[aharlap]

Stacked on #31, targeting feat/composer-skill.

Generate the composer's library use/avoid table from pinned READMEs and fail CI when the table differs. Remove the old hand-copied guidance while retaining Studio support, barriers, fallbacks and composition guidance.

Following Rubato's source-revision and SHA-256 convention, keep an MIT-licensed README snapshot and per-file hashes under docs/provenance/. The snapshot comes from Praxity/learning-patterns@e9d4fb6989135a44d75f0765c61d3a596a5d4428, the merged headings change from Praxity/learning-patterns#51. The check runs offline, requires no dependencies, and rejects hash mismatches before regenerating. No live clone or moving branch is used in CI.

Validation: 34 affected tests pass, covering stale composer/source text, deliberate pin updates, malformed sections, missing files and source tampering; composer check, manifest/release validation, syntax checks and whitespace check pass. Verified all 15 copied files against the source commit. Regenerated adapter golden hashes. Full suites left to CI; CI pending at submission.

Depends on #31 and Praxity/learning-patterns#51. The final library review moved setup notes out of the selection headings. The checker excludes those notes and preserves course-lookup's empty avoid cell without inventing guidance. No merge requested.

[aharlap]

Closing for now (controller decision, 2026-10-20). The Opus review showed that generating the composer's pattern table from the current README sections makes the composer's guidance worse: six READMEs are thin (several avoid cells are only 'It gives no grade', course-lookup has none), and the copied-README-plus-hashes check catches only hand edits, not upstream changes. Revisit after the READMEs carry real use/avoid guidance, with a simpler design: learning-patterns publishes the extracted sections at a pinned revision and Rubato reads that one artifact. The hand-written patterns.md on main stays the composer's source until then.