code-review
Use this skill when you need a risk-driven code review of a PR/diff with severity-ranked findings and actionable fixes; triggers include code review, PR review,.
Install
npx skills add https://github.com/naodeng/awesome-qa-skills/tree/main/skills/en/testing-types/code-review
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install naodeng-awesome-qa-skills@llmmart
git clone https://github.com/naodeng/awesome-qa-skills.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole naodeng/awesome-qa-skills collection as a plugin from our marketplace. Git is the plain clone.
README
Code Review
Skill Overview
Risk-driven review of a PR / diff with severity-ranked findings and actionable fixes — catch logic, security, financial-loss, and maintainability defects before merge.
How to Use
- Open
SKILL.mdin this folder and confirm this skill fits your task. - In your AI tool, call
@skill code-review, then add the diff, business goal, stack, and upstream/downstream context. - If you need a specific output format (table, checklist, report), include it directly in your request.
One-Click Install Script
Run from the repository root:
macOS / Linux
bash ./scripts/install-skills-mac.sh --tool codex --lang en --skill code-review
Windows PowerShell
powershell -ExecutionPolicy Bypass -File .\scripts\install-skills-windows.ps1 -Tool codex -Lang en -Skill code-review
Skill manifest
Code Review
Chinese version: See the corresponding Chinese skill.
When to Use
- Need to review a PR / diff / commit and catch logic, security, financial-loss, or maintainability risks before merge.
- Need a P0/P1/P2-ranked report with locations and actionable fix guidance.
- Need a QA / engineering-quality lens beyond author self-review.
Workflow
- Read and follow the main prompt listed under Progressive disclosure (coverage, structure, quality bar).
- Before reviewing, confirm both an identifiable code version and its reviewable changes; if either is missing, return a blocked result and request the exact material.
- Add only project context that changes the result: change scope, business goal, stack, upstream/downstream deps, known risks, team norms.
- Treat role reports as optional, source-identified context; use Product and UI/UX reports only when this change touches their concerns.
- Default to Markdown; switch formats only when the user asks.
Core Constraints
- Risk-driven: prioritize production failures, financial loss, security, and core maintainability — not naming/indent noise.
- Evidence-based: prefer file path, line, or snippet plus trigger path and impact for each finding.
- Two gates: require both identifiable code version (for example repository + PR / commit / branch / tag / revision) and reviewable change (for example diff/patch, changed-file contents, or an accessible base-to-head range). Code identity, change content, and role reports cannot substitute for one another.
- If either gate is missing, explicitly return
status: blockedand distinguishmissing_code_identityfrommissing_reviewable_change; do not claim review completion, recommend merge, or invent code findings. - Role reports are optional. When using them, retain
source_role. Product reports may add business-rule, state-flow, or acceptance context; UI/UX reports may add UI-state, feedback, responsive, or accessibility context only when relevant. Never present a role view as code fact. - Strict severity: P0 blocks merge, P1 should fix this iteration, P2 can be tech debt.
- Separate confirmed facts from assumptions; do not invent endpoints, fields, environments, or root causes the user did not provide.
- Critique the code, not the author; respect the current stack — do not demand framework/architecture rewrites without authorization.
- Keep output executable: every finding needs a fix direction or before/after example.
Progressive Disclosure
- Before producing output, read and follow
prompts/code-review.md(minimum coverage, output structure, quality bar). - When Excel/CSV/JSON/Word is requested: read
output-formats.mdand honor the format. - When a ready-made template fits: use matching files under
output-templates/. - For deeper review dimensions or severity rubrics: read
references/review-dimensions.md. - For examples or calibration: read matching files under
examples/. - For format conversion or helper checks: prefer existing
scripts/over reinventing. - For the shortest path: read
quick-start.md. - For evaluating/regressing this skill: use
evals/with skill-up.
