The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches content from external URLs to extract release notes, which is a potential surface for indirect injection.\n
Ingestion points: External web pages fetched via the agent's web tools or curl in Step 4.\n
Boundary markers: No specific delimiters or "ignore instructions" prompts are mandated for the extraction process.\n
Capability inventory: Includes network requests (fetch, curl), file system access via Bun.file, and subprocess execution (bun, curl).\n
Sanitization: The skill performs specific data cleanup (e.g., version placeholders and media URL normalization) but lacks general text sanitization for the extracted content.\n- [COMMAND_EXECUTION]: The skill relies on executing local scripts and standard CLI tools to manage the ingestion workflow.\n
Evidence: Instructions include running bun preflight.ts, curl for API interactions, and various releases admin management commands.\n- [DYNAMIC_EXECUTION]: A JavaScript snippet is used at runtime to facilitate efficient data transmission.\n
Evidence: Step 5 provides a bun -e execution pattern to batch process release records from a local JSON file and send them to the API endpoint.\n- [DATA_EXFILTRATION]: The skill transmits extracted data to an external API endpoint as part of its core functionality.\n
Evidence: Extracted releases are POSTed to the URL defined in the RELEASES_API_URL environment variable using curl and fetch with an accompanying API key stored in the environment.

local-ingest