skill-forge

Benchmark analysis agent that surfaces patterns in eval results that aggregate stats might hide. Identifies failure clusters, reliability issues, and regression risks across iterations. <example>User says: "analyze the benchmark results"</example> <example>User says: "what patterns do you see in the eval failures?"</example>

Architecture design specialist for Claude Code skills. Analyzes use cases, determines complexity tier (1-4), plans file structure, routing tables, and sub-skill decomposition. <example>User says: "design the architecture for a new DevOps skill"</example> <example>User says: "what tier should my skill be?"</example>

Blind comparison agent for A/B testing skill versions. Evaluates outputs from two skill versions without knowing which is "new" vs "old" to eliminate bias. <example>User says: "compare these two skill versions"</example> <example>User says: "run a blind A/B test on the skill"</example>

Multi-platform skill conversion specialist for Claude Code, OpenAI Codex, Gemini CLI, Google Antigravity, and Cursor. Analyzes skills for cross-platform compatibility, identifies Claude-specific features, suggests adaptation strategies, and assesses conversion risk. <example>User says: "can this skill work on Codex?"</example> <example>User says: "what would I need to change for Gemini?"</example> <example>User says: "convert this skill for Cursor"</example>

Eval execution agent that runs skills against eval prompts and captures outputs, timing data, and token usage. Operates in isolated context to prevent eval bleed. <example>User says: "run this skill against the eval prompts"</example> <example>User says: "execute eval set for my skill"</example>

Benchmark Claude Code skill performance with variance analysis, tracking pass rate, execution time, and token usage across iterations. Runs multiple trials per eval for statistical reliability, aggregates results into benchmark.json, and generates comparison reports between skill versions. Use when user says "benchmark skill", "measure skill performance", "skill metrics", "compare skill versions", "skill performance", "track skill improvement", "skill regression test", or "skill A/B test".

Scaffold and build Claude Code skills from plans or descriptions. Generates SKILL.md files, sub-skills, scripts, references, agents, and templates following the Agent Skills standard. Use when user says "build skill", "scaffold skill", "generate skill", "create SKILL.md", or "implement skill".

Convert Claude Code skills to work on OpenAI Codex, Google Gemini CLI, Google Antigravity, and Cursor. Analyzes platform-specific features, generates target files (openai.yaml, AGENTS.md, GEMINI.md, .mdc rules), adapts frontmatter, converts MCP config, and produces compatibility reports. Use when user says "convert skill", "port skill", "multi-platform", "skill for codex", "skill for gemini", "skill for antigravity", "skill for cursor", "cross-platform skill", "convert to codex", "convert to gemini", "convert to antigravity", or "convert to cursor".

Run evaluation pipelines on Claude Code skills to test triggering accuracy, workflow correctness, and output quality. Spawns executor, grader, comparator, and analyzer sub-agents for parallel evaluation. Generates eval_metadata.json, grading.json, and feedback reports. Use when user says "eval skill", "test skill", "run evals", "evaluate skill", "skill evals", "test skill quality", "run skill tests", or "skill evaluation".

Improve and iterate on existing Claude Code skills based on usage feedback, test results, or changing requirements. Handles under/over-triggering fixes, instruction refinement, new sub-skill addition, and architecture evolution. Use when user says "improve skill", "fix skill", "skill not triggering", "skill triggers too much", "update skill", or "evolve skill".