Code Review

Output: one line stating what is in scope (e.g. "Reviewing PR #12", "Reviewing branch diff vs master", "Reviewing staged changes") and one line warning about any out-of-scope changes if applicable (e.g. "⚠ Staged changes are not part of this PR and will not be reviewed").

1. Identify requirements and intention of changes

Check the following sources in order:

• Conversation context and user messages
• If PR: read PR title and description for intent, linked workitems, and acceptance criteria
• If branch: look for branch name, commit messages
• If workitem found (WI\d+, can be in PR title or description, branch name): load workitem details via ediprod MCP/cli tools

2. Summarize

Fetch the diff determined in step 0. Return: title or scope description, requirements and intention of changes from step 1, summary of changes.

3. Launch ALL 4 Review Agents IN PARALLEL and wait for results

Input to all: title, description, diff. ./agents/spec-compliance.md ./agents/bugs-and-regression.md ./agents/code-quality.md ./agents/design-quality.md

4. Validate (parallel subagents, one per issue)

Input: title, description, issue description. Confirm issue is real Validate code facts (e.g. symbol actually undefined) and repository rule scope.

5. Filter

Retain only confirmed issues.

6. Output

## Code Review — {PR title or changes scope description}

### Summary

{2–3 sentences: what changed and overall verdict}

### Issues

**High**

> No issues found. (if no issues identified, otherwise list as below)

- **H{n} | {Title}** · `{file}:{Lstart-Lend}`
  {State WHY it is a problem. If confidence is medium, explain why. Provide enough detail that an engineer reading this as a PR inline comment immediately grasps the issue and how to fix it. If the fix is small and self-contained, include a suggestion block with the exact code to commit.}
  {new line between issues}

**Medium**

- **M{n} | {Title}** · `{file}:{Lstart-Lend}`
  {same format}

**Low**

- **L{n} | {Title}** · `{file}:{Lstart-Lend}`
  {same format}

### Verdict

**Approve** | **Approve with suggestions** | **Request changes** — {one-sentence justification}

Source: SNIKO/agent-skills — distributed by TomeVault.