Add Claude skill evaluation guidance and pilot review

[snissn]

Summary

Implements the next issue #13 slice:

• manual pilot review of the generated Claude-compatible skill bundles
• focused canonical-pack vs generated-skill evaluation guidance and example artifact
• a generator fix discovered during pilot review
• homepage/product copy update for the dual-distribution model

What changed

• fixed the compiler so generated SKILL.md purpose sections say This skill teaches... instead of reusing canonical This pack teaches...
• regenerated the pilot skill bundles and synced the static website copies
• added manual pilot-review notes under eval/
• added a focused React canonical-vs-skill evaluation artifact under eval/
• updated the evaluation tutorial with explicit canonical-pack vs generated-skill guidance
• updated homepage copy so the product surface explains canonical packs, Claude-compatible skills, and MCP retrieval together

Validation

• npm run generate:claude-skill:pilot
• npm run check:claude-skill
• npm run check:mcp-distributions
• npm run build

Issue progress

This PR moves issue #13 forward on:

• Phase 3 manual pilot review
• Phase 3 generator defect remediation and regeneration
• Phase 4 homepage/product copy
• Phase 8 evaluation guidance
• Phase 8 one focused evaluation set

Real Claude-compatible environment validation is still blocked separately by the local Claude Code login requirement from PR #15.

Refs: #13