Dispatching Parallel Agents

Name: dispatching-parallel-agents
Rating: 92
Author: yohayetsion

Use this skill when you have multiple independent tasks that can be solved concurrently rather than sequentially.

When to Use

•Multiple test failures with different root causes
•Separate subsystems broken independently
•Problems that don't require understanding shared system context
•Tasks that won't interfere with each other

When NOT to Use

•Failures are related or have common root cause
•Full system understanding is required
•Exploratory debugging (don't know what's wrong yet)
•Agents would modify shared resources

The Pattern

Step 1: Analyze & Group

Categorize problems into independent domains:

markdown

## Problem Analysis

### Domain A: [Name]
- Task: [Specific scope]
- Files: [Affected files]
- Independent because: [Why it won't conflict]

### Domain B: [Name]
- Task: [Specific scope]
- Files: [Affected files]
- Independent because: [Why it won't conflict]

Step 2: Create Focused Tasks

Each agent needs:

Element	Good	Bad
Scope	"Fix auth validation in `auth.ts`"	"Fix all tests"
Goal	"Make `test_login_flow` pass"	"Fix authentication"
Constraints	"Don't modify `session.ts`"	(no constraints)
Expected Output	"PR-ready code with passing test"	(vague)

Step 3: Dispatch Concurrently

markdown

## Parallel Dispatch

### Agent 1: [Domain A]
**Task:** [Specific task description]
**Files:** [List of files to work on]
**Success criteria:** [How to verify]
**Constraints:** [What NOT to touch]

### Agent 2: [Domain B]
**Task:** [Specific task description]
**Files:** [List of files to work on]
**Success criteria:** [How to verify]
**Constraints:** [What NOT to touch]

Step 4: Review & Integrate

When agents return:

•Verify no conflicts — Check for overlapping changes
•Run full test suite — Ensure combined changes work
•Merge carefully — Handle any unexpected interactions

Task Specification Template

markdown

## Agent Task: [Name]

### Objective
[One sentence: what must be true when done]

### Scope
- Files to modify: [List]
- Files to NOT modify: [List]

### Context
[Minimum necessary background]

### Success Criteria
- [ ] [Specific verifiable outcome]
- [ ] [Specific verifiable outcome]

### Verification Command
```bash
[Command to prove success]

code


---

## Common Mistakes

### Too Broad
❌ "Fix the failing tests"
✅ "Fix `test_user_creation` in `tests/user.test.ts` — the mock is returning wrong type"

### Missing Context
❌ "Update the config"
✅ "Update `config/auth.ts` to include the new `SESSION_TIMEOUT` env var (default: 3600)"

### Unclear Constraints
❌ "Fix the API"
✅ "Fix the `/users` endpoint. Don't modify `/auth` endpoints — another agent is handling those."

### Vague Success
❌ "Make it work"
✅ "Running `npm test -- --grep 'user creation'` should show 3 passing tests"

---

## Integration Checklist

After all agents complete:

- [ ] All individual verifications passed
- [ ] No file conflicts between agents
- [ ] Full test suite passes
- [ ] No unexpected side effects
- [ ] Changes are logically coherent together

---

## Example: Three Parallel Agents

```markdown
## Dispatch: Fix Dashboard Module

### Agent 1: Chart Component
**Objective:** Fix rendering crash when data is empty
**Files:** `components/Chart.tsx`, `components/Chart.test.tsx`
**Don't touch:** Any other components
**Verify:** `npm test -- Chart` passes

### Agent 2: Data Fetching
**Objective:** Fix timeout on slow networks
**Files:** `services/dashboard.ts`, `services/dashboard.test.ts`
**Don't touch:** Components or UI code
**Verify:** `npm test -- dashboard.test` passes

### Agent 3: State Management
**Objective:** Fix stale state on navigation
**Files:** `store/dashboard.ts`, `store/dashboard.test.ts`
**Don't touch:** Services or components
**Verify:** `npm test -- store/dashboard` passes

---

### Integration
After all complete:
```bash
npm test -- --grep 'Dashboard'

Expected: All dashboard tests pass

code


---

## Related Skills

- `/writing-plans` — Create plans for each agent
- `/verification-before-completion` — Each agent must verify
- `/brainstorming` — Understand problem before dispatching