alexgreensh/token-optimizer v5.7.14

on GitHub

What's New

Skill Description Analysis Overhaul (fixes #47)

Claude Code raised the skill description truncation limit from 250 to 1,536 characters in v2.1.105. Token Optimizer's Skills Auditor was using the old cap, incorrectly flagging valid descriptions as broken.

Now:

• Descriptions over 1,536 chars flagged as TRUNCATED (correctness bug, high severity)
• Descriptions over 200 chars flagged as verbose (efficiency opportunity, not a bug)
• Combined description + when_to_use length measured together (matching Claude Code behavior)
• Codex sessions now correctly detect truncated descriptions

New Awareness

• skillListingBudgetFraction: warns when 30+ skills may be silently dropped in long sessions (default 4% context budget)
• disallowed-tools frontmatter: detects skills using this v2.1.152+ feature to reduce tool overhead

Fixes

• Haiku 3 retirement warning updated to past tense (retired April 2026)
• Dead verbose_count field removed from schema
• Version synced across all manifests (Claude, Codex, measure.py)

Full Changelog: v5.7.13...v5.7.14