write-skill

What problem does it solve? Authoring Agent Skills that work consistently across Claude Code, Codex, and Grok Build is error-prone: skills drift from repository policy, triggers collide with neighboring skills, and structural or behavioral defects go undetected until runtime. This Skill provides one governed workflow to create, update, and verify skills against a shared authoring contract. ## Core Features & Use Cases - Three routed actions: create a new reusable skill, update existing skills to align with policy, or verify structural, policy, and trigger compliance. - Portable authoring contract: enforces self-contained skill directories, strict frontmatter rules, concrete requirements.intelligence levels, and root-relative references that work across all three harnesses. - Cold-start release gate: requires fresh-agent sessions with digest-pinned evidence before any behavioral self-sufficiency claim can pass, separating paper reasoning from observed results. - Bundled validator: scripts/quick_validate.ts checks frontmatter policy, adversarial YAML forms, Markdown link integrity, and runs the official Claude plugin validator when resolvable. - Use Case: A maintainer needs to add a new skill for a recurring workflow. The Skill derives representative tasks, drafts the smallest self-contained skill, runs the portable validator, and gates release on cold-start sessions proving a fresh agent can complete the tasks from the skill alone. ## Quick Start Ask the agent to create a new Agent Skill for your workflow, or to verify an existing skill directory against the repository's authoring policy.