am2rican5

Code quality specialist for deduplication, naming consistency, and dead code detection. Use proactively after large code changes, during refactoring, or when user says "clean up code", "find duplicates", "deduplicate", "improve code quality", "naming consistency", "dead code", or "code sweep".

Performance optimization specialist using benchmark-driven profiling. Use proactively when user says "optimize performance", "make this faster", "profile this", "benchmark", "reduce latency", "improve throughput", or "performance bottleneck".

Translate functional specifications into codebase-aware implementation plans. Analyzes existing patterns, makes architecture decisions with rationale, and decomposes work into ordered atomic tasks. Use proactively when a spec exists and needs an implementation plan, or when user says "plan from spec", "create implementation plan", "translate spec to tasks", or "how should I implement this spec".

Validate implementation against a functional specification. Checks acceptance criteria coverage, edge case handling, and detects spec drift. Use proactively after implementation is complete, or when user says "validate against spec", "check spec conformance", "does code match spec", "spec drift check", or "verify implementation".

Autonomous task execution agent that works through a task queue using lock files. Use when user says "run tasks autonomously", "pick up next task", "work through the task list", "manage task queue", "autonomous task execution", or "process task backlog".

Guided first-principles reasoning that decomposes problems into foundational truths and reconstructs solutions from scratch. Use when user says "first principles", "first-principle", "decompose this problem", "break this down", "challenge my assumptions", "why does this work this way", "rethink from scratch", or wants to question conventional approaches before implementing. Do NOT use for brainstorming, general ideation, spec writing, or implementation tasks.

Orchestrate multiple Claude agents working in parallel on a shared codebase with task locking, conflict avoidance, and autonomous work loops. Use when user says "run parallel agents", "orchestrate agents", "multi-agent workflow", "coordinate agents on shared codebase", "parallelize this work", or "run multiple agents simultaneously". Key capabilities include task decomposition, lock-file coordination, git worktree isolation, and branch-per-agent strategies. Do NOT use for single-agent task execution (use task-runner agent instead) or for learning about testing patterns (use test-driven-autonomous-dev instead).

Orchestrate the full spec-driven development lifecycle through four phases (Specify, Plan, Implement, Validate). Supports three rigor levels and lightweight mode for small changes. Use when user says "spec driven development", "SDD workflow", "specify then implement", "spec first development", "write spec then build", or "develop with specification". Do NOT use for writing specs only (use spec-writer) or for validating existing code against specs (use spec-validator agent directly).

Run a self-improvement cycle that analyzes Claude Code usage patterns and applies targeted improvements to CLAUDE.md, hooks, and skills. Audits current setup, identifies gaps from usage data, and proposes grouped changes with rationale. Use when user says "improve my setup", "self-improve", "optimize my Claude config", "fix my workflow", or after running /insights and wanting to act on the results. Do NOT use for general Claude Code help, configuration questions, or manual editing of individual files.

Create or update Claude Code skills following Anthropic's official guide best practices, with automated validation. Walks through frontmatter, instructions, examples, and troubleshooting sections, then runs validate-skill.sh to verify. Use when user says "create a skill", "build a skill", "new skill", "update this skill", "improve this skill", "fix this skill", or "validate my skill". Do NOT use for creating agents (use agents/ directory directly) or for general Claude Code configuration.