Highlights
- Adds opt-in sensitive pattern redaction before LCM storage, FTS, summaries, and externalized payloads.
- Adds the deterministic LCM benchmark harness, benchmark policy comparison metrics, and the Codex GPT benchmark policy.
- Adds the inspectable preset surface for runtime visibility.
- Fixes storage rebind when Hermes home changes, disabled/numeric-zero compression threshold handling, automatic compaction noise, lcm_grep schema compatibility, and quarantined assistant replay safety.
Known dependency
- Model-visible
lcm_*tools with saved Hermes Agent toolsets depend on upstream Hermes Agent fix NousResearch/hermes-agent#31194. - The blocked registry fallback from #201 is not included in this release.
Changes
- feat: add opt-in sensitive pattern redaction (#198) (6376272)
- feat: add inspectable preset surface (#195) (d242b0f)
- Fix quarantined assistant replay safety (#197) (51ecd9b)
- feat: add Codex GPT benchmark policy (#194) (9148297)
- feat: add benchmark policy comparison metrics (#193) (f8ceaa2)
- fix: rebind storage when Hermes home changes (#179) (8d631d1)
- fix: treat numeric zero as disabled compression (#192) (84fd9dc)
- fix: ignore disabled Hermes compression threshold (#191) (1df4423)
- feat: add deterministic LCM benchmark harness (#190) (a8049c1)
- fix: mark LCM automatic compaction as silent (#188) (eb7d27b)
- fix: avoid union type arrays in lcm_grep schema (#185) (71dd8a1)