coordinator-mode
Skill9 starsChanged 3 months agoApache-2.0View on GitHub
Advanced multi-agent orchestration with parallel workers, synthesis protocols, and coordinator lifecycle. Use when complex tasks require multiple agents working in parallel with intelligent result synthesis. When the user needs multi-agent coordination, parallel task execution, complex multi-domain work, or when /coordinate or /orchestrate is invoked. NOT for single-domain tasks.
7 other projects use a version of this, 2 of them edited.
Adopted by
Most recently taken up by
- CodeAlive-AI/pragmatic-orchestrationtheir own version, first seen Nov 15, 2026
- myths-labs/musetheir own version, first seen Nov 6, 2026
- vudovn/ag-kitedited, first seen Oct 8, 2026
- vudovn/antigravity-kitedited, first seen Oct 8, 2026
- Aankirz/coordinator-modetheir own version, first seen Sep 23, 2026
- sefaertunc/anthropic-watchtheir own version, first seen Apr 30, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 2
- vudovn/ag-kit★ 8.2k98% same
- vudovn/antigravity-kit★ 8.2k98% same
Wrote their own version · 5
- CodeAlive-AI/pragmatic-orchestration★ 71same name
- myths-labs/muse★ 38same name
- Aankirz/coordinator-mode★ 4same name
- sefaertunc/Worclaude★ 4same name
- sefaertunc/anthropic-watch★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository36 other files it gives its agents.
- .claude/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- api-designSkill
- architectureSkill
- bash-linuxSkill
- batch-operationsSkill
- behavioral-modesSkill
- claude-code-worktreesSkill
- clean-codeSkill
- context-compressionSkill
- database-designSkill
- deployment-proceduresSkill
- documentation-templatesSkill
- error-handlingSkill
- github-issuesSkill
- github-pmSkill
- github-projectSkill
- git-workflowSkill
- goal-builderSkill
- graphifySkill
- intelligent-routingSkill
- lint-and-validateSkill
- lint-fixSkill
- memory-systemSkill
- performance-profilingSkill
- powershell-windowsSkill
- pre-commit-verifySkill
- red-team-tacticsSkill
- release-vsixSkill
- run-testsSkill
- server-managementSkill
- setup-testingSkill
- testing-patternsSkill
- test-triageSkill
- uncle-bob-craftSkill
- vulnerability-scannerSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 7 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 139
Don't peek: Don't read the fork's output file mid-flight.
- Distinctive: line 141
Don't race: After launching, you know nothing about what the fork found.
- Distinctive: line 161
After all workers complete:
“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
<!-- Vendored from ag-kit (github.com/vudovn/ag-kit) @ 20a13da6d4414c7c6ae33db050a9c606eaef9f40 :: .agents/skills/coordinator-mode/SKILL.md. MIT (c) vudovn. -->Coordinator Mode — Multi-Agent Orchestration
Distilled from production-proven coordinator patterns. Transforms sequential agent chains into intelligent parallel orchestration.
Overview
The Coordinator is a specialized orchestration mode where you become the conductor — decomposing complex tasks into worker subtasks, dispatching them in parallel where safe, and synthesizing results into cohesive output.
You are NOT a worker. You are the coordinator. Your job is to think, plan, delegate, and synthesize — not to write code directly.
Coordinator Lifecycle
User Request
↓
1. DECOMPOSE — Break task into worker subtasks
↓
2. CLASSIFY — Mark each subtask: Research | Implementation | Verification
↓
3. DISPATCH — Launch workers (parallel for reads, sequential for writes)
↓
4. MONITOR — Track worker completion notifications
↓
5. SYNTHESIZE — Combine results into unified response
↓
6. VERIFY — Ensure completeness before reporting to userPhase-Based Workflow
| Phase | Purpose | Concurrency | Worker Type |
| Research | Gather information, explore codebase | ✅ Fully parallel | Read-only agents |
| Synthesis | Analyze findings, plan approach | ❌ Coordinator only | No workers |
| Implementation | Make changes to code/files | ⚠️ Sequential per file set | Write-capable agents |
| Verification | Test, lint, validate changes | ✅ Parallel (independent) | Test/security agents |
🔴 Rule: NEVER skip the Synthesis phase. Research → direct Implementation = poor results.
