Dispatching parallel agents
Skill charlieviettq/awesome-agent-skill/.cursor/skills/ai-agent-systems/dispatching-parallel-agents
Curated skill pack for LLM agents in engineer and science workflow (Cursor & Claude ready).
npx -y skills add charlieviettq/awesome-agent-skill --skill dispatching-parallel-agentsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 22 stars22 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Split independent work across parallel agents with explicit boundaries and a parent integration review. Use only when user allows subagents and tasks do not share mutable files. Triggers: "parallel agents", "fan out tasks", "run in parallel".
SKILL.md
2.0 KB, as published. Nobody here has run it
Dispatching parallel agents
Gate (required)
Use this skill only if:
- User or parent explicitly allows parallel subagents, and
- Tasks are independent (no shared files, no ordering dependency on partial output)
If either condition fails, use sequential incremental-implementation instead.
When to use
- Independent reviews (security + tests + API contract) on the same commit
- Parallel research spikes with separate output files
- Large plan with disjoint file ownership per task
When not to use
- Single shared module being edited by multiple agents
- Database migrations or sequential deploy steps
- User did not approve multi-agent execution
Workflow
- Partition — list tasks with disjoint
Files:sets - Brief — each agent gets scope, constraints, verify command, output format
- Dispatch — run parallel agents; cap concurrency to what the harness supports
- Collect — gather outputs without assuming correctness
- Integrate — parent reconciles conflicts, runs full verification, single commit strategy per user rules
Integration checklist
- No conflicting edits to the same paths
- Combined diff reviewed by parent (or
requesting-code-review) - Full test/lint suite on integrated result
- User informed of what each agent did
Anti-patterns
- Parallel implementers on the same feature branch without file boundaries
- Merging subagent output without reading diffs
- Spawning agents to avoid reading the codebase
Related
subagent-driven-development, incremental-implementation, requesting-code-review, verify-before-done
Orchestration patterns inspired by obra/superpowers (MIT).