hyfdev-adversarial-review
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions allow subagents to run 'git diff' and other arbitrary 'checks' to discover and verify changes within the repository.
- [INDIRECT_PROMPT_INJECTION]: Untrusted data from user requests or external files is ingested verbatim into subagent prompts. Ingestion points include the 'Request' field in the reviewer brief. The skill uses a structured template as a boundary but does not specify sanitization for the interpolated text.
- [DYNAMIC_EXECUTION]: The skill manages the lifecycle of multiple reviewer subagents, spawning them with specific briefs to analyze deliverables in parallel. It follows the security best practice of ensuring subagents are fresh and do not inherit sensitive context from the main session.
Audit Metadata