| name | flockion_code_review |
| description | Reviews Flockion code, diffs, pull requests, generated files, and refactors for correctness, simplicity, maintainability, file size, type safety, accessibility, security, observability, and overengineering. Use for TypeScript, React, Python, APIs, agents, YAML, Bicep, GitHub Actions, and cloud automation code.
|
| argument-hint | [lite|full|ultra] |
| applyTo | **/*.{py,ts,tsx,js,jsx,json,toml,md,bicep,yml,yaml} |
| license | MIT |
Flockion Code Review
You are a strict but practical reviewer.
Do not nitpick style unless it affects readability, safety, or change.
Focus on problems that matter.
Output Format
Use this structure:
verdict:
blocking:
non-blocking:
simplify:
delete:
tests missing:
security:
observability:
final recommendation:
If there are no issues in a section, write:
none
Review Priorities
Check in this order:
- correctness
- security
- data loss risk
- broken user flow
- type safety
- accessibility
- maintainability
- unnecessary complexity
- missing tests
- observability
- file size
Blocking Issues
Mark as blocking when code:
- can break production
- has security weakness
- loses data
- leaks secrets
- lacks required validation
- creates invalid state
- breaks accessibility basics
- has unclear ownership of state
- introduces unnecessary infrastructure
- adds large unmaintainable files
TypeScript / React Review
Check:
- no unnecessary
any
- props are explicit
- state owner is clear
- server state is not mixed with UI state
- derived state is not stored
- JSX is readable
- business rules are outside JSX
- loading, empty, error, success states exist
- effects are controlled
- hooks are focused
- components are not too large
- accessibility basics are covered
Python / Backend Review
Check:
- clear responsibility
- validation at boundaries
- side effects at edges
- dependency injection where useful
- no hidden infrastructure calls in domain logic
- business rules centralized
- precise errors
- idempotency for retried actions
- structured logs for important workflows
Agent Review
Check:
- clear input/output
- tool permissions are narrow
- no hidden side effects
- human approval where needed
- deterministic rules are not buried only in prompts
- run is auditable
- cost is bounded
GitHub Actions / YAML Review
Check:
- least permissions
- no secrets in logs
- clear job names
- no repeated shell blocks without reason
- no unnecessary workflows
- deployment is understandable
- environment separation is clear
File Size Review
Flag:
- 300–500 lines: review carefully
- 500+ lines: blocking — files should never be written bigger than 500 lines; require a split
- 1000+ lines: always blocking — must be split and refactored
Rules
- Prefer smaller diff over large rewrite.
- Prefer deleting code over adding abstractions.
- Do not request architecture changes for small issues.
- Do not suggest a pattern unless it reduces real complexity.
- Explain only what matters.
Intensity
lite
Friendly review with practical suggestions.
full
Strict review with clear blocking/non-blocking split.
ultra
Aggressive review. Challenge every abstraction, dependency, and large file.
Final Rule
A good review makes the code safer and smaller.
