harness-review_skill

Overview

This skill performs context-aware reviews of code, plans, and scope to surface security, performance, accessibility, and quality issues. It automatically determines the appropriate review type from recent activity and supports forced review modes for code, plan, or scope. Use it to get structured, prioritized review feedback without performing implementation work. It is not intended for feature development, bug fixes, or setup tasks.

How this skill works

The skill inspects provided inputs (explicit files, git diffs, or inline code content) and runs a quality-gate analysis to identify high-priority review areas such as missing tests, auth/API changes, accessibility hotspots, and performance risks. In default mode a single reviewer engine produces consolidated findings; in codex mode it spins up parallel expert reviewers (Security, Performance, Quality, Accessibility for code; Clarity, Feasibility, Dependencies, Acceptance for plans; Scope-creep, Priority, Feasibility, Impact for scope). It leverages LSP, AST-grep, and structured git logs when available to perform impact analysis, pattern scans, and reference lookups.

When to use it

• When you mention reviews, code review, plan review, or scope analysis.
• When PRs, diffs, or change sets need a multi-aspect review before merge.
• When you want automated checks for security, performance, accessibility, SEO/OGP, or test coverage.
• When you need prioritized, actionable findings rather than implementation.
• When you want parallel expert perspectives (codex mode) on larger changes.

Best practices

• Provide explicit files when possible—if files are passed, only those are reviewed.
• Include git_diff or commit-range for diffs; the tool estimates impacted files when files are not provided.
• Run the quality-gate first: flag missing tests, auth/api changes, a11y, and performance hotspots.
• Enable MCP tools (AST-Grep, harness_lsp_*) for more accurate pattern and impact analysis when available.
• For large reviews enable codex mode to run per-expert parallel reviewers and aggregate results.

Example use cases

• Pre-merge review of a pull request that touches auth and API layers to surface OWASP-related concerns and LSP-based impact analysis.
• Review a proposed plan from an agent run to check clarity, feasibility, dependency gaps, and acceptance criteria.
• Scope analysis after task additions to detect scope-creep, priority conflicts, and downstream impact.
• Audit a set of changed pages for SEO/OGP and accessibility regressions, including targeted page-range document reads for large specs.
• Run an AST-grep scan for common code smells (console.log in production, empty catches, magic numbers) across changed files.