正文
获取原文件CHANGELOG.md
Baseline: v0.2.5 SHA-256 746dee780bf91a25a3164075caf0428634a4853ada691b6d528f90d14fa5be76.
Fixed and tested
- Pairwise context-aware relations; context is part of claim identity.
- Explicit version/knowledge-scope distinctions; missing, overlapping or unclear context remains REVIEW.
- Complete expected teaching entities, usable field status, explicit teaching review and assessment policy.
- Original source bytes vs local notes; reject/withdrawn and duplicate-origin coverage safeguards.
- Source snapshots and curation hash/rights checks; strict paths/JSON; bounded ZIP/TAR/TAR.GZ/GZIP; ignore/review not extracted.
- Clinical and nonclinical dataset branches coexist; role-aware handoff sidecars and candidate review queue.
- Frozen relative-path package inventory; re-finalization validates the unchanged package; mutations are rejected.
- Consolidated SKILL.md rather than retaining contradictory incremental release paragraphs.
Regression fixture changes
Original self_test was preserved in the baseline archive. Positive test fixtures now explicitly declare artifact origin; synthetic teaching approvals are added only to synthetic complete test cases. No real-source medical expected values were rewritten. Four earlier source packages are replayed as frozen metadata/claims; local title notes are explicitly local_note rather than original excerpts. No current medical labels, epidemiology, doses or textbook bibliographic assertions were independently reverified in this release.
Compatibility
Skill name/CLI entry names preserved; manifest schema 0.4, software 0.2.6. New runs required; do not edit old manifests. D001–D014 unchanged. No new Skill or third-party medical runtime introduced.
Not established
General clinical accuracy, exhaustive PHI detection, legal license judgment, real-data download success, Codex client orchestration, full D001–D014 integration and production certification remain out of scope.