subagent-context-delegation

Package context capsules and delegate tasks to Kimi Code subagents with parallel explore patterns.

Updated Dec 24, 2025
One-click install
npx skills add https://github.com/JoyJoin-Tech-Limited/JoyJoin_app_v0.1 --skill subagent-context-delegation-joyjoin-tech-limited
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: subagent-context-delegation
Source: https://github.com/JoyJoin-Tech-Limited/JoyJoin_app_v0.1/tree/main/.github/skills/subagent-context-delegation
Command: npx skills add https://github.com/JoyJoin-Tech-Limited/JoyJoin_app_v0.1 --skill subagent-context-delegation-joyjoin-tech-limited

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Subagents spawned via Agent tool calls do not inherit parent context, so they often fail from missing background or repeat work the parent already did, while parent sessions bloat with tool results. This Skill provides templates and rules for packaging context, delegating work, and keeping the parent session lean. ## Core Features & Use Cases - Context Capsule Templates: Six copy-paste prompt templates covering bug investigation, feature implementation, architecture planning, parallel explore swarms, resume continuation, and code review delegation. - Parallel Explore Swarms: Launch 2-4 explore agents concurrently with non-overlapping, self-contained scopes for cross-cutting research. - Resume and Parent Hygiene: Rules for resuming existing subagents instead of respawning, plus discipline for summarizing subagent output before it re-enters parent context. - Use Case: When investigating a bug spanning matching logic, venue assignment, and feedback data, launch three parallel explore agents with focused capsules, then hand a consolidated capsule to a coder agent for the fix. ## Quick Start Ask the AI to launch a subagent for your task using a context capsule that summarizes prior decisions, relevant file paths, and the specific instruction.

Frequently Asked Questions about subagent-context-delegation

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

FAQPage Schema
How do I give a subagent enough context when it starts fresh?

Subagents do not inherit parent context, so include a context capsule at the start of the prompt. Cover what the parent already knows, decisions already made, and exact file paths involved, keeping the capsule under 400 words.

How do I run multiple research agents in parallel?

Launch 2-4 explore agents concurrently, each with a self-contained prompt and a distinct, non-overlapping scope. Set a 3-5 word description for log readability and use run_in_background only for truly independent tasks.

When should I resume a subagent instead of spawning a new one?

Resume a subagent when the new task is a natural continuation of its previous work, preserving accumulated context. If the topic has shifted significantly or its context has grown too large, extract findings and spawn a fresh agent with a new capsule.

What is the difference between explore, coder, and plan subagent types?

Explore handles read-only research with short focused scopes, coder handles implementation with full capsules and acceptance criteria, and plan handles architecture proposals with broader context. The default is coder when subagent_type is omitted.

Why does my subagent say it does not have enough context?

The capsule was incomplete because subagents never see your previous tool calls. Add prior findings, decisions already made, and exact file paths or module names to the prompt before delegating.

When should I not delegate to a subagent?

Skip delegation for single file reads or one-line fixes where overhead exceeds benefit, and when working entirely within the parent session. Delegation pays off once the parent accumulates more than about 10 tool-use blocks.