The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted input from user arguments and stores it in the filesystem, creating a surface for indirect prompt injection in future sessions. • Ingestion points: User-provided text content via the $ARGUMENTS variable. • Boundary markers: Absent; the content is stored in markdown templates without explicit isolation markers to prevent the agent from interpreting stored text as instructions in future interactions. • Capability inventory: The skill uses grep for file reading and performs file system writes within the agents-notes directory. • Sanitization: No validation or sanitization of the input content is performed before storage.
[COMMAND_EXECUTION]: The skill executes the grep command to perform keyword searches within the local agents-notes directory for note organization and updates.

note