The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script overhead.js from the plugin's source directory using the Bash tool to generate reports on token usage and tool activity.
[DATA_EXFILTRATION]: It accesses local project files (CLAUDE.md) and session transcript metadata to itemize context costs. This data is used solely for local reporting and optimization recommendations.
[PROMPT_INJECTION]: The instructions include a safety requirement that the agent must only execute tool removal commands if the user explicitly agrees after reviewing the audit report.

cco-overhead