v0.5.3 - Agent-owned semantic generalization
Immutable
release. Only release title and notes can be modified.
What changed
- Generalizes repeated narrow failures by Agent-owned behavior shape: responsibility, trigger, reachable execution path, intended effect, and observable verification.
- Uses summary-first, bounded evidence loading before opening full records.
- Requires subsumed rule/proposal IDs, preserved domain details, counterexamples, behavior-loss analysis, net context reduction, and explicit human approval.
- Adds fresh-context positive and negative acceptance cases with schema validation and canonical-LF digest binding.
Compatibility and boundaries
- Workspace Schema remains v1.
- No new public command, runtime similarity threshold, embeddings/vector database, background/global scan, or semantic auto-merge.
- Semantic replacements remain proposal-only until explicitly approved.
Verification
- GitHub Actions: Windows and Ubuntu passed for both push and pull-request runs.
- Exact merge-commit archive: 198 tests, 178 passed, 20 expected Windows skips, 0 failed.
- Upgrade validation is performed from the published v0.5.2 asset to this release asset before completion.
Upgrade
Use the existing release-based updater from v0.5.2. Run UpdateDryRun first, then apply the exact returned plan hash.