The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes metadata records from the RNDT catalog (geodati.gov.it), which is an external source where content could be influenced by third parties. Maliciously crafted dataset titles or descriptions could theoretically perform prompt injection against the agent.
Ingestion points: Metadata is retrieved using openrndt search and openrndt get commands, as seen in SKILL.md and references/workflows.md.
Boundary markers: The instructions do not specify the use of clear delimiters or guardrails when handling the external metadata content.
Capability inventory: The skill involves the execution of powerful CLI tools including openrndt, curl, jq, and ogr2ogr to interact with and process geographical data.
Sanitization: There is no explicit logic described for sanitizing or escaping the data retrieved from the external catalog before it is interpreted by the agent.
[COMMAND_EXECUTION]: The skill relies on the execution of shell commands (openrndt, jq, curl, ogr2ogr, python3) to perform its functions. These commands target legitimate Italian government services and are used to search for and verify geographical data as part of the skill's primary purpose.

rndt-explorer