All Skills

LLM cost optimization pipeline: automatic model routing by text length and item count thresholds, immutable cost tracking with frozen dataclasses, budget enforcement with early stopping, exponential backoff retry, and prompt caching strategies.

Systematic prompt improvement pipeline: structured prompt analysis, A/B testing patterns, metric-driven iteration, and prompt versioning for production LLM applications.

Strategic context window compaction: compact after research before implementation, after milestones before next phase, after debugging before feature work, after failed approaches before new ones. Never mid-implementation.

Audit context window consumption across agents, skills, MCP servers, and rules. Identifies token bloat, redundant components, and produces prioritized savings recommendations. Target: under 10 MCPs, under 80 tools active.

Production deployment workflows: rolling/blue-green/canary strategies, CI/CD pipeline patterns, Docker multi-stage builds, health checks, readiness probes, rollback strategies, and environment-specific configuration management.

Head-to-head coding agent comparison tool: YAML task definitions with judge criteria, git worktree isolation per agent run, pass rate/cost/time/consistency metrics, and reproducible benchmarking across Claude Code, Aider, Codex, and other agents.

Pre-PR verification pipeline: sequential build check, type check, lint, test with coverage, security scan, and diff review. Hard stop on build failure. Report all type errors before proceeding.

Eval-driven development framework: define pass/fail criteria before implementation, capability/regression/consistency eval types, pass@k reliability metrics, grader patterns, and continuous eval integration during development.

Playwright E2E testing patterns: Page Object Model, test configuration, CI/CD integration with artifact capture (screenshots, videos, traces), flaky test quarantine, and critical user flow coverage strategies.

Test-driven development enforcement: write-tests-first methodology, RED/GREEN/REFACTOR cycle, 80%+ coverage requirement across unit/integration/E2E, user journey definitions, test pyramid structure, and coverage verification gates.

Audit skill and command libraries for quality: quick scan mode for changed skills, full stocktake with sequential subagent batch evaluation, quality scoring, and gap analysis.

Docker and Docker Compose patterns: multi-service orchestration for local development, bind mounts with anonymous volume isolation, container networking, multi-stage Dockerfile optimization, container security hardening, and volume backup strategies.

OWASP Top 10 security checklist with code patterns for secrets management, input validation, SQL injection prevention, XSS, CSRF, rate limiting, and pre-deployment security verification.

PostgreSQL quick reference: B-tree/GIN/BRIN index selection, data type cheat sheet, Row Level Security patterns, connection pooling, and Supabase-specific best practices.

Zero-downtime database migration patterns: forward-only production migrations, concurrent index creation, nullable column additions, separate schema vs data migrations, rollback planning.

Backend architecture patterns: repository/service/controller layers, database query optimization, Redis caching, background jobs, middleware chains, and structured error handling.

REST API design patterns: resource naming, HTTP methods, status codes, pagination, cursor-based filtering, error response envelopes, versioning strategies, and rate limiting configuration.

Regression testing strategies for AI-assisted development. Sandbox-mode API testing without database dependencies, automated bug-check workflows, patterns to catch blind spots where the same model writes and reviews code.

SHA-256 content-hash caching for expensive file processing. Path-independent keys auto-invalidate on content change. Frozen dataclass cache entries with file-based O(1) storage.

Scans agent harness configurations for security vulnerabilities, misconfigurations, injection risks, secrets leaks, and permission escalation vectors.

Designs production-grade multi-crate Rust workspace architectures with enforcement boundaries, cryptographic integrity, and deterministic policy evaluation.

Enforces plan-first development via brainstorm, write-plan, execute-plan workflow. No implementation code until plan is approved.

Forces rigorous hypothesis-driven debugging sequence before writing any fix. Prevents guessing. Works across Rust, Flutter, TypeScript, React Native, Python, and SQL.

Generates complete design systems from brand name, vibe, or color input. Outputs CSS variables, typography, spacing, component tokens, and dark/light modes.