger could execute. Include absolute paths, the exact goal, the acceptance criterion, and what the delegate must not touch, because it cannot see your conversation.

When should I use mixture_of_agents instead of a single model call?▼

Use mixture_of_agents for open judgements, design trade-offs, or summaries where a single pass is unreliable. Avoid it for deterministic work like running commands or reading files, where voting among models wastes cost and can produce a merged answer no model actually gave.

How do I run a long task in the background with an AI agent?▼

Call submit_job with the script and a readable label, which returns a job id immediately. Continue other work, then poll with check_job_status and read the output before reporting anything, since a job id alone is not a result.

Why does my delegated agent say it cannot find a file that exists?▼

The delegate does not share your working directory, so relative paths fail. Pass absolute paths in the task or context fields and retry with the corrected brief.

What happens when I cancel a background job?▼

Cancelling stops the job from continuing but does not undo work already done, such as files written or API calls made. After cancelling, inspect the actual state before reporting, and never claim nothing happened without verifying.

When should I not delegate a task at all?▼

Do not delegate anything you can finish in two or three tool calls yourself, since delegation costs a full context build. Also avoid delegating ambiguous tasks or the entire user request, because you lose the ability to verify the result.