awsflow-s3-fileops

Pass

Audited by Gen Agent Trust Hub on Sep 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by downloading untrusted content from remote S3 buckets.
  • Ingestion points: Untrusted data enters the agent context through the DownloadFile and DownloadFolder commands, which fetch objects from S3 keys and prefixes.
  • Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent from interpreting or executing commands that might be embedded within the downloaded files.
  • Capability inventory: The skill possesses capabilities for reading and writing to the local file system, as well as performing network-based S3 operations across different commands.
  • Sanitization: The skill does not implement any visible sanitization or validation of the content retrieved from external S3 sources before the agent processes it.
  • [DATA_EXFILTRATION]: The UploadFile and UploadFolder tools allow for the transfer of local data to external storage. While this is the intended functionality, it provides a vector for exfiltrating sensitive local files (such as credentials, configuration, or personal data) to a remote, potentially attacker-controlled S3 bucket if the agent is misled by a malicious prompt.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 5, 2026, 11:07 AM
Security Audit — agent-trust-hub — awsflow-s3-fileops