dispatching-parallel-agents

Dispatching Parallel Agents

Core Principle

One agent per independent problem domain. Parallelize only when independence is real.

When To Use

• Multiple unrelated test files fail.
• Independent subsystems need investigation.
• Research questions can be answered separately.
• Plan tasks touch different files and share no mutable state.

Do Not Use When

• Failures may share one root cause.
• Tasks edit the same files.
• A sequential migration or release step is involved.
• Agents need to communicate continuously.

Dispatch Pattern

1. Group work by independent domain.
2. Write one focused prompt per agent.
3. Include scope, constraints, relevant errors, and expected output.
4. Dispatch agents concurrently.
5. Review summaries for conflicts.
6. Run integration verification after all results return.

Agent Prompt Checklist

• One clear goal.
• One domain or file group.
• Explicit constraints.
• Exact expected output.
• Verification command or evidence requested.

Output Contract

## Independent Domains
## Agent Assignments
## Results
## Conflicts
## Integration Verification
## Next Steps