multi-agent-feature-workflow

Prompt file imported from LuswetiNick/scrappy (.codex/prompts/multi-agent-feature-workflow.md). Copyright stays with the author.

I want to build the following feature:

[Describe feature here]

Use this multi-agent workflow:

1. Spawn planner first.

   • Inspect the repo and produce a feature plan.
   • Use the code-structure skill if the feature touches repeated operational logic or shared mechanics.
   • Decide what belongs in actions versus services.
   • Do not modify files.

2. After the plan is ready, spawn feature_coder.

   • Implement only the approved plan.
   • Use the code-structure skill if implementation would duplicate operational logic.
   • Keep domain rules in actions.
   • Extract only reusable mechanics into services.
   • Keep the diff small.

3. Then spawn test_engineer.

   • Add or update relevant tests.
   • If service extraction happened, test all affected callers.
   • Run the smallest useful test suite first, then broader checks if appropriate.

4. Finally spawn code_reviewer.

   • Review the final diff.
   • Use the code-structure skill to check action/service boundaries.
   • Do not modify files.
   • Return approve / approve with comments / request changes.

5. Optional PR hardening:

   • If a PR/MR/CL exists and I ask for Greptile optimization, spawn greploop_reviewer.
   • Use the greploop skill.
   • Trigger Greptile review.
   • Fix actionable comments.
   • Run tests.
   • Push or re-shelve changes.
   • Stop at 5/5 confidence with zero unresolved comments or max iterations.

Final summary should include:

• plan summary
• action/service architecture decisions
• implementation summary
• tests added and results
• review verdict
• Greptile result, if greploop was used
• remaining risks