wheel-size-analyzer

Pass

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection Surface: The skill ingests data from external commit messages and CI build logs to track binary sizes. This involves the agent reading untrusted strings from GitHub commits and AWS S3 logs, which could potentially contain instructions designed to influence agent behavior.
  • Ingestion points: External data is retrieved from GitHub via the gh api tool and from ossci-raw-job-status.s3.amazonaws.com for build log analysis.
  • Boundary markers: The instructions do not specify the use of delimiters or "ignore instructions" wrappers to isolate the ingested external text from the agent's system prompt.
  • Capability inventory: Across its scripts and instructions, the skill uses the gh tool for repository interaction and performs network requests to retrieve build diagnostic data.
  • Sanitization: The skill relies on extracting specific patterns, such as release strings and SHAs, which inherently limits the processing of arbitrary instructions found in the data.
  • Authorized Resource Interaction: The skill communicates with official PyTorch infrastructure and well-known services like Amazon S3 and the GitHub API. These interactions are necessary for the skill's core purpose of binary size analysis and follow standard development practices for the PyTorch ecosystem.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 29, 2026, 03:23 AM
Security Audit — agent-trust-hub — wheel-size-analyzer