← ClaudeAtlas

coordinator-patternlisted

Decompose complex tasks into parallel sub-agent work using the Research → Synthesis → Implementation → Verification pattern. Inspired by Claude Code's Coordinator Mode. Use when a task is too large for a single agent pass, needs multiple perspectives, or benefits from parallel execution. Works with any agent that can spawn sub-tasks.
VictorGjn/agent-skills · ★ 1 · AI & Automation · score 70
Install: claude install-skill VictorGjn/agent-skills
# Coordinator Pattern Decompose complex tasks into parallel sub-agent work — inspired by Claude Code's Coordinator Mode. ## Concept When a task is too big for one agent pass, break it into phases. The coordinator (you, the main agent) never does the grunt work directly. You decompose, delegate, synthesize, and verify. The pattern: **Research → Synthesis → Implementation → Verification** ## When to Use - Task touches **3+ files or areas** that can be investigated independently - Work can be **parallelized** (independent changes, research threads, analysis axes) - Task needs **verification** — implementation should be checked by a fresh agent with no implementation bias - You need **multiple perspectives** (technical, product, user, security) ## When NOT to Use - Single-file edits or simple lookups - Tasks that are strictly sequential (each step depends on the previous) - Tasks where the user wants to stay hands-on at every step ## The 4 Phases ### Phase 1: Research (Parallel Workers) Spawn workers to investigate the problem space. Each worker gets a **focused, self-contained mandate**. Workers have no shared state — each prompt must contain all context needed. ``` Workers (launched concurrently): Worker A: "Find all files related to authentication. List them, summarize the patterns used, note any inconsistencies. Write findings to /tmp/research-auth.md" Worker B: "Map the database schema for users and sessions. List all migrations