offload-r2

Warn

Audited by Gen Agent Trust Hub on Aug 18, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [DATA_EXPOSURE]: The skill instructions direct the agent to load sensitive AWS/R2 credentials from a local .env file. Accessing environment files containing secrets is a high-risk activity that can lead to credential leakage.\n
  • Evidence: The workflow explicitly references the path /Users/jonathanburger/remotion/packages/remotion-media/.env to obtain AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY.\n- [DYNAMIC_EXECUTION]: The skill utilizes the bun runtime to execute an inline script that performs the upload to the R2 bucket.\n
  • Evidence: The instruction uses bun -e followed by a script that instantiates an S3Client and writes file data to the network.\n- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where it ingests local files and transmits them to a remote endpoint without sanitization or boundary markers.\n
  • Ingestion points: The skill reads file paths provided as <local-file> in SKILL.md (Step 4).\n
  • Boundary markers: The instructions lack markers to distinguish between data and instructions within the processed files.\n
  • Capability inventory: Includes shell command execution (bun, curl, git), file system read access, and network write capabilities via the Cloudflare R2 endpoint.\n
  • Sanitization: No input validation or content escaping is performed on the files being processed.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 18, 2026, 09:13 PM
Security Audit — agent-trust-hub — offload-r2