Github review
Personal AI artifacts, agent skills, and reusable workflows
npx -y skills add shenxianpeng/skills --skill github-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 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
Use when the user provides a GitHub pull request number, PR ID, or pull request URL and asks for a merge-readiness review. This skill reviews the PR description, comments, linked issues, implementation diff, tests, and code quality to decide whether the PR can be merged.
SKILL.md
4.8 KB, 926 tokens by cl100k_base, as published. Nobody here has run it
GitHub Review
Purpose
Determine whether a GitHub pull request is ready to merge by tracing it back to the validated issue or requirement, checking whether the implementation actually solves the problem, and reviewing code quality.
Inputs
The user may provide:
- A full GitHub pull request URL.
- A PR number or ID. If only a number is provided, infer the repository from the current worktree or conversation. Ask for the repository only when it cannot be determined.
Review Workflow
1. Read the PR
Collect the PR context before reviewing code:
- PR title, body, author, target branch, and current state.
- Files changed and diff.
- PR conversation comments.
- Review comments and unresolved review threads.
- CI/check status when available.
Use the GitHub connector or gh CLI when available. Prefer thread-level review data when determining whether comments are resolved.
2. Trace Linked Issues
Find issues referenced by GitHub closing keywords in the PR title, body, commits, and comments. Include common forms such as:
fix,fixes,fixedclose,closes,closedresolve,resolves,resolved
For each referenced issue:
- Open the issue and read the original report or request.
- Read issue comments and discussion that clarify scope, acceptance criteria, reproduction steps, or maintainer decisions.
- Confirm the issue exists and appears validated. Treat an issue as validated when maintainers, maintainers' labels, reproducible evidence, accepted requirements, or subsequent discussion confirm it is legitimate.
If no linked issue exists, review against the PR's stated requirement, but call out the missing issue traceability as a risk when it matters.
3. Understand the Requirement
Before judging the implementation, write down the actual requirement in your own words:
- What user-visible or technical problem must be solved?
- What behavior is expected after the fix?
- What cases are explicitly in or out of scope?
- What tests or evidence would prove the fix?
Do not approve merge-readiness if the underlying issue is missing, inaccessible, invalid, stale, superseded, or contradicted by discussion.
4. Verify the Solution
Review the diff and relevant surrounding code. Check whether:
- The code path changed is the one exercised by the issue.
- The fix covers the reported reproduction path and important edge cases.
- The implementation avoids regressions in adjacent behavior.
- Tests were added or updated at the right level. When tests are absent, decide whether that is acceptable based on risk and project norms.
- Existing CI or local test results support the change.
When feasible, run focused tests or static checks. If you cannot run them, state that explicitly and base the verdict on code review evidence.
5. Assess Code Quality
Evaluate the code according to the language, framework, and repository conventions:
- Correctness, error handling, concurrency, security, and data validation.
- Simplicity and maintainability.
- Consistency with existing architecture and naming.
- Backward compatibility, migrations, configuration, and documentation when relevant.
- Avoidance of unrelated refactors or behavior changes.
Review comments already raised by humans should be treated as part of the quality bar. A PR with unresolved blocking review comments is not merge-ready unless the comments are clearly obsolete or non-blocking.
Verdict Rules
Return exactly one primary verdict:
Can merge: The linked issue or requirement is valid, the implementation solves it, quality is acceptable, and no blocking concerns remain.Cannot merge: The issue is invalid or unverified, the implementation does not solve the requirement, important cases are missing, quality is below the project bar, tests or evidence are insufficient for the risk, CI is failing for relevant reasons, or blocking comments remain.
If evidence is incomplete for a mandatory part of the workflow, use Cannot merge and explain what must be checked or fixed before approval.
Output Format
Start with the verdict, then provide concise evidence:
Verdict: Can merge | Cannot merge
Requirement: <one-paragraph summary of the validated issue or PR goal>
Evidence:
- <what was reviewed and what supports the verdict>
Blocking issues:
- <required fixes if cannot merge; use "None" if can merge>
Non-blocking notes:
- <optional improvements, risks, or test gaps>
For Cannot merge, be specific: name the file, behavior, missing test, unresolved comment, failing check, or issue-discussion mismatch that must be addressed.
What ships with it: 1 file
211 B alongside SKILL.md
agents/
- openai.yaml211 B