multi-agent-collab
oleh vdc-k
AI & LLMsRingkasan
Universal multi-agent collaboration methodology for Claude Code.
Deskripsi
Apa yang Dilakukan Skill Ini
A coordination protocol for teams of AI agents working on shared projects. Assigns roles by model capability: a lead model handles architecture and decisions, an engineer model handles execution, and a maintainer model handles archiving. Three shared documents (TASK.md, CHANGELOG.md, CONTEXT.md) act as the sync layer, with QMD indexing enabling on-demand context retrieval instead of full history injection.
Tiered model roles cut token costs by reserving expensive models for high-level decisions while cheaper models handle routine execution and archiving.
Kapan Menggunakannya
- Running an expensive model for architecture decisions while cheaper models write code
- Tracking task status across multiple agent sessions with a shared TASK.md
- Generating weekly progress reports by aggregating tagged CHANGELOG entries
- Flagging repeated agent workflows as candidate reusable skills
- Coordinating parallel agents on a shared codebase without context conflicts
Contoh Alur Kerja
User asks: implement a new API endpoint using multiple agents
- 1
Lead model queries QMD index to retrieve relevant TASK.md fragments
- 2
Lead model breaks the feature into subtasks and updates TASK.md
- 3
Engineer model implements the endpoint and writes tests
- 4
Engineer model appends a tagged CHANGELOG entry and marks tasks complete
- 5
Maintainer model archives completed entries and flags any pattern occurring 3 or more times as a candidate skill
Endpoint implemented with synchronized TASK.md, a new CHANGELOG entry, and a candidate skill flagged if the workflow pattern recurred
Persyaratan
- qmd indexing tool
Instalasi
clawhub install vdc-k/multi-agent-collabJalankan perintah CLI-nya, atau tempel prompt ke Claude Code, Cursor, Codex, atau coding agent apa pun yang Anda pakai.