Read only gh pr review
Skill jawwadfirdousi/agent-skills/read-only-gh-pr-review/skills/read-only-gh-pr-review
Review backend pull requests for correctness, security, performance, maintainability, and test coverage using GitHub CLI plus local repository inspection. Use when asked to review service-layer/API/database changes, audit backend branch diffs, summarize backend risk, or produce actionable must-fix/should-fix feedback.From its SKILL.md
npx -y skills add jawwadfirdousi/agent-skills --skill read-only-gh-pr-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.
- 13 stars13 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.
SKILL.md
6.5 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it
PR Review (Backend, GitHub CLI)
Overview
Review backend pull requests end-to-end using local code analysis and GitHub CLI API calls. Report only actionable, high-signal findings.
Tool Constraints
- Use only:
SemanticSearch,WebSearch,Grep,LS,Glob,Read,Shell,GitHub CLI. - Before any
ghcommand, source the read-only environment script to enable security enforcement:
Replacesource "<SKILL_DIR>/scripts/activate-gh-readonly.sh"<SKILL_DIR>with the absolute path to this skill directory. - After sourcing, use
ghcommands directly—they are intercepted by the read-only wrapper. - Verify CLI auth first with
gh auth status. If not authenticated, ask the user to rungh auth login. - Enforce strict read-only mode at all times.
- Never attempt any write operation, including comments, reviews, edits, assignments, merges, closes, reopens, or API mutations.
- If a requested command is blocked by the wrapper, do not try alternatives that can mutate state.
- The read-only wrapper blocks
command ghand other bypass attempts.
Workflow
- Enable read-only environment.
- Source the environment script:
source "<SKILL_DIR>/scripts/activate-gh-readonly.sh" - All subsequent
ghcommands in this shell session are now protected.
- Source the environment script:
- Prepare review context.
- Confirm identity and auth:
gh auth status,gh api user. - Resolve repository owner/name from the current repo or pass
-R <OWNER>/<REPO>.
- Confirm identity and auth:
- Resolve the target PR.
- Use
gh pr view <PR_NUMBER> [--json <fields>]when PR number is known. - Otherwise shortlist with
gh pr list [flags]and pick the target PR.
- Use
- Sync local repository to the latest PR branch code.
- Fetch the latest remote state for the PR head branch before reviewing code.
- Example flow:
- Get head branch name from PR metadata (
headRefName). - Run
git fetch --prune origin <HEAD_BRANCH>. - Review files from
FETCH_HEADor check out a local review branch from it.
- Get head branch name from PR metadata (
- Gather full PR evidence before judging.
- Metadata:
gh pr view <PR_NUMBER> [--json <fields>] - Diff:
gh pr diff <PR_NUMBER> [--patch|--name-only] - Changed files:
gh api repos/<OWNER>/<REPO>/pulls/<PR_NUMBER>/files --paginate - Reviews:
gh api repos/<OWNER>/<REPO>/pulls/<PR_NUMBER>/reviews --paginate - Checks:
gh pr checks <PR_NUMBER> [--json <fields>] - Comments:
gh pr view <PR_NUMBER> --commentsgh api repos/<OWNER>/<REPO>/issues/<PR_NUMBER>/comments --paginategh api repos/<OWNER>/<REPO>/pulls/<PR_NUMBER>/comments --paginate
- Metadata:
- Inspect changed backend code deeply.
- Read all high-risk touched files locally (
Read,Grep) and correlate with diff hunks. - Prioritize request handlers/controllers, business services, authorization logic, database queries, migrations, background jobs, and queue/event handlers.
- Verify idempotency, transaction safety, concurrency behavior, retry behavior, and backward compatibility for public API contracts.
- Use
gh api repos/<OWNER>/<REPO>/contents/<PATH>?ref=<REF>when exact remote content is needed (content is usually base64 in.content).
- Read all high-risk touched files locally (
- Apply review checklist with risk-first ordering.
- Use
references/review-checklist.md. - Cover security, correctness, data integrity, API compatibility, performance, and test sufficiency before style concerns.
- Use
- Produce actionable review output.
- Report only issues that are likely defects, regressions, or maintainability risks.
- Every issue must pass the Evidence and Verification rules below before it is reported.
- Include exact
file:line, impact, and concrete fix guidance. - End with residual risk and missing validation/testing assumptions.
- Return findings in chat only; do not write any comment or review back to GitHub.
Evidence and Verification
Every reported issue must separate verified facts from assumptions. Never present an assumption as a fact.
- Fact: behavior you verified by reading the actual code, diff, or command output. Cite
file:linefor every fact. - Assumption: anything inferred but not verified—runtime behavior, unseen configuration, external service behavior, data volumes, deployment topology, caller behavior outside the diff.
Rules:
- Before reporting an issue, trace the full code path: callers, dependencies, and existing guards. Do not report an issue that a guard or validation elsewhere already prevents; find and read that code first.
- Do not report an issue based only on the diff hunk when the surrounding file or callers are available to read.
- Label every assumption a finding depends on, and state how to confirm it.
- If a suspicion cannot be verified with available evidence, either report it as a question with the assumption labeled explicitly, or drop it. Do not report it as a defect.
- Never fabricate line numbers, symbols, code snippets, or behavior. Quote real code only.
- If evidence is incomplete (file truncated, command failed, ref unavailable), say so in the finding instead of filling gaps with guesses.
Response Format
Use this section order:
Critical Issues (Must Fix)Important Issues (Should Fix)Suggestions (Consider)
For each issue, use:
Issue: <brief description>
Location: <file:line>
Severity: <Critical|High|Medium|Low>
Facts: <verified behavior with file:line evidence>
Assumptions: <unverified inferences this finding depends on, and how to confirm them; "None" if fully verified>
Problematic Code: <snippet quoted from the actual code>
Suggestion: <specific fix>
Example: <optional patch-style snippet>
Findings with Assumptions: None are confirmed defects. Findings that depend on assumptions must be phrased conditionally (e.g., "If X is not enforced upstream, then...").
GitHub CLI API Equivalents
Use command mappings in references/github-cli-map.md.
Review Tone
- Be constructive and specific.
- Explain impact and rationale.
- Assume positive intent.
- Prefer concise, high-confidence feedback.
- State uncertainty plainly; a labeled assumption is better than a confident guess.
What ships with it: 6 files
20.1 KB alongside SKILL.md, 3 of them executable
agents/
- openai.yaml227 B
references/
- github-cli-map.md2.3 KB
- review-checklist.md3.1 KB
scripts/
- activate-gh-readonly.shruns4.4 KB
- gh-readonly.shruns4.9 KB
- test-readonly.shruns5.2 KB