agent-delegation
Rules for when and how to spawn sub-agents: Explore for 3+ file reads, parallel agents for independent tasks, summaries only — raw output never enters main context
Install
npx skills add https://github.com/yeisonrestrepo/code-conductor/tree/main/skills/agent-delegation
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install yeisonrestrepo-code-conductor@llmmart
git clone https://github.com/yeisonrestrepo/code-conductor.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole yeisonrestrepo/code-conductor collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Agent Delegation
Keep the main context clean. Raw file contents, grep output, and intermediate data never enter the main context. Sub-agents process data and return summaries only.
When to Spawn
Explore Sub-Agent
Condition: Task requires reading 3+ files to answer one question, or codebase exploration precedes implementation.
// Use the Agent tool with these parameters:
Agent({
subagent_type: "Explore",
description: "<10-word description of the lookup>",
prompt: "<question>. Search across relevant files. Return a summary of ≤200 words. Do not include raw file contents."
})
Parallel Agents
Condition: 2+ implementation tasks with no shared state, each describable independently.
Send both Agent calls in a single message so they run in parallel:
// Use the Agent tool twice in the same message — they run in parallel:
Agent({ description: "Task A", prompt: "... Return a summary of ≤200 words." })
Agent({ description: "Task B", prompt: "... Return a summary of ≤200 words." })
Output Contract
- Every sub-agent returns ≤200 words to the main context.
- The main context acts only on summaries, not raw data.
- If a sub-agent needs user input, it surfaces one question to the main context and stops.
When NOT to Spawn
- Single Grep or Glob can answer the question.
- Answer is already in
.claude/memory/project.md. - Task requires back-and-forth with the user (stay inline).
Files (code-conductor)
-
SKILL.md 1.7 KB
--- name: agent-delegation description: "Rules for when and how to spawn sub-agents: Explore for 3+ file reads, parallel agents for independent tasks, summaries only — raw output never enters main context" type: skill --- # Agent Delegation Keep the main context clean. Raw file contents, grep output, and intermediate data never enter the main context. Sub-agents process data and return summaries only. ## When to Spawn ### Explore Sub-Agent **Condition:** Task requires reading 3+ files to answer one question, or codebase exploration precedes implementation. ```javascript // Use the Agent tool with these parameters: Agent({ subagent_type: "Explore", description: "<10-word description of the lookup>", prompt: "<question>. Search across relevant files. Return a summary of ≤200 words. Do not include raw file contents." }) ``` ### Parallel Agents **Condition:** 2+ implementation tasks with no shared state, each describable independently. Send both Agent calls in a single message so they run in parallel: ```javascript // Use the Agent tool twice in the same message — they run in parallel: Agent({ description: "Task A", prompt: "... Return a summary of ≤200 words." }) Agent({ description: "Task B", prompt: "... Return a summary of ≤200 words." }) ``` ## Output Contract - Every sub-agent returns ≤200 words to the main context. - The main context acts only on summaries, not raw data. - If a sub-agent needs user input, it surfaces one question to the main context and stops. ## When NOT to Spawn - Single Grep or Glob can answer the question. - Answer is already in `.claude/memory/project.md`. - Task requires back-and-forth with the user (stay inline).
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.