interrogate
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill orchestrates a workflow where findings from automated reviewers (which process untrusted code) include shell commands that the lead agent is instructed to execute. This creates a risk that malicious code under review could influence a reviewer to generate harmful commands.
- Ingestion points: Reviewer findings containing a command in the "evidence" field are ingested by the lead agent in Phase 3 of the SKILL.md.
- Boundary markers: Absent. The instructions do not define any delimiters or warnings to ignore embedded instructions within the command strings.
- Capability inventory: The lead agent has the capability to execute shell commands to verify findings.
- Sanitization: Absent. The lead agent is explicitly told to "re-run each finding's command yourself" to verify the reproduction of findings, bypassing potential validation logic.
- [DYNAMIC_EXECUTION]: The skill relies on the execution of shell commands constructed at runtime by AI models. Phase 3 of SKILL.md directs the agent to re-run commands provided in the "evidence" field of the reviewer output, which constitutes the execution of dynamically generated code from a potentially untrusted source.
- [COMMAND_EXECUTION]: The skill uses arbitrary shell command execution as a verification tool. While this is the intended primary purpose for "stress-testing" code, it lacks the necessary safeguards to prevent the execution of malicious commands if the reviewer's output is compromised or hallucinated.
Audit Metadata