Skip to content

v0.18.1: fix double verified tags, normalize memory, rule learn, cognitive restructure

Choose a tag to compare

@joryeugene joryeugene released this 21 Feb 08:57
· 255 commits to main since this release

Features

  • hive rule learn: Friction-to-rules pipeline. Reads /insights friction data from ~/.claude/usage-data/facets/, maps repeated friction patterns to behavioral rules, deduplicates via trigram overlap, queues candidates in .pending-rules.md for human review.
  • Verify batch processing: hive v processes all facts in batches of 8 with continue/stop prompts between batches. Increased max_turns from 12 to 25 to prevent failures on complex fact batches.
  • Cognitive restructure: Stats pipeline consolidated. Adaptive help shows recently-used commands first with a Discover section for unused commands. Dashboard polish pass.

Fixes

  • Double [verified:] tags: All memory write paths (precompact auto-correct, precompact auto-add, reflect apply, reflect edit, reflect contradiction update) now strip existing [verified:] tags before appending a new one. Prevents [verified:X] [verified:X] accumulation.
  • normalize_memory() cleanup pass: Runs automatically after hive v. Fixes double tags, removes resolved TODOs, corrects - - malformed prefixes, and deduplicates identical lines. Reports a cleanup summary.

Tests

  • 1335 tests passing.
  • New: tests/test_normalize.py (16 tests for tag stripping, normalization, and write-path fixes)
  • New: tests/test_rule_learn.py (23 tests for friction-to-rules pipeline)

Install/upgrade: uv tool install keephive or hive up