Parallel Agents

Dispatch multiple agents for independent problems.

Announce: "I'm using the parallel-agents skill to handle these independent tasks."

When to Use

Use when:

• 3+ test files failing with different root causes
• Multiple independent bugs/features
• No shared state between the work

Don't use when:

• Failures might be related (fix one = fix others)
• Agents would edit the same files
• Need to understand full context first

Step 0: Load Context

Get project config and check for prior strategies:

ec_search:
  query: project config
  type: config

ec_search:
  query: parallel agent strategy
  type: learning

The Pattern

1. Identify Independent Domains

Group problems by what's broken:

• File A tests: One subsystem
• File B tests: Different subsystem
• File C tests: Another subsystem

If fixing A might fix B, they're not independent.

2. Check for Shared Dependencies

Search EC for known coupling:

ec_search:
  query: [component A] depends
  type: learning

ec_search:
  query: [component B] coupling
  type: pattern

If EC indicates these components are coupled, don't parallelize.

3. Create Focused Tasks

Each agent gets:

• Specific scope - One file or subsystem
• Clear goal - Make these tests pass
• Constraints - Don't change other code
• EC context - Relevant patterns/learnings for that area
• Expected output - Summary of findings and fixes

4. Set Up Progress Tracking

Choose ONE approach (don't mix Tasks and TodoWrite):

Option A: Tasks (Preferred) If TaskCreate/TaskUpdate tools are available:

TaskCreate: "Fix file-a.test failures"
TaskCreate: "Fix file-b.test failures"
TaskCreate: "Fix file-c.test failures"

No dependencies needed since these are independent. Agents can share the task list via CLAUDE_CODE_TASK_LIST_ID - updates broadcast in real-time.

Option B: TodoWrite (Fallback) If Tasks aren't available, use TodoWrite to track each workstream.

5. Dispatch in Parallel

Task("Fix file-a.test failures")
Task("Fix file-b.test failures")
Task("Fix file-c.test failures")

All three run concurrently.

6. Review and Integrate

When agents return:

1. Read each summary
2. Check for conflicts (same files edited?)
3. Run full test suite
4. Integrate changes
5. Mark Tasks as completed (if using Tasks)

Agent Prompt Template

Fix the failing tests in [file]:

Failures:
1. [test name] - [error summary]
2. [test name] - [error summary]

EC Context:
- Test command: {test_command}
- [Relevant patterns for this component]
- [Known gotchas for this area]

Your task:
1. Read the test file
2. Identify root cause (@debugging)
3. Fix the issue (@tdd)
4. Verify tests pass (@verifying)

Constraints:
- Only modify files in [scope]
- Don't change [shared files]

Return: Summary of what you found and fixed.

Red Flags

Never:

• Dispatch agents for related failures
• Let agents edit the same files
• Skip the integration verification
• Trust agent success without checking

After Agents Return @verifying

Run full suite to verify no conflicts:

{test_command}

If conflicts exist, resolve manually.

Store Learnings

If parallelization revealed something useful:

ec_add:
  type: learning
  area: parallel-agents
  content: [What worked or didn't work about this parallelization]
  rationale: Discovered during parallel task execution

Worth storing:

• Components that seemed independent but weren't
• Effective task boundaries for this codebase
• Agent prompt patterns that worked well