The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to run shell commands such as find, wc, and xargs to calculate file sizes within the project and the user's global configuration directory (~/.claude).\n- [COMMAND_EXECUTION]: An embedded Python script is used in Step 3 to read and parse ~/.claude/settings.json and .claude/settings.json to extract hook definitions for token overhead analysis.\n- [DATA_EXFILTRATION]: The skill accesses sensitive configuration and history files, including ~/.claude/settings.json and MEMORY.md. This access is limited to local file size and structure analysis necessary for the token audit and does not involve network activity or exfiltration.\n- [INDIRECT_PROMPT_INJECTION]: The skill reads and processes user-defined markdown files in .claude/rules/ and ~/.claude/ for classification, which could contain instructions intended to influence the agent.\n
Ingestion points: Contents of files located in .claude/rules/*.md, ~/.claude/CLAUDE.md, and project MEMORY.md.\n
Capability inventory: Bash tool for shell command execution and file reading capabilities.\n
Boundary markers: The instructions do not define delimiters or special markers to isolate the content of audited files from the agent's logic during the classification step.\n
Sanitization: No sanitization or filtering is applied to the content of the markdown files before they are processed by the agent.

token-audit