← ClaudeAtlas

dev-grilllisted

Stress-tests a plan by asking hard questions about what could go wrong — edge cases, missing assumptions, and failure modes. Use before starting any significant implementation to catch problems in the plan rather than in the code. Asks 10 or more adversarial questions and does not proceed until all are answered satisfactorily.
talentedgeai/infinite-leverage · ★ 0 · AI & Automation · score 63
Install: claude install-skill talentedgeai/infinite-leverage
# Dev Grill Adversarial plan interrogation. Hard gate before implementation starts. ## The Rule Do not write a single line of implementation code until this skill completes and the plan survives all questions. If any question cannot be answered, the plan is incomplete — refine it first. ## Interrogation Protocol Generate at least 10 questions across these categories: ### Scope and Boundaries - What is explicitly OUT of scope? (name it) - What adjacent systems could be affected that aren't mentioned? - Is there existing code that already does part of this? ### Data and State - What data does this touch? Where does it live? - What happens to existing data during migration/deployment? - Are there race conditions or concurrent access patterns? ### Error and Edge Cases - What happens when inputs are null, empty, or malformed? - What is the failure mode if a dependency is unavailable? - What does the user experience when this fails? ### Testing - How will you verify this works? (specific test scenarios) - How will you verify this didn't break anything? (regression plan) - Is there a component here that's hard to test automatically? ### Deployment - Does this require a migration or schema change? - Is this a breaking change for any callers? - Can this be deployed incrementally or is it all-or-nothing? ### Assumptions - What are you assuming is true that you haven't verified? - What would invalidate this approach entirely? ## Gate For each question, provide an answer. If