The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes metadata (titles, descriptions, keywords) from external, untrusted open-data platforms.
Ingestion points: Dataset APIs and feeds (CKAN, DCAT, Socrata, OpenDataSoft, ArcGIS) accessed via WebFetch and curl as described in SKILL.md.
Boundary markers: The instructions do not define explicit boundary markers or delimiters for the agent when it displays or processes ingested metadata to prevent the agent from following instructions embedded in that data.
Capability inventory: The skill utilizes Write, Edit, Bash(npm, node, curl, git), and WebFetch capabilities.
Sanitization: While SKILL.md mentions a 'description sanitizer' and 'metadata hygiene rules', these are intended for data quality and do not fully prevent adversarial prompt injection embedded in source data.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch data from remote sources.
Evidence: It uses WebFetch and curl to reach external platform APIs (e.g., data.gov, data.europa.eu) as well as vendor-specific resources on GitHub (github.com/datopian/*). These actions are aligned with the skill's primary purpose of data migration.
[COMMAND_EXECUTION]: The skill instructions involve executing shell commands to perform the migration and verify the environment.
Evidence: Use of npx tsx to run metadata harvesting scripts, npm install for project dependencies, and git-lfs for resource management. These are standard operations for the target PortalJS environment.
[CREDENTIALS_UNSAFE]: The skill requires access to a CKAN_API_KEY for write operations.
Evidence: The skill explicitly instructs users to provide this key via environment variables and states it should 'never [be] hardcoded', which aligns with secure development practices.

portaljs-migrate