github alexgreensh/token-optimizer v4.4.0
v4.4.0 - Memory Review, OpenClaw Coach, Token Accuracy Fix

latest releases: v5.7.11, v5.7.10, v5.7.9...
one month ago

Memory Review (Issue #15)

New memory-review subcommand for structural auditing of MEMORY.md and CLAUDE.md. Detects orphaned topic files, broken links, inline content that should be in topic files, stale entries, duplicate rules, task-tracking leakage, and truncation waste. Includes rule inventory for in-session contradiction detection and net-result projections.

Dashboard gets two new Overview cards: CLAUDE.md Health and MEMORY.md Health.

OpenClaw Coach Tab (OC v2.1.0)

Full Coach tab with health score, Working Well/Setup Summary pattern split, costly prompts with cache breakdown, per-turn token parser, never-used skill detection, topic extraction, and agent cost analysis.

Token Accuracy

Fixed three measurement bugs causing ~5.8K token overcount:

  • Core system constant updated from 15K to 12.9K (measured against v2.1.94+)
  • Removed phantom skill wrapper overhead (35 tokens/skill that Claude Code never injects)
  • Commands now correctly counted as on-demand (5 tokens menu entry, not 50 tokens full load)
  • Dashboard shows staleness warning when data is over 1 hour old

Security

Path traversal containment on --project-dir, symlink safety check on backup writes, XSS escaping on health cards. All from 5-agent torture room QA gauntlet.

Don't miss a new token-optimizer release

NewReleases is sending notifications on new releases.