judge-trajectory
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [NO_CODE]: The skill repository does not include any executable code, shell scripts, or binary artifacts. The
scripts/directory only contains a README confirming that no deterministic helper is required. This significantly reduces the attack surface as no arbitrary commands are introduced into the agent's environment. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data in the form of agent trajectories, tool traces, and transcripts.
- Ingestion points:
SKILL.mddefines inputs such as 'Candidate trajectories', 'tool traces', and 'transcripts' which originate from external sources. - Boundary markers: The instructions do not define specific delimiters or escaping mechanisms to prevent the LLM from executing instructions that might be embedded within the trajectories being analyzed.
- Capability inventory: The skill itself does not define any tools, subprocess calls, or network operations, limiting the impact of any potential injection to the context of the specific judging task.
- Sanitization: No explicit sanitization or filtering logic is provided for the input data.
Audit Metadata