Context audit
An audit for an agent's skill library and memory: what loads at the start of every session, what is silently dropped, and a link ledger that proves a clean-up lost nothing.
- Status
- SHIPPED
- Year
- 2026
- Stack
- Python · Claude Code skills
It measures what each skill costs to discover and to open, finds descriptions that are missing, cut off or too alike to tell apart, and checks the memory index against its limits and its growth rate. Before any change it records every path, URL and link in the library, and it runs the same check afterwards.
On my own setup it found 19 of 42 skills without a visible description, two broken headers, an index nine days from overflowing and a second memory store nobody knew about.