The Agent Skills Directory

[SAFE]: The skill performs local-only audits of project files to calculate context window consumption. All scanning and analysis are performed on existing project metadata and instructions without external data transmission.
[DATA_EXFILTRATION]: While the skill reads local configuration files such as .mcp.json and CLAUDE.md, this access is strictly for calculating token overhead and identifying redundancy. No sensitive data is extracted or transmitted externally.
[COMMAND_EXECUTION]: The skill defines a command for auditing context space, which involves standard file system read operations for metadata and content length. It does not invoke high-risk shell commands or perform unauthorized system modifications.

context-budget