Guidelines For Performing Code Reviews After Completing Multiple Complex Software Development Tasks
- Spawn parallel sub-agents with tasks to perform a critical self-review of the changes you've made.
- Compile findings into a concise numbered list with severity (critical/medium/low)
- Verify each finding against actual code (no false positives)
- Implement all fixes and run the appropriate lint/test/build pipeline
If you get stuck on any especially complex or recurring issue consider using the systematic debugging skill to investigate further and unblock yourself.
Sub Agent Guidelines
- Instruct sub-agents to keep outputs concise, token-efficient, relevant and actionable focused on your changes and not to nitpick on minor style issues.
- Provide concise relevant context to the sub-agents to help them understand the context of the changes they are reviewing but avoid wasting lots of tokens on context they can infer from the code itself.
- Appropriately scope the review to your changes with clear boundaries.