The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like wc, ls, grep, and sed to analyze local file system metadata. These operations are restricted to the agent's own project configuration and memory state files (e.g., ~/.claude/ and CLAUDE.md).
[INDIRECT_PROMPT_INJECTION]: The skill extracts file paths from MEMORY.md to verify their existence. While this involves processing potentially untrusted text stored in the agent's memory, the resulting action (checking if a file exists) is benign and does not lead to command injection or unauthorized data access.
[DATA_EXPOSURE]: The skill accesses line counts and file lists from the agent's internal memory and rule files. This data is used solely for generating a health report and is not exfiltrated to any external destination.

status