The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local shell script with user input: bash "$PLUGIN_ROOT/scripts/memory-manager.sh" search "query". This is susceptible to command injection if the query string contains shell metacharacters.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface due to reading untrusted data from local database tables. * Ingestion points: SQLite tables like commitments, decisions, and ideas in secretary.db. * Boundary markers: Absent; retrieved content is not delimited or quarantined from the agent's instructions. * Capability inventory: Access to Bash, Read, Glob, and Grep. * Sanitization: None; data retrieved from the database is interpolated directly into the response format without escaping or sanitization.

assistant