SKILL.md
Parallel Agent Dispatch
Critical rules
- Domains must be independent. If failures might share a root cause, use one agent.
- Each agent gets specific scope, clear goal, constraints, and expected output.
- Never ask agents to "make tests pass" without finding the real cause.
- After return: read summaries, check conflicts, run full suite; verify via
verification-before-completion. - Before dispatching, read [references/patterns.md](references/patterns.md).
Workflow
- Confirm independence (no shared state/files; fixing one cannot affect another).
- Group by domain; write one self-contained brief per domain (see patterns).
- Dispatch agents concurrently.
- Review each summary; resolve conflicts.
- Run the full test suite; integrate only after fresh verification.
Decision: multiple failures → independent? no → single agent; yes → can parallelize? no → sequential; yes → parallel dispatch.
Resources
- [references/patterns.md](references/patterns.md) — brief template, when to parallelize, mistakes. Read before dispatch.
Validation
Before dispatch:
- Independence confirmed
- Each prompt has scope, goal, constraints, errors, return format
After return:
- Summaries read; conflicts checked
- Full suite passes after integration
Constraints
- Related:
debugging-methodology,verification-before-completion,tdd-workflow,git-worktrees. - Do not parallelize exploratory debugging when you do not yet know what is broken.