github alexgreensh/token-optimizer v3.1.0
v3.1.0 — The Intervention Update

latest releases: v5.6.11, v5.6.10, v5.6.9...
one month ago

What's New

Efficiency Grading (S/A/B/C/D/F)

All quality scores now include a letter grade. See ContextQ:A(82) in your status line, grade badges in the dashboard, and grades in CLI output.

PreToolUse Read-Cache

Detects and optionally blocks redundant file reads with structural digests. Opt-in: set TOKEN_OPTIMIZER_READ_CACHE=1. Default warn mode shows you what is happening; upgrade to TOKEN_OPTIMIZER_READ_CACHE_MODE=block when confident.

Git-Aware Context Suggestions

New measure.py git-context command analyzes your git state to suggest which files should be in context: test companions, frequently co-changed files, and import chains.

.contextignore

Block files from being read with gitignore-style patterns. Place .contextignore in your project root or ~/.claude/.contextignore for global rules.

Performance

  • PostToolUse archive-result now uses tool matchers (saves ~15-30s per session)
  • Read-cache runs as a standalone 45ms script, not through the 9.5K-line monolith

Security

  • Path canonicalization on all cache keys
  • 0o600/0o700 permissions on cache files and directories
  • mtime re-verification on every cache hit
  • LRU pruning at 500 entries per session

Dual Platform

All features ship for both Claude Code (Python) and OpenClaw (TypeScript).


Claude Code: 3.0.2 -> 3.1.0 | OpenClaw: 1.2.0 -> 1.3.0

Don't miss a new token-optimizer release

NewReleases is sending notifications on new releases.