cloud-agent

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFEDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill is designed to upload local files and metadata to cloud-based storage (S3/OSS) to enable processing by a remote agent. This operation is central to the skill's purpose of delegating local tasks to the cloud.
  • [REMOTE_CODE_EXECUTION]: The skill allows the agent to run code (including Python, Node.js, and Shell) within isolated Docker containers on a cloud server. While this involves executing arbitrary code, the documentation specifies that this occurs within a secured, isolated sandbox environment.
  • [PROMPT_INJECTION]: As the skill ingests untrusted data from user tasks and uploaded files, it presents a surface for indirect prompt injection where instructions inside files could influence the remote agent's behavior.
  • Ingestion points: Data enters the cloud context via the task, context, and files parameters in the cloud_agent tool call.
  • Boundary markers: The provided documentation does not define specific delimiters or instructions to ignore embedded commands within the ingested data.
  • Capability inventory: The remote environment can execute code, build and deploy projects to the public internet, and access cloud storage.
  • Sanitization: There is no evidence of local sanitization or filtering of content before it is transmitted to the cloud.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 04:19 PM