Deterministic linter and scorer for instruction files — SKILL.md, AGENTS.md, CLAUDE.md. No model in the loop: the same bytes score the same everywhere. Use for linting, scoring, auditing or CI-gating an instruction file, and for checking whether the commands a file promises actually resolve in the repo. Trigger phrases: "lint my skill", "score my skill", "audit skill", "review my skill", "harden skill", "make this skill better", "optimize my skill", "improve [metric] from X to Y", "benchmark skill", "check my AGENTS.md", "is my AGENTS.md lying", or paste a SKILL.md for auto-analysis. Also use when a user shares an instruction file without explicit instructions. Do NOT use for authoring a file from scratch — use a skill-creator first, then schliff. Do NOT use for application-code linting, SQL tuning, or runtime behaviour testing.
Deterministic skill linter and scoring engine for Claude Code — the Ruff for SKILL.md files. 7-dimension structural scoring (structure, triggers, quality, edges, efficiency, composability, clarity) with anti-gaming detection, ~32% rule-based patches, and cross-session episodic memory. An autoresearch loop that measures first, then fixes — not the other way around. Use for linting, scoring, and autonomously improving any Claude Code skill: trigger accuracy, output quality, edge coverage, token efficiency, composability, or custom metrics. Works with community, custom, project-local, or global skills. Trigger phrases: "make this skill better", "optimize my skill", "iterate on this skill overnight", "improve [metric] from X to Y", "audit skill", "review my skill", "harden skill", "benchmark skill", "lint my skill", "score my skill", or paste SKILL.md for auto-analysis. Also use when user shares skill without explicit instructions. Do NOT use for brand-new skills from scratch — use skill-creator first, then come