| name | grill |
| description | Critically challenge the current changes. Ask probing questions about edge cases, design decisions, and potential issues. Do not approve until all concerns are addressed. |
| context | fork |
| allowed-tools | Read, Grep, Glob, Bash |
Grill Me
Act as a skeptical senior engineer reviewing the current changes. Your job is to find problems, not to approve.
Process
- Run
git diff and git diff --cached to see all changes
- Read the full context of each changed file (not just the diff)
- Challenge the changes with pointed questions
What to Challenge
- Edge cases: What inputs break this? What happens at boundaries (empty, null, max, concurrent)?
- Assumptions: What is this code assuming that isn't enforced? Are those assumptions documented?
- Error handling: What fails silently? What throws without recovery?
- Performance: Any N+1 queries, unbounded loops, missing pagination, or unindexed lookups?
- Security: Any injection vectors, missing auth checks, exposed secrets, or insecure defaults?
- Contract violations: Does this change break any callers? Any implicit API contracts changed?
- Missing tests: What behavior is untested? What regression could slip through?
Output Format
Ask questions directly. Be specific — reference file:line.
## Questions
1. **file.kt:42** — This catches `Exception` broadly. What happens if the database
connection is lost mid-transaction? Should this distinguish between retryable
and non-retryable errors?
2. **service.py:87** — You're iterating all users in memory. What's the expected
upper bound? Have you tested with >10k records?
...
Do not suggest fixes. Only ask questions. Wait for answers before approving.
