offload-subagent

Delegates implementation tasks to subagents while the main session handles design and review.

2|Updated Sep 25, 2011
One-click install
npx skills add https://github.com/pokutuna/dotfiles --skill offload-subagent-pokutuna
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: offload-subagent
Source: https://github.com/pokutuna/dotfiles/tree/main/claude/skills/offload-subagent
Command: npx skills add https://github.com/pokutuna/dotfiles --skill offload-subagent-pokutuna

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Long implementation sessions consume large amounts of context tokens, leaving little room for design reasoning and careful review. This Skill switches the session into an orchestration mode where implementation work is offloaded to subagents, preserving the main session's context for design, auditing, and review. ## Core Features & Use Cases - Task Decomposition: Breaks work into independent implementation units that can run in parallel via the Agent tool. - Model Selection Guidance: Uses sonnet for routine implementation and considers opus for complex investigation or large dependency analysis. - Mandatory Review Loop: Requires the main session to read and verify every subagent deliverable against the design rather than trusting reported success. - Use Case: During a large refactoring, you invoke this Skill so the main session plans the architecture, dispatches file-by-file changes to subagents in parallel, and then reviews each diff for consistency and quality. ## Quick Start Invoke the offload-subagent skill with a description of your task to have it decomposed and executed by subagents under this session's supervision.

Frequently Asked Questions about offload-subagent

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

FAQPage Schema
How do I delegate coding tasks to subagents in Claude Code?

Invoke this Skill with a task description, and the session decomposes the work into independent units dispatched through the Agent tool. Parallel-safe units are launched together in a single message, and each result is reviewed before acceptance.

When should I use sonnet versus opus for subagent tasks?

Use sonnet for straightforward implementation and mechanical changes, which covers most delegated work. Reserve opus for tasks involving complex investigation, design judgment, or large dependency analysis that a simpler model would handle poorly.

Does the main session still write code after offloading to subagents?

Only in limited cases. The main session may directly implement parts that are especially difficult or involve inseparable design decisions; otherwise it focuses on planning, instruction writing, and reviewing subagent output.

What happens if a subagent produces incorrect code?

The main session must read every subagent deliverable and check the diff against the design intent. It never trusts a subagent's claim of success without verifying consistency and quality itself.

What are the limitations of subagent-based implementation?

Subagents need explicit goals, background, file paths, and completion criteria; vague instructions produce poor results. Highly coupled changes that resist decomposition are better handled directly in the main session.