The Agent Skills Directory

DATA_EXFILTRATION (HIGH): The skill explicitly maps sensitive file paths in the 'Critical Paths' section. Specifically, it targets ~/.claude/settings.json (which often contains API keys or configuration) and ~/.claude/MEMORY/ (containing potentially sensitive session history). Identifying these paths facilitates the exposure of sensitive user data.
COMMAND_EXECUTION (LOW): The 'Voice Integration' section provides a functional template for using curl to send data to a local server. While targeting localhost, this establishes a pattern for the agent to use subprocesses to transmit internal data to external processes.
PROMPT_INJECTION (LOW): The 'Response Format' and 'First-Person Voice' sections contain strict formatting requirements (e.g., 'CRITICAL', 'IMPORTANT'). While these are functional instructions, they can be used as a surface for formatting-based jailbreaks or to override default agent persona constraints.
Indirect Prompt Injection (LOW):
Ingestion points: The skill is designed to ingest data from untrusted or external files like Contacts.md and MEMORY/ as shown in the examples.
Boundary markers: Absent. The skill provides no instructions for the agent to treat data read from these files as untrusted or to use delimiters.
Capability inventory: File system read access and the ability to execute network requests via curl.
Sanitization: Absent. There is no logic provided to sanitize or validate the content of the files before processing or displaying them.

CORE