The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it retrieves and processes untrusted data from multiple external API endpoints.
Ingestion points: The agent uses the WebFetch tool to ingest JSON data from numerous public datasets on data.cityofnewyork.us (NYC Open Data), covering landmarks, permits, and violations.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between its own logic and potentially malicious instructions that could be embedded within the retrieved property records.
Capability inventory: The agent has access to WebFetch, Write, Read, and Bash tools, which could be leveraged if an injection were successful.
Sanitization: The skill does not perform validation or filtering on the external content before interpolating it into the final markdown report and inline summary.
[DATA_EXFILTRATION]: The skill makes network requests to data.cityofnewyork.us. As this is a well-known service for public data and the skill's purpose matches this behavior, these operations are considered safe. No data exfiltration to suspicious or unknown domains was observed.

nyc-property-report