All Skills

Run a comprehensive 8-point security audit on vibe-coded apps. Catches the critical vulnerabilities that AI code generators miss — API key exposure, input validation, rate limiting, auth/authz, SQL injection, password reset flaws, error handling, and logging.

Review your vibe-coded app for the 6 backend concepts that AI code generators skip — caching, retry logic with exponential back-off, database transactions, idempotency, race conditions, and proper auth patterns.

Scan your project for exposed secrets, API keys, and environment variable misconfigurations. Checks source code, git history, frontend bundles, and deployment config.

Design multi-agent AI systems — orchestrator/sub-agent patterns, task decomposition, guardrails, session management, and hooks. Know when to use single vs. multi-agent and how to enforce rules programmatically.

Design Claude API tool definitions with proper descriptions, input schemas, and error handling. The description is KING — Claude chooses tools based on description, not name. Includes tool_choice strategy and MCP considerations.

Fix inconsistent Claude output by diagnosing the root cause — is it a few-shot problem, a tool description problem, or a code enforcement problem? Then apply the right fix to the right layer.

Structured code review with explicit severity criteria — CRITICAL, WARNING, INFO. Reports only concrete, verifiable issues. Uses the separate-session principle: never review code you wrote.

Structured feature planning workflow — explore the codebase, ask clarifying questions, build a step-by-step plan, get approval, then execute. Prevents wasted refactors and wrong architectural turns.

Before executing any task, Claude asks all the clarifying questions needed to deliver the best result. Prevents assumptions, surfaces edge cases, and produces better architecture decisions.