์Šคํ‚ฌ agent-coordination
๐ŸŽฏ

agent-coordination

๋‚ฎ์€ ์œ„ํ—˜ โšก ์Šคํฌ๋ฆฝํŠธ ํฌํ•จโš™๏ธ ์™ธ๋ถ€ ๋ช…๋ น์–ด

Coordinate Agent Teams

Managing multiple coding agents requires clear task definitions and coordination patterns. This skill provides Chief of Staff workflows, two-tier delegation patterns, and VibeKanban integration to orchestrate autonomous agent teams efficiently.

์ง€์›: Claude Codex Code(CC)
๐Ÿฅ‰ 75 ๋ธŒ๋ก ์ฆˆ
1

์Šคํ‚ฌ ZIP ๋‹ค์šด๋กœ๋“œ

2

Claude์—์„œ ์—…๋กœ๋“œ

์„ค์ • โ†’ ๊ธฐ๋Šฅ โ†’ ์Šคํ‚ฌ โ†’ ์Šคํ‚ฌ ์—…๋กœ๋“œ๋กœ ์ด๋™

3

ํ† ๊ธ€์„ ์ผœ๊ณ  ์‚ฌ์šฉ ์‹œ์ž‘

ํ…Œ์ŠคํŠธํ•ด ๋ณด๊ธฐ

"agent-coordination" ์‚ฌ์šฉ ์ค‘์ž…๋‹ˆ๋‹ค. Act as my Chief of Staff. I need to add dark mode support to the app.

์˜ˆ์ƒ ๊ฒฐ๊ณผ:

  • โœ“ Chief of Staff mode activated. Ready for task delegation.
  • Creating investigation task: 'Investigate: Dark mode implementation scope'
  • Task #142 created with investigation tier. Agent exploring codebase...
  • Once investigation completes, I'll create implementation tasks using the detailed plan.

"agent-coordination" ์‚ฌ์šฉ ์ค‘์ž…๋‹ˆ๋‹ค. Create a task to fix the login bug on the login page

์˜ˆ์ƒ ๊ฒฐ๊ณผ:

  • I'll create a bug fix task with clear acceptance criteria.
  • Title: Bug: Login button unresponsive on Safari iOS
  • Context: Users report login failures on iOS Safari browsers
  • Acceptance: Login works on Safari iOS 15+, no console errors

๋ณด์•ˆ ๊ฐ์‚ฌ

๋‚ฎ์€ ์œ„ํ—˜
v5 โ€ข 1/16/2026

Documentation-focused skill for agent coordination. The only executable script (monitor-pr-ci.py) safely wraps gh CLI commands with hardcoded arguments. Static findings are false positives triggered by markdown documentation code blocks and YAML frontmatter. No actual security risks identified.

7
์Šค์บ”๋œ ํŒŒ์ผ
1,894
๋ถ„์„๋œ ์ค„ ์ˆ˜
2
๋ฐœ๊ฒฌ ์‚ฌํ•ญ
5
์ด ๊ฐ์‚ฌ ์ˆ˜

์œ„ํ—˜ ์š”์ธ

โšก ์Šคํฌ๋ฆฝํŠธ ํฌํ•จ (1)
scripts/monitor-pr-ci.py:1-329
โš™๏ธ ์™ธ๋ถ€ ๋ช…๋ น์–ด (1)
scripts/monitor-pr-ci.py:60-73
๊ฐ์‚ฌ์ž: claude ๊ฐ์‚ฌ ์ด๋ ฅ ๋ณด๊ธฐ โ†’

ํ’ˆ์งˆ ์ ์ˆ˜

68
์•„ํ‚คํ…์ฒ˜
100
์œ ์ง€๋ณด์ˆ˜์„ฑ
85
์ฝ˜ํ…์ธ 
22
์ปค๋ฎค๋‹ˆํ‹ฐ
90
๋ณด์•ˆ
91
์‚ฌ์–‘ ์ค€์ˆ˜

๋งŒ๋“ค ์ˆ˜ ์žˆ๋Š” ๊ฒƒ

Orchestrate Agent Teams

Delegate complex features to specialized agents while maintaining visibility and quality control

Monitor CI at Scale

Track CI status across multiple PRs with automated waiting and failure detection

Break Down Features

Transform feature requests into actionable tasks with clear scope and acceptance criteria

์ด ํ”„๋กฌํ”„ํŠธ๋ฅผ ์‚ฌ์šฉํ•ด ๋ณด์„ธ์š”

Activate CoS Mode 
Act as my Chief of Staff. I want you to coordinate coding agents for my project.
Create Investigation Task 
Create an investigation task to understand the scope of refactoring our authentication system.
Delegate Implementation 
Based on the investigation, create implementation tasks for the refactor. Use Haiku for simple changes and Sonnet for complex ones.
Monitor Multiple PRs 
Check the CI status for PR #456 and PR #458. Wait for both to complete and report results.

๋ชจ๋ฒ” ์‚ฌ๋ก€

  • Use the two-tier model: expensive models for investigation, cheap models for implementation
  • Always include acceptance criteria and scope boundaries in task descriptions
  • Verify VibeKanban MCP tools are available before starting coordination
  • Track blocked agents and escalate decisions that exceed their scope

ํ”ผํ•˜๊ธฐ

  • Writing code instead of delegating after CoS mode activation
  • Creating vague tasks without context or acceptance criteria
  • Assigning tasks that are too large or lack clear scope boundaries
  • Ignoring agent blockers for more than 30 minutes

์ž์ฃผ ๋ฌป๋Š” ์งˆ๋ฌธ

Which AI tools are supported?
Claude and Claude Code are supported. The skill works with VibeKanban MCP tools for task management.
What is the two-tier delegation model?
Use Opus/Sonnet for investigation tasks that require planning, then Haiku/Sonnet for implementation tasks that follow the plan.
How does CoS mode persist?
Once activated, CoS mode persists for the entire conversation. All requests become task delegations until explicitly reverted.
Is my data safe?
Yes. The skill only accesses VibeKanban MCP tools and runs read-only gh CLI commands for CI monitoring.
Why is gh CLI required?
The monitor-pr-ci.py script uses gh CLI for efficient GitHub API access. No direct network calls are made.
How is this different from regular Claude Code?
As Chief of Staff, you delegate rather than execute. Create tasks instead of writing code directly.

๊ฐœ๋ฐœ์ž ์„ธ๋ถ€ ์ •๋ณด

์ž‘์„ฑ์ž

ClementWalter

๋ผ์ด์„ ์Šค

MIT

๋ฆฌํฌ์ง€ํ† ๋ฆฌ

https://github.com/ClementWalter/rookie-marketplace/tree/main/chief-of-staff/skills/agent-coordination

์ฐธ์กฐ

main

ํŒŒ์ผ ๊ตฌ์กฐ

๐Ÿ“ examples/

๐Ÿ“„ task-templates.md

๐Ÿ“ references/

๐Ÿ“„ cos-workflow.md

๐Ÿ“„ delegation-patterns.md

๐Ÿ“„ vibekanban-api.md

๐Ÿ“ scripts/

๐Ÿ“„ monitor-pr-ci.py

๐Ÿ“„ SKILL.md