mcpField
Q
qa-review-pr

qa-review-pr

github.com/testomatioTesting & quality1671 installsUpdated 6d ago
Repository

About qa-review-pr

A skill for reviewing pull requests from a senior QA perspective, focused on user-facing risks and scenarios. It organizes findings by completion, compatibility, merge risks, and items that need verification.

Key features

  1. 01Analyzes requirement alignment
  2. 02Surfaces edge cases and interactions
  3. 03Prioritizes user-impacting findings
  4. 04Separates proven risks from scenarios
  5. 05Structures concise review sections

Use cases

  1. 01Reviewing a feature pull request
  2. 02Finding cancellation and boundary cases
  3. 03Checking security and feature interactions
  4. 04Preparing a prioritized QA review

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/testomatio/skills --skill qa-review-pr

After installing, type/qa-review-prin Claude Code to confirm it loaded.

Details

From the project README

QA Review of a Pull Request

Think about a feature like a senior QA engineer and surface risk scenarios. Gather intent with the qa-pr-requirements-analyzer skill first.

Think

  • Does the provided implementation represent the original task or requirement.
  • Possible ambiguities in implementation.
  • Possible contradictions with existing practices, features.
  • Possible duplication of existing features or patterns.
  • Unobvious usage: edge cases, repeated actions, boundary values, cancellations.
  • Combinations: how this feature interacts with other features.
  • Security vulnerabilities.

Prioritize facts

  • Every finding appears in exactly one section. Never restate a finding in another section.
  • A defect proven by the code goes to Is it done or Merge Risks, not both.
  • Anything not proven by the code goes to What must be verified as a scenario, or is dropped.
  • Order every list by impact on end-users, most severe first.
  • Limits are maximums, not targets. Fewer points are better than padded ones.
  • Whole review: at most 180 words. If over, drop the lowest-severity item.

Output

  • Your output should be readable by a person who doesn't understand or doesn't look into code.
  • Use QA language, avoid coding jargon.
  • Avoid mentioning internal variable names, syntax, queries, not relevant for QAs.
  • Never mention HTTP status codes, test coverage, or unit tests.
  • Use high-level business domain specific terms and not low level coding details.
  • If needed mention class names, file names, but never get into deeper internal details.
  • Explain risks and ambiguities in terms of the persona using the software.
  • Try to resolve ambiguities based on your code and requirements understanding.
  • State facts directly. No hedging words: "may", "might", "possibly", "should be sanity-checked", "unverified".
  • Use bold only for the key point of each item.
  • Reply with only the requested sections, named exactly as provided. No preface, no conclusions.
  • Prefer simple wording and short sentences.

Requested sections

  • Section Is it done: does the code meet the original request (PR title, issue description, etc).
    • Verdict is Yes, No, or Partially.
    • If the reported problem still reproduces on any path, verdict is Partially — never "Yes, with a caveat".
    • If any Merge Risks point says the original problem still happens somewhere, verdict is Partially.
    • Issue summary: at most 15 words. Reasoning: at most 30 words.
  • Section Backwards Compatibility: does the change alter behavior of existing features for existing users.
    • If not, write one sentence and stop.
    • New optional inputs and their new errors are not compatibility issues.
  • Section Merge Risks: problems merging this PR introduces, proven by the code.
    • At most 3 points. Empty section is allowed: write No risks found.
    • Each point: impact on user, at most 8 words — who hits it and when, at most 15 words.
  • Section What must be verified: at most 4 riskiest usage scenarios for end-users.
    • Each scenario: What if {persona} {action} — expected result, at most 10 words.
    • Avoid scenarios that are technical and can be unit tested.

Output Format

### 👷‍♀️ Is it done

**<Yes | No | Partially>.** <original issue summary, ≤ 15 words>

<reasoning, ≤ 30 words>

### 🦕 Backwards Compatibility

<'No breaking changes.' + why, ≤ 15 words — OR one line per changed existing behavior>

### 🌋 Merge Risks

1. **<impact on user, ≤ 8 words>** — <who hits it and when, ≤ 15 words>

<0 to 3 points, most severe first>

### 🔬 What must be verified

- **What if <persona> <action>** — <expected result, ≤ 10 words>

<up to 4 scenarios, most risky first>

Comments

Markdown supported