apify-integration-development
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building tools that ingest web content via Apify scrapers. It correctly identifies this as an attack surface and explicitly recommends mitigations, including the use of boundary markers like "<<<EXTERNAL_UNTRUSTED_CONTENT>>>", content sanitization, and truncation of results to prevent prompt injection and context window blowouts. These practices demonstrate a high level of security awareness for the target use case.\n
- Ingestion points: Dataset items from Actors (referenced in references/ai-harness-plugin.md and references/workflow-automation.md).\n
- Boundary markers: Recommended in references/ai-harness-plugin.md for all scraped data.\n
- Capability inventory: Tool capabilities include starting Actor runs, reading/writing to key-value stores, and fetching datasets.\n
- Sanitization: Instructions suggest sanitizing markers and capping payload sizes (e.g., 50,000 characters) to defend against adversarial input.\n- [EXTERNAL_DOWNLOADS]: The skill fetches documentation and technical specifications (e.g., agents.md and openapi.json) from apify.com. These are official resources provided by the vendor for integration development and discovery, and do not present a security risk.\n- [COMMAND_EXECUTION]: The skill includes instructions to check for the presence of the Apify CLI (using "apify --help") and perform local development tasks. These are standard environment checks and workflows for a developer-oriented skill and do not involve execution of untrusted code.
Audit Metadata