The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses multiple bash commands to perform system inspections, including find, grep, file, and which. It also identifies and executes the --version flag on all binaries named claude found within the system path.
[DATA_EXFILTRATION]: The skill accesses sensitive configuration files and shell profiles, including ~/.zshrc, ~/.bashrc, ~/.bash_profile, and ~/.claude/settings.json. These files often contain private environment variables, authentication tokens, and user-specific configurations.
[DATA_EXFILTRATION]: Performs a broad recursive search across the user's home directory (find ~) for files named CLAUDE.md, which exposes the existence and location of various projects and configurations to the agent.
[INDIRECT_PROMPT_INJECTION]: The skill ingests content from local configuration and project files using grep and find, presenting the results to the agent without sanitization or protective delimiters.
Ingestion points: Output from grep and find commands reading CLAUDE.md, settings.json, and shell profiles (~/.zshrc, etc.) in SKILL.md (Phases 2 and 3).
Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish command output from instructional content.
Capability inventory: The skill utilizes bash for system-wide file searches and content inspection across all five phases.
Sanitization: No filtering, escaping, or validation is performed on the data retrieved from the file system before it enters the agent's context.

check-cache-bugs