Pre-delivery Checklist
- Followed the main prompt's output structure
- Confirmed both code-identity and reviewable-change gates; if blocked, did not issue a completed review or merge recommendation
- Minimum coverage focus: change summary, overall risk rating, P0/P1/P2 list, testability/observability, API/contract compatibility, fix order, residual risks and assumptions… (details in main prompt)
- Covered the minimum checklist, or explained omissions
- High-risk items have explicit P0/P1 severity with rationale
- Did not invent details the user did not provide
- Assumptions and gaps are marked
Common Pitfalls
- Do not treat a code snippet or role report as code identity, or a PR / commit identifier as the diff; block when either gate is missing.
- Do not activate Product or UI/UX concerns merely because a report exists; first establish relevance and retain its source.
- Do not treat every item as equally important, or dump low-value style nits.
- Do not skip assumptions and information gaps.
- Do not force refactors outside the change under review.
- Do not dump generic theory unrelated to this change.
Files (awesome-qa-skills)
-
agents
-
openai.yaml 382 B
Loading…
-
-
evals
-
cases
-
basic-success.yaml 1.9 KB
Loading… -
edge-incomplete-input.yaml 1.6 KB
Loading… -
edge-missing-code-version.yaml 3.3 KB
Loading… -
edge-risk-priority.yaml 2 KB
Loading… -
review-correct-patch-control.yaml 1.3 KB
Loading… -
review-reachable-null-defect.yaml 1.4 KB
Loading… -
review-upstream-guard-control.yaml 1.4 KB
Loading…
-
-
eval.yaml 646 B
Loading…
-
-
examples
-
README.md 221 B
Loading… -
sample-diff-review.md 1.4 KB
Loading…
-
-
output-templates
-
template-csv.csv 206 B · in bundle
-
template-excel.tsv 358 B · in bundle
-
template-json.json 393 B
Loading… -
template-markdown.md 606 B
Loading… -
template-word.md 302 B
Loading… -
template-xmind.md 317 B
Loading…
-
-
prompts
-
code-review.md 6.5 KB
Loading…
-
-
references
-
review-dimensions.md 2.1 KB
Loading…
-
-
scripts
-
batch_convert_templates.py 2.7 KB
Loading… -
convert_formats.py 10.2 KB
Loading… -
convert_output_formats.py 255 B
Loading… -
convert_to_csv.py 270 B
Loading… -
convert_to_excel.py 272 B
Loading… -
convert_to_json.py 271 B
Loading… -
convert_to_markdown.py 275 B
Loading… -
convert_to_word.py 271 B
Loading… -
convert_to_xmind.py 272 B
Loading… -
parse_csv.py 272 B
Loading… -
parse_excel.py 274 B
Loading… -
parse_formats.py 5.3 KB
Loading… -
parse_json.py 273 B
Loading… -
parse_markdown.py 277 B
Loading… -
parse_output_formats.py 253 B
Loading… -
parse_word.py 273 B
Loading… -
parse_xmind.py 274 B
Loading… -
xlsx_reader.py 5.2 KB
Loading…
-
-
output-formats.md 377 B
Loading… -
quick-start.md 962 B
Loading… -
README.md 852 B
# Code Review ## Skill Overview Risk-driven review of a PR / diff with severity-ranked findings and actionable fixes — catch logic, security, financial-loss, and maintainability defects before merge. ## How to Use 1. Open `SKILL.md` in this folder and confirm this skill fits your task. 2. In your AI tool, call `@skill code-review`, then add the diff, business goal, stack, and upstream/downstream context. 3. If you need a specific output format (table, checklist, report), include it directly in your request. ## One-Click Install Script Run from the repository root: ### macOS / Linux ```bash bash ./scripts/install-skills-mac.sh --tool codex --lang en --skill code-review ``` ### Windows PowerShell ```powershell powershell -ExecutionPolicy Bypass -File .\scripts\install-skills-windows.ps1 -Tool codex -Lang en -Skill code-review ``` -
SKILL.md 4.7 KB
