Code review
Personalised workflow, tools, agents, skills and other stuff in one place
npx -y skills add Perdolique/workflow --skill code-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Review changed code and report evidence-backed findings. Use whenever the user asks for a code review or wants code inspected for problems, including pull requests and other diffs.
The file declares its own license as Unlicense. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
1.7 KB, as published. Nobody here has run it
Code review
Review code without modifying it unless the user asks for fixes.
Scope
- Read repository instructions and relevant domain skills.
- Review changed code and only the context needed to understand it.
- Review the full repository only when requested.
- Do not run repository linters, type checks, builds, or tests during review; use focused non-mutating reproductions only to verify specific issues.
- Report only verified issues introduced, exposed, or worsened by the change, including pre-existing issues it depends on.
Review areas
Correctness
- Flag contract violations, reachable logic errors, broken edge cases, invalid state transitions, data loss, and regressions.
Style
- Flag violations of project conventions and concrete readability or consistency problems.
- Include subjective style suggestions, clearly labeled as preferences.
Waste
- Flag unused, dead, duplicate, temporary, or speculative code and artifacts.
- Recommend deletion when they have no current purpose.
Security
- Flag plausible vulnerabilities with a concrete attacker path and consequence.
Performance
- Flag material problems in supported workloads and state the triggering workload and impact.
- Do not suggest optimizations without a material benefit.
Findings
- Use one finding per root cause at the narrowest relevant
file:line. - Explain the trigger, impact, and smallest correction.
- If no findings exist, say
No actionable findings.