Agent Orchestration
Guidance on when to apply specialized agents and how to run tasks in parallel.
When to Use This Skill
- •User asks for a plan, review, security check, or refactor
- •After implementing a non-trivial feature (suggest code review)
- •Before committing sensitive or high-risk code (suggest security review)
- •When a task clearly benefits from planning or multiple perspectives
Agent Roles (Conceptual)
| Role | Use for |
|---|---|
| Planner | Complex features, refactors, multi-step work |
| Architect | System design, API shape, tech choices |
| Code reviewer | After writing or changing code |
| Security reviewer | Auth, payments, secrets, new APIs |
| TDD guide | New feature or bug fix (tests first) |
| Refactor/cleaner | Dead code, duplication, clarity |
| Doc updater | Updating docs to match code |
When to Invoke (No Explicit Prompt)
- •Complex feature or refactor → Propose a short plan (or “planner” flow) first
- •Code just written or modified → Suggest a quick self-review or “code reviewer” pass
- •Bug fix or new feature → Suggest TDD-style tests where relevant
- •Architecture or API design → Suggest an “architect” perspective or design doc
Parallel Work
For independent subtasks, do them in parallel instead of strictly one-by-one:
- •Example: “Security pass on auth module, performance pass on list API, type-check on shared utils” can be three parallel analyses, then merge findings.
Multi-Perspective Review
For critical or ambiguous work, consider:
- •Factual correctness (does it match requirements?)
- •Code quality and consistency
- •Security and data handling
- •Redundancy and dead code
You can run these as separate passes or a single structured review.