devs-roundtable
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interpolates user-supplied input via $ARGUMENTS directly into the instructions for both the main agent and 5 parallel sub-agents.
- Ingestion points: The $ARGUMENTS variable in SKILL.md is the primary entry point for untrusted data.
- Boundary markers: There are no delimiters or specific 'ignore embedded instructions' warnings around the user-supplied engineering problem.
- Capability inventory: The skill allows sub-agents to read files, design architectures, and generate 'REAL code' implementation sketches across multiple files.
- Sanitization: No escaping or validation is performed on the input before it is passed to the sub-agents.
- [EXTERNAL_DOWNLOADS]: The 'AZ Skill Feedback' section recommends running a command that downloads the hivenet package from the NPM registry.
- Evidence: npx --yes hivenet@latest in SKILL.md.
- This represents a download and execution of code from an external repository associated with the vendor's feedback system.
- [COMMAND_EXECUTION]: The skill provides a shell command intended for reporting feedback and telemetry to the developer.
- Evidence: The sh code block in the 'AZ Skill Feedback' section of SKILL.md executes npx with specific parameters directed to a vendor-owned hivenet.app endpoint.
Audit Metadata