The Agent Skills Directory

[COMMAND_EXECUTION]: The file memory_integration.py uses dynamic module loading to integrate with a taxonomy system. It constructs a path to a sibling directory and executes the code found at that path using importlib.util.spec_from_file_location and exec_module. This dynamic execution of computed paths is a significant capability that can be misused if the local environment is manipulated.
[PROMPT_INJECTION]: The skill processes untrusted external data from contact lists and is susceptible to indirect prompt injection because it interpolates this data into tool calls.
Ingestion points: Data enters the system through the enrich command in run.sh, which reads user-provided CSV and YAML files.
Boundary markers: The skill does not use explicit delimiters or instructions to the agent to ignore embedded commands within the contact data during interpolation into queries like /dogpile "{first_name} {last_name} ...".
Capability inventory: The skill utilizes run.sh for subprocess execution and has the ability to write enriched data to the /memory system and local disk storage.
Sanitization: No sanitization or validation of the input strings is performed before they are interpolated into prompts for the downstream research tools.
[EXTERNAL_DOWNLOADS]: The run.sh script ensures required dependencies are present by installing the loguru package from the standard registry via uv pip during initialization.

discover-contacts