qoder-cloud-agents
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill documents and provides tools for executing shell commands within the Qoder Cloud Agents platform. This includes the 'Bash' tool for agents and the 'setup_script' field in environment configurations, which allows users to define idempotent shell scripts that run during the provisioning of cloud sandboxes. These capabilities are restricted to the remote execution environment provided by the service.
- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by enabling agents to ingest and process data from external, potentially untrusted sources. This risk is inherent to the function of an AI agent that interacts with user messages and web resources.
- Ingestion points: Data enters the agent context through 'user.message' events, file resources mounted from '/mnt/session/uploads/', and results from 'WebFetch' and 'WebSearch' tools.
- Boundary markers: The platform uses structured JSON for event delivery, but the skill does not specify particular delimiters or prompt instructions to isolate untrusted content within those strings.
- Capability inventory: The agents configured by this skill possess significant capabilities including remote shell access ('Bash'), file system operations ('Read', 'Write', 'Edit'), network operations ('WebFetch', 'WebSearch'), and GitHub integration ('gh' CLI).
- Sanitization: There are no documented mechanisms for the sanitization or validation of external content before it is processed by the agent.
Audit Metadata