--- name: code-review description: Use this skill when you need a risk-driven code review of a PR/diff with severity-ranked findings and actionable fixes; triggers include code review, PR review,. license: PolyForm-Noncommercial-1.0.0 metadata: version: "1.0.0" compatibility: "" author: "Awesome QA Skills contributors" language: "en" category: "testing-types" source: "https://github.com/naodeng/awesome-qa-skills" license-url: "https://polyformproject.org/licenses/noncommercial/1.0.0" --- # Code Review **Chinese version:** See the corresponding Chinese skill. ## When to Use - Need to review a PR / diff / commit and catch logic, security, financial-loss, or maintainability risks before merge. - Need a P0/P1/P2-ranked report with locations and actionable fix guidance. - Need a QA / engineering-quality lens beyond author self-review. ## Workflow 1. Read and follow the main prompt listed under Progressive disclosure (coverage, structure, quality bar). 2. Before reviewing, confirm both an identifiable code version and its reviewable changes; if either is missing, return a blocked result and request the exact material. 3. Add only project context that changes the result: change scope, business goal, stack, upstream/downstream deps, known risks, team norms. 4. Treat role reports as optional, source-identified context; use Product and UI/UX reports only when this change touches their concerns. 5. Default to Markdown; switch formats only when the user asks. ## Core Constraints - Risk-driven: prioritize production failures, financial loss, security, and core maintainability — not naming/indent noise. - Evidence-based: prefer file path, line, or snippet plus trigger path and impact for each finding. - Two gates: require both identifiable code version (for example repository + PR / commit / branch / tag / revision) and reviewable change (for example diff/patch, changed-file contents, or an accessible base-to-head range). Code identity, change content, and role reports cannot substitute for one another. - If either gate is missing, explicitly return `status: blocked` and distinguish `missing_code_identity` from `missing_reviewable_change`; do not claim review completion, recommend merge, or invent code findings. - Role reports are optional. When using them, retain `source_role`. Product reports may add business-rule, state-flow, or acceptance context; UI/UX reports may add UI-state, feedback, responsive, or accessibility context only when relevant. Never present a role view as code fact. - Strict severity: P0 blocks merge, P1 should fix this iteration, P2 can be tech debt. - Separate confirmed facts from assumptions; do not invent endpoints, fields, environments, or root causes the user did not provide. - Critique the code, not the author; respect the current stack — do not demand framework/architecture rewrites without authorization. - Keep output executable: every finding needs a fix direction or before/after example. ## Progressive Disclosure - Before producing output, read and follow `prompts/code-review.md` (minimum coverage, output structure, quality bar). - When Excel/CSV/JSON/Word is requested: read `output-formats.md` and honor the format. - When a ready-made template fits: use matching files under `output-templates/`. - For deeper review dimensions or severity rubrics: read `references/review-dimensions.md`. - For examples or calibration: read matching files under `examples/`. - For format conversion or helper checks: prefer existing `scripts/` over reinventing. - For the shortest path: read `quick-start.md`. - For evaluating/regressing this skill: use `evals/` with skill-up. ## Pre-delivery Checklist - [ ] Followed the main prompt's output structure - [ ] Confirmed both code-identity and reviewable-change gates; if blocked, did not issue a completed review or merge recommendation - [ ] Minimum coverage focus: change summary, overall risk rating, P0/P1/P2 list, testability/observability, API/contract compatibility, fix order, residual risks and assumptions… (details in main prompt) - [ ] Covered the minimum checklist, or explained omissions - [ ] High-risk items have explicit P0/P1 severity with rationale - [ ] Did not invent details the user did not provide - [ ] Assumptions and gaps are marked ## Common Pitfalls - Do not treat a code snippet or role report as code identity, or a PR / commit identifier as the diff; block when either gate is missing. - Do not activate Product or UI/UX concerns merely because a report exists; first establish relevance and retain its source. - Do not treat every item as equally important, or dump low-value style nits. - Do not skip assumptions and information gaps. - Do not force refactors outside the change under review. - Do not dump generic theory unrelated to this change.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.