Skip to content

SKILL.md

Latest commit

 

History

History
134 lines (94 loc) · 5.19 KB

SKILL.md

File metadata and controls

134 lines (94 loc) · 5.19 KB
Error in user YAML: (<unknown>): did not find expected key while parsing a block mapping at line 1 column 1
---
allowed-tools: Bash(git:*), Bash(wc:*), Bash(grep:*), Bash(find:*), Bash(head:*), Bash(tail:*), Bash(printf:*), Bash(shasum:*), Read
argument-hint: [repository](default: sebt-self-service-portal) [base-branch|commit-ref] (default: main)
description: Analyze code changes and generate a summary and prioritized human review guide. Use for larger PRs or preparing code for review.
---

PR Review Priority Analysis

Generate a prioritized review guide for human reviewers. Our analysis will identify which code changes warrant the most human attention.

Execution Steps

1. Determine Diff Source

Parse $ARGUMENTS to determine comparison target:

  • Empty or "main" → git diff main...HEAD
  • Branch name → git diff <branch>...HEAD
  • Commit ref (SHA, HEAD~N) → git diff <ref>
  • "staged" → git diff --cached
  • "unstaged" → git diff

Verify the ref exists before proceeding. If invalid, report available branches.

2. Gather Change Metrics

# Get list of changed files with stats
git diff <ref> --stat

# Get detailed diff for analysis
git diff <ref> --unified=3

Extract:

  • Files added, modified, deleted
  • Lines added/removed per file
  • Total change volume

3. Identify Review Units

For each changed file, identify discrete units requiring review:

  • New functions/methods: Extract function name and line range
  • Modified functions: Note what changed
  • Configuration changes: Flag separately
  • Documentation changes: identify semantically distinct chunks of docs
  • New files: Entire file is one unit

Use language-appropriate patterns:

  • Python: def , class , async def
  • Ruby: def, class, module
  • JavaScript/TypeScript: function , const.*=.*=>, class , export
  • Go: func
  • Java or C#: Method signatures, class declarations

4. Score Each Unit

Evaluate each review unit against three criteria:

Security Risk (Weight: Highest)

🔴 High: Authentication, authorization, secrets/credentials, storing personal information, file system access, network requests, and so on 🟡 Medium: Configuration, permissions, logging (if potential for PII leaks), session handling, rate limiting 🟢 Low: Pure computation, display logic, tests, documentation, styling

Complexity (Weight: Medium)

🔴 High: Deep cyclomatic complexity, multiple code paths, recursive logic, complex conditionals, concurrent/async patterns, 🟡 Medium: Moderate branching, callbacks, error handling paths 🟢 Low: Linear flow or simple conditionals, simple CRUD, straightforward transformations

Novelty (Weight: Medium)

🔴 High: New libraries, significant departures from existing code 🟡 Medium: first-of-kind in codebase or moderate variations on existing patterns 🟢 Low: Boilerplate, repeated patterns; stuff that clearly builds on well-known patterns or code that already exists in code base

Also generate an aggregate rating and, for medium or high risk, concisely explain (2-6 words) the reason for attention

5. Generate Output

Format: Markdown Report

## PR Review Priority Guide

**Risk Level**: [🔴 High | 🟡 Medium | 🟢 Low] | **Files**: N | **Lines**: +X / -Y

[1-2 sentence summary: what this PR does and where reviewer attention is needed]

---

### Priority Review Table

| Priority | File | Function/Change | Lines | Sec | Cpx | Nov | Notes |
|:--------:|------|-----------------|:-----:|:---:|:---:|:---:|-------|
| 🔴 | `path/to/file.py` | `authenticate_user()` | 45-89 | 🔴 | 🟡 | 🔴 | New auth flow, handles tokens |
| 🔴 | `api/medicaid.py` | `process_household()` | 112-156 | 🔴 | 🟢 | 🟡 | New data model with sensitive PII |
| 🟡 | `utils/cache.py` | `invalidate_all()` | 23-45 | 🟢 | 🟡 | 🟡 | New cache pattern |
| 🟢 | `models/user.py` | `UserProfile` class | 1-34 | 🟢 | 🟢 | 🟢 | Standard model |

---

### Summary by Priority

**🔴 Critical Review (N items)**
- List critical items with brief rationale

**🟡 Recommended Review (N items)**  
- List medium-priority items

**🟢 Low Priority (N items)**
- Boilerplate, tests, config (can be skimmed)

---

### Review Recommendation

[Actionable guidance: "Start with auth/oauth.py lines 45-89, then review payment logic. Test files can be skimmed for coverage gaps."]

Output Guidelines

  • Sort table by priority (🔴 → 🟡 → 🟢), then by file path
  • Files should be links, where the link text is a path from the repo root, and the URL will jump directly to the first line of the range in the PR changeset. Use shasum if needed to generate the links.
  • Line numbers should be clickable-friendly (exact ranges).
  • Keep notes column concise (under 40 chars)
  • For very large PRs (50+ review units), group by directory and show top 20 critical items with a note about remaining
  • If no 🔴 items exist, explicitly state "No critical security concerns identified"
  • Always include the aggregate risk assessment

Edge Cases

  • No changes found: Report "No changes detected against "
  • Binary files: Note as "binary file changed" with 🟡 priority
  • Renamed files: Track as low priority unless content also changed
  • Deleted files: List separately; generally low priority unless removing security controls