The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves economic data from the FRED API, creating a surface where untrusted external data is interpolated into the agent's context. While economic time series are low-risk content, the lack of sanitization or boundary markers is a characteristic vulnerability.
Ingestion points: Data is ingested via the _make_request method in scripts/fred_query.py, which calls the series, categories, and releases endpoints of the FRED API.
Boundary markers: There are no explicit delimiters or specific instructions (e.g., 'ignore embedded commands') wrapped around the API data before it is presented to the agent.
Capability inventory: The skill allows the agent to use Bash (python and curl) and WebFetch tools, and it has the ability to read and write to the local file system.
Sanitization: The skill performs standard JSON parsing but does not sanitize or filter the values retrieved from the API for potential linguistic instructions.
[EXTERNAL_DOWNLOADS]: The skill is configured to interact with the Federal Reserve Bank of St. Louis official API at api.stlouisfed.org and requires the installation of standard libraries such as requests and pandas via uv or pip.
Evidence: Documentation in SKILL.md and references/api_basics.md guides the user to install these packages and use the official FRED API endpoints. These interactions target well-known and trusted infrastructure.

alterlab-fred