All 94 skillsQuality11 of 13
stryx:eval
When you must measure how a prompt, an agent or an AI feature behaves.
/stryx:evalInside its piece: 11 specks, one for each file on its shelf. Its twelve group-mates glow on the open book behind it. The light splits toward orange, the band of Quality.
In numbers
counted 2026-10-12 in the installed set- 7reference files
- 2helper scripts
- 11files in the skill
- 134KB on disk
- 7chapters
- 1agent works by this book
Use it when
in the skill’s own wordsWhen measuring how an AI system behaves rather than whether deterministic code is correct
- Writing evals for a prompt, system prompt, agent configuration or an app's LLM feature (chat, summary, extraction, classification, reports)
- Eval-driven development (evals before implementation)
- After a prompt, instruction, model or temperature change
- Before shipping an agent configuration, or building a regression suite for a skill
- Comparing prompts, models or approaches on the same cases
- Tracking whether AI output improves or regresses
- When the user says
- "eval"
- "evals"
- "evaluation"
- "EDD"
- "eval suite"
- "pass rate"
- "pass@k"
- "pass^k"
- "regression check"
- "LLM-as-judge"
- "grader"
- "rubric score"
- "golden set"
- "baseline"
Not for
another skill does it- Tests of deterministic code→ stryx:test
- A single done-claim→ stryx:verify
- The build-time test of a Stryx skill→ stryx:skill-forge
What it covers
5 parts- capability and regression evals
- code
- model and human graders
- pass@k and pass^k targets
- baselines and the eval runner
Inside the book
7 chapters, by title only- Core philosophy
- Words this skill uses
- The procedure
- Graders - pick the cheapest one that decides
- Metrics quick reference
- Storage structure
- Best practices
Its helper
1 agent works by this book- stryx:prompt-engineerWhen a prompt for Claude needs writing or tightening: a skill, an agent, a CLAUDE.md.
Works with
the skills it names most- stryx:skill-forge→When creating or changing a skill, agent, slash command or hook, or a skill did not fire.
- stryx:ai-power→When you are not sure which skill fits, or you say "load aipower": it is the map of all skills.
- stryx:local-ai→When running AI models on your own graphics card: Ollama, LM Studio, ComfyUI.
- stryx:mcp-server→When building a server that gives Claude new tools (an MCP server).
- stryx:test→When tests must be written or fixed, or you want the test written first (TDD).
- stryx:verify→When Claude is about to say done, fixed or ready to commit: fresh proof comes first.
Claude loads it by itself when the moment comes; by hand, type the line at the top. It comes with the whole set: Installs all 94 skills, 32 agents and the brain program as one set. The skill’s own text is private source and is not shown here.