The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata (such as titles, descriptions, and notes) and resource content from external open data portals, which constitutes an injection surface.\n
Ingestion points: Untrusted data enters the agent context through CKAN MCP tools (ckan_package_search, ckan_package_show) and external REST API responses from data.europa.eu and api.openknowledgegraphs.com.\n
Boundary markers: The skill includes a dedicated "Security" section instructing the agent to treat all returned content as untrusted third-party data and to ignore any embedded instructions found within metadata or resources.\n
Capability inventory: The agent has access to SQL execution via ckan_datastore_search_sql, SPARQL queries via sparql_query, and shell commands (curl, duckdb) for fetching and analyzing data.\n
Sanitization: No programmatic sanitization or schema validation for ingested metadata is specified, relying instead on the agent following its instructional guardrails.\n- [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch data from various open data sources and semantic search APIs.\n
Targets: Fetches data from the official European Union data portal (data.europa.eu), the vendor's semantic search API (api.openknowledgegraphs.com), and various national/regional CKAN instances (e.g., dati.gov.it, data.gov.uk). These downloads are legitimate and necessary for the skill's primary function.\n- [COMMAND_EXECUTION]: The instructions direct the agent to use CLI tools like curl, jq, and duckdb to interact with external APIs and analyze tabular data resources.\n
Usage: Shell commands are used to perform on-the-fly analysis of CSV, JSON, and Parquet resources identified during dataset discovery.

ckan-mcp