Improve repo structure, CI pipeline, and contributor experience

[marc0olo]

Summary

• Shared parser module — Extract duplicated frontmatter parsing across 5 generation scripts into scripts/lib/parse-skill.js (single source of truth)
• Structural validation — Add scripts/validate-skills.js checking frontmatter fields, required sections, dependency graph integrity, circular dependency detection, and code block annotations. Runs in CI and locally via npm run validate
• CI pipeline hardened — Split into three workflows: _checks.yml (reusable validation, freshness, version bump checks), ci.yml (PR-only, shows as "PR / checks / validate"), and deploy.yml (push-to-main only, builds and deploys to GitHub Pages). Deploy permissions (pages: write, id-token: write) scoped exclusively to the deploy job
• Frontmatter improvements — Add requires and tags fields to all 12 skills; remove redundant > version: body lines (frontmatter as single source of truth); patch-bump all skill versions
• Auto-generated ai-plugin.json — Now produced dynamically alongside agent.json with accurate skill count and list
• Deterministic sitemap — Moved from committed public/ to build-time dist/ generation using git commit dates (no more new Date() drift)
• Contributor docs aligned — CONTRIBUTING.md adds Setup section, generate step, and versioning rules. New .claude/CLAUDE.md for AI agent contributors
• Schema and template — skills/skill.schema.json and skills/_template/SKILL.md.template for consistent new skill creation
• README updated — Programmatic Access section, accurate tool usage instructions, restored planned REST API section

[marc0olo]

Follow-up: Reconsider ## Prerequisites section vs requires frontmatter field

There's overlap between the requires frontmatter field (machine-parseable) and the ## Prerequisites markdown section (human-readable). Both list tool/package dependencies, creating a drift risk where contributors update one but forget the other.

Since skills are primarily consumed by agents (who parse frontmatter), we should consider:

• Dropping ## Prerequisites as a required section
• Making requires the single source of truth for dependencies
• Optionally enriching requires to capture install hints (e.g. icp-cli >= 0.1.0 + how to install)

This would require updating the template, validator (REQUIRED_SECTIONS), CLAUDE.md, and all 12 existing skills — better suited for a dedicated PR.

Would love to hear thoughts from anyone who has opinions on this. Please comment if you have ideas on how the requires field should evolve or whether ## Prerequisites still adds enough value to keep.

[JoshDFN]

Looks good — pushed a fix commit bumping ic-cdk references from 0.18 to 0.19 in the template, schema, README, and https-outcalls.

[JoshDFN]

Looks good — pushed a fix commit bumping ic-cdk references from 0.18 to 0.19 in the template, schema, README, and https-outcalls.