Fix stale glossary adoption in selective retranslation

[cursor]

Bug and impact

run_state.py plan adopted every existing output_chunk*.md without a run_state.json record as record_only by default. If a user had stale translated outputs and then introduced or edited a glossary before the first run-state record, glossary-sensitive chunks could be recorded as current and become permanently unchanged, silently preserving stale terminology in final book outputs. Alias-only glossary edits for already-selected terms were also ignored because term_hash did not include aliases.

Root cause

The untracked-output path in scripts/run_state.py did not check whether the source chunk currently receives glossary terms before choosing record. scripts/glossary.py::term_hash only hashed source/target/category, omitting prompt-affecting aliases.

Fix

• Re-translate untracked existing outputs when their source chunk currently selects glossary terms.
• Continue record-only adoption for untracked outputs with no selected glossary terms, preserving resumability.
• Include sorted aliases in per-term hashes only when aliases exist, avoiding no-alias run-state churn.
• Updated SKILL/README docs and regression tests.

Validation

• Pre-fix CLI repro showed stale chunk0001 planned as record_only, then unchanged after recording.
• Post-fix CLI repro: chunk0001 is in translation_chunk_ids, glossary-free chunk0002 remains record_only.
• Alias-only edit CLI repro triggers glossary_term_hash_changed for chunk0001.
• python3 -m unittest tests.test_run_state tests.test_glossary -v — 69 tests OK.
• python3 -m unittest discover -s tests -p 'test_*.py' -v — OK, 3 skipped.
• python3 -m compileall scripts tests — OK.
• Standard Alice mock full pipeline: 38 chunks converted/merged; HTML/DOCX/PDF generated successfully. EPUB failed only for the documented Ubuntu Calibre 7.6.0 bytes/str issue.