code-scrutiny

Multi-expert review system for Claude Code. Runs parallel specialist agents to review code, designs, UX flows, algorithms, financial logic, and more — then iteratively fixes issues until clean.

Install

claude plugins add gh:kofort9/code-scrutiny

Commands

/scrutinize [target] [--panel]

Runs parallel expert reviews. Smart-routes to the right panel based on content, or use a flag to force one.

/scrutinize src/api/auth.ts              # Auto-detect (code panel)
/scrutinize docs/user-journey.md         # Auto-detect (UX panel)
/scrutinize src/scoring.ts --algorithm   # Force algorithm + incentives panel
/scrutinize --staged                     # Review staged git changes
/scrutinize src/api/ --full              # Run ALL applicable agents

Panels: --code, --ux, --financial, --algorithm, --fullstack, --mobile, --data, --devops, --full

/scrutinize-loop [target] [--panel] [--rounds N]

Iterative tightening loop: review, classify findings, fix actionable issues, verify (run tests), commit, repeat. Stops when all agents approve or max rounds reached.

/scrutinize-loop src/api/auth.ts             # Iterate until clean
/scrutinize-loop src/api/ --fullstack        # Full-stack panel loop
/scrutinize-loop --rounds 3                  # Cap at 3 rounds
/scrutinize-loop src/scoring.ts --dry        # Preview findings only

Features: auto-fix with verify gate, oscillation detection, agent disagreement protocol, out-of-scope issue filing.

/design-loop [url-or-path] [--mode] [--rounds N]

Iterative frontend design refinement using a generator-evaluator pattern. Screenshots your running app, scores against 4 criteria (design quality, originality, craft, functionality), then refines or pivots.

/design-loop http://localhost:5173           # Evaluate and iterate
/design-loop http://localhost:3000/donate    # Specific route
/design-loop --creative                      # Weight originality higher
/design-loop --dry                           # Single evaluation, no changes
/design-loop --pivot                         # Force aesthetic pivot

Agents Included

Agent	Expertise
code-reviewer	Bugs, logic errors, code quality
red-team	Security vulnerabilities, attack vectors
ml-specialist	Algorithms, scoring, statistical logic
ux-reviewer	User flows, friction, accessibility
economist	Incentive design, behavioral economics
accountant	Financial compliance, tax receipts
fullstack-reviewer	API design, auth, database modeling
mobile-reviewer	iOS/Android, lifecycle, offline resilience
data-reviewer	Pipelines, ETL, schema evolution
devops-reviewer	CI/CD, containers, monitoring
visual-reviewer	Screenshots, responsive testing, Lighthouse
design-evaluator	Design quality scoring (generator-evaluator pattern)

Dependencies

• Chrome DevTools MCP (optional): Required for design-loop and visual review in scrutinize. Without it, design-loop won't work and visual review silently degrades. Install via claude mcp add chrome-devtools.

License

MIT