vaccinate
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates on user-provided artifacts (source code, manuscripts, replication packages) which serve as an ingestion point for external data.\n
- Ingestion points: Artifacts provided by the user for seeding defects (e.g., code files, manuscripts, and manuscript writeups).\n
- Boundary markers: The protocol does not specify the use of delimiters or explicit 'ignore instructions' warnings for the content of the artifacts being processed.\n
- Capability inventory: The skill utilizes Read, Write, Bash, Glob, Grep, and Agent tools to interact with artifacts and execute tests.\n
- Sanitization: No sanitization, escaping, or filtering of the ingested content is mentioned in the protocol.\n- [COMMAND_EXECUTION]: The skill instructions direct the agent to execute local shell scripts (e.g., 'bash scripts/check-model-versions.sh') using the Bash tool to perform verification and qualification tasks.\n- [DYNAMIC_EXECUTION]: The skill uses the Agent tool with 'context: fork' to spawn fresh agent sessions for blind testing of review capabilities, which involves the dynamic execution of instructions in an isolated environment.
Audit Metadata