skillcheck Update: Scorer Fixes, Cleaner Failures, Honest Token Numbers

skillcheck is a static analyzer for SKILL.md files used by agents like Claude Code, Copilot, Codex, and Cursor to load reusable skills. It validates frontmatter, scores description discoverability, checks file references, enforces token budgets, and flags cross-agent compatibility issues. The latest update focused on hardening and accuracy, not new features.

Description scores have improved significantly, with median ratings rising from 75 to 90 across the reference corpus. The scorer now provides more detailed feedback, indicating which patterns are met or missed instead of just a numerical score. This update ensures the scoring accurately predicts whether an agent will find and trigger the skill.

Corrupt files now provide clear error messages, including file names and byte offsets, rather than crashing with a Python traceback. The configuration discovery process is more robust, preventing a single bad file from affecting the entire scan. The documentation has been updated regarding token estimates, advising users to install the tiktoken optional extra if near a budget limit.

The offline heuristic, used for budget checks, has been validated against tiktoken, with the documented numbers now reflecting the measured results. The pass included 1058 tests covering all rule modules, with coverage raised from 75% to 80%, now sitting at 90%. The update aims to ensure the tool works as intended without introducing any new elements.