Concurrency Rules
Safe to Parallelize
- ✅ Multiple agents reading different files
- ✅ Security audit + performance audit (read-only)
- ✅ Test runner + linter (independent)
- ✅ Exploring different directories
Must Be Sequential
- ❌ Two agents writing to the same file
- ❌ Implementation that depends on another agent's output
- ❌ Database migration + code that uses the new schema
- ❌ API change + frontend that consumes the API
Worker Prompt Writing Guide
🔴 The Golden Rule: Never Delegate Understanding
❌ WRONG: "Based on your findings, fix the bug"
❌ WRONG: "Based on the research, implement it"
❌ WRONG: "Look at the code and do what's needed"
✅ RIGHT: "The bug is in src/auth/jwt.ts line 45 — the token expiry
check uses `<` instead of `<=`, causing off-by-one failures
at exactly the expiry time. Change line 45 from
`if (now < expiry)` to `if (now <= expiry)`"Why: Phrases like "based on your findings" push synthesis onto the worker instead of doing it yourself. Write prompts that prove YOU understood — include file paths, line numbers, what specifically to change.
Writing Effective Worker Prompts
Brief the worker like a smart colleague who just walked into the room:
- Explain what you're trying to accomplish and why
- Describe what you've already learned or ruled out
- Give enough context about the surrounding problem for judgment calls
- Be specific about scope: what's in, what's out, what another agent handles
- State the output format you expect ("report in under 200 words")
Prompt Templates
Research Worker
Investigate [specific question] in [file/directory scope].
Context: We're trying to [goal] because [reason].
I've already checked [what you checked] and found [what you found].
Report: List [specific deliverable]. Under 200 words.Implementation Worker
Modify [specific file(s)] to [specific change].
Context: [Why this change is needed].
The current code at [file:line] does [X], change it to [Y].
Constraints: Don't touch [out-of-scope files]. Another agent handles [related area].
Verify: [How the worker should confirm success].Verification Worker
Verify that [specific change] works correctly.
Run: [specific commands].
Check: [what success looks like].
Report: Pass/fail with details on any failures.Fork Semantics
When to Fork (vs. Spawn Fresh Agent)
| Scenario | Action | Why |
| Open-ended research question | Fork (omit agent type) | Fork inherits context, shorter prompt |
| Independent parallel research | Fork multiple in one message | Each fork shares cache |
| Specialized domain work | Spawn (specify agent type) | Fresh specialist starts clean |
| Second opinion on your work | Spawn | Independent perspective |
Fork Rules
- Don't peek. Don't read the fork's output file mid-flight. You'll get a completion notification. Reading the transcript pulls tool noise into your context.
- Don't race. After launching, you know nothing about what the fork found. Never fabricate or predict fork results. If the user asks before notification lands, say "the fork is still running."
- Keep prompts short. Forks inherit your full context — write a directive (what to do), not a briefing (what the situation is).
Anti-Patterns
| Anti-Pattern | Why It's Bad | Fix |
| "Look at the code and fix it" | Zero context = workers make wrong assumptions | Provide specific file paths, line numbers, and what to change |
| Skipping Research phase | Implementation without understanding = rework | Always research first, even briefly |
| Launching 10+ workers at once | Overwhelms synthesis, diminishing returns | 2-5 workers per round, synthesize, then more if needed |
| Fabricating worker results | User gets wrong information | Wait for actual notification, say "still running" if asked |
| Coordinator writes code directly | Loses orchestration benefits | Always delegate to appropriate specialist |
Synthesis Protocol
After all workers complete:
## Coordinator Synthesis
### Task: [Original Request]
### Workers Dispatched
| Worker | Type | Status | Key Finding |
|--------|------|--------|-------------|
| agent-1 | Research | ✅ | Found X in file Y |
| agent-2 | Research | ✅ | Identified Y pattern |
### Consolidated Analysis
[Your synthesis — not a copy-paste of worker output]
### Decision & Rationale
[What you decided based on ALL worker findings and why]
### Implementation Plan
1. [Specific change with file path]
2. [Specific change with file path]
### Remaining Risk
- [Anything workers flagged but couldn't resolve]Continue vs. Spawn Decision
| Situation | Action |
| Worker completed, need follow-up on same topic | Continue (send message to worker ID) |
| Worker completed, need different domain work | Spawn new specialist |
| Worker failed, need retry with same context | Continue with corrected instructions |
| Worker failed, need fundamentally different approach | Spawn new |
Best Practices
- Start with 2-3 workers — add more after synthesis if needed
- Research before implementation — always, even for "simple" tasks
- Synthesize, don't copy — your summary should add insight, not repeat
- Verify independently — verification workers shouldn't trust implementation workers
- Track state — note which workers are pending/completed/failed
- Share scratchpad — use a known directory for cross-worker artifacts
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

