python-pipeline
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external content from URLs, CSV files, and images using Large Language Models (Gemini). It proactively addresses this risk with a comprehensive security section.
- Ingestion points: Data enters the system via
read_input(local CSVs),fetch_one(remote URLs), andclassify_image(binary image data) as seen in SKILL.md. - Boundary markers: The skill explicitly defines an 'Untrusted content boundary' contract, instructing the agent to use
<EXTERNAL_DATA>delimiters and treat all third-party material as data rather than instructions. - Capability inventory: The pipeline patterns include the ability to perform HTTP requests (
httpx), write to the local filesystem (write_output,ProgressTracker), and interact with the Google Sheets and Gemini APIs. - Sanitization: The instruction set mandates structured extraction with schema validation and explicit user confirmation before any high-stakes actions like credential use or publication.
Audit Metadata