What's Changed
- Agent-dictation eval Phase 2: end-to-end harness (TTS → voxmlx → stop-commit polish), eval-e2e lane + nightly workflow by @T0mSIlver in #127
- PolishHelper: pin mlx-swift-lm past the safetensors-index loader bug by @T0mSIlver in #137
- Agent-dictation eval: human baseline, inspection report, and stage ablations by @T0mSIlver in #136
- Improve terminal-agent polish fidelity by @T0mSIlver in #138
- Pin polishing model weights to an exact HF revision (fixes red main) by @T0mSIlver in #140
Full Changelog: v0.8.1...v0.8.2