| name | code-reviewer |
| description | Review repository code changes with a findings-first focus on correctness, regressions, compatibility, and missing validation. Use for code review, risk review, or model adaptation review requests. |
| license | Apache-2.0 |
| metadata | {"author":"Lihui","version":"3.0"} |
Code Reviewer
Use this skill when the user asks for code review, risk review, regression review, or quality review.
Scope
- Review only. Do not edit code unless the user explicitly switches from review to implementation.
- Prioritize real defects and behavior risks over style-only comments.
- Keep the review grounded in the requested scope. If the user names files, models, or directories, do not expand the scope without saying so.
Review Priorities
Review in this order:
- Correctness and behavioral regressions
- Compatibility with existing APIs, configs, and model registration
- Cache, shape, dtype, and sequence-length contracts
- Reliability, error handling, and edge cases
- Missing or weak validation
- Maintainability and readability
Repository-Specific Checks
- For code under
xhmodel_merak/xh_llm/models, check registration, wrapper replacement, cache flow, and output semantics.
- For CLI changes, check command routing, argument compatibility, and help behavior.
- For config or example changes, check that names, paths, and parameters still match the code they exercise.
- If the review is about model adaptation compatibility, also read
.github/skills/xhquant_llm_reviewer/SKILL.md.
Model Review Routing
When the user asks to review a specific adapted model, default review targets are:
xhmodel_merak/xh_llm/models/<model_name>/
examples_merak/llm/<model_name>/ if it exists
When the user asks for a broad LLM review, enumerate model directories under xhmodel_merak/xh_llm/models/ and review them one by one instead of giving a vague aggregate answer.
Output
Return findings first, ordered by severity.
For each finding include:
- file or symbol
- concrete issue
- impact
- minimal fix direction
Then include:
- open questions or assumptions
- residual risk or validation gaps
- a short overall conclusion
If there are no findings, say that explicitly and still mention residual risks or unverified areas.