Token Saver Master

What problem it solves

As skill folders grow, SKILL.md files can become verbose, references/ can collect duplicate notes, and the same skill may exist in multiple directories. These issues often fail silently, but they increase context size, raise model-call costs, and make it harder to tell which copy is current. This skill treats token waste as a diagnostic target: it finds avoidable overhead without changing core behavior.

How it works

It runs through diagnose, identify, confirm, and clean-up:
- Diagnostic scan: reads the configured skill directory and measures SKILL.md line count, description length, number of references/ files, and total size.
- Redundancy detection: compares results against thresholds such as SKILL.md ≤100 lines, description ≤120 characters, references/ ≤10 files, and total size ≤500KB, flagging oversized, duplicated, stale, or repeated entries.
- Suggestion report: presents compression, merge, uninstall, and duplicate-directory cleanup as proposed actions instead of applying them immediately.
- Authorized execution: only after user confirmation does it delete, move, or clean up files, with backups created before mutation.

Boundaries and cautions

It is best for periodic local skill audits, especially when SKILL.md has bloated, references/ contains overlapping material, or multiple copies exist. It depends on a clear local path and folder layout. Deletion requires explicit approval, and moving content into references/ only reduces main-file footprint; it does not improve functionality. If skills are generated or synced by another system, verify the source of truth before cleaning.