Context Bloat Reduction via Skill Cleanup
Audit loaded skills and tools, identify unused ones, and disable them to reduce context size.
Overview
As AI systems grow, they accumulate tools and skills that were loaded once but never used again. This bloat consumes context space in every session, reducing the model's ability to focus on relevant tools. A periodic audit identifies unused skills and disables them to keep the context lean.
How It Works
The audit reviews session logs and memory files to identify which skills have been used in the past 30 days. Any skill with zero usage is flagged for removal. Bundled skills are disabled via configuration; user-installed skills are moved to a .disabled directory. The system retains a list of disabled skills so they can be re-enabled if needed later.
Tools Used
Outcome
Removed 12 unused skills from context in February 2026. Reduced per-session token usage by ~8% (fewer skill descriptions in system prompt). Improved tool selection accuracy — the model no longer considers irrelevant tools when choosing how to solve a task.