Convert guidance documents into guardrail-ready YAML eval artifacts.
npx skills add https://github.com/tnunamak/dotfiles --skill skill-to-evals-v2
Please help me install this Agent Skill. Skill: skill-to-evals-v2 Source: https://github.com/tnunamak/dotfiles/tree/main/ai/skills/local/skill-to-evals Command: npx skills add https://github.com/tnunamak/dotfiles --skill skill-to-evals-v2
This Skill turns mixed, prose-like guidance and rule sets into guardrail-ready evals that are atomic, observable, evidence-friendly, and backed by concrete counterexamples.
Use the skill-to-evals-v2 Skill to convert your existing rule or guidance text into YAML eval artifacts that you can turn into hooks, harness gates, or benchmark assertions.
High-intent search queries and answers about installing and using this skill.
To compile prose guidance into enforceable evals, normalize rules into atomic structures, classify them as hard constraints or soft scores, and emit a YAML block with stable IDs and provenance.
Evidence targeting in guardrail evals assigns the strongest practical evidence source, such as output, diff, ast, dom, test, or manual, to make checks reliably gradeable and discriminating.
Add hookability to hard constraint evals by classifying atomic rules as hard constraints and generating a YAML schema that includes a specific hook_suggestion for the enforcement path.
Yes, rule-to-eval compilation requires normalization into atomic rules and emits a single YAML schema block that preserves section and priority data for stable, auditable guardrails.
Creating counterexample-driven evals requires converting normative actions and prohibitions into structured outputs that use counterexamples to discriminate between passing and failing checks.
Yes, rule-to-eval compilation supports soft score classifications alongside hard constraints, allowing measurable thresholds to be normalized into atomic rules for nuanced grading.
memectExtract atomic verification rules from regulatory text with source references.
getsentryValidate XcodeBuildMCP structured output schemas and manifest outputSchema metadata.
trailofbitsValidate cryptographic implementations against Wycheproof test vectors for AES-GCM, ECDSA, EdDSA, and RSA.
uu201Assess narrative structure, consistency, and prose with multi-agent review.
