use-jev-cli
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data through an LLM evaluation tool, which represents a potential injection surface.\n
- Ingestion points: Untrusted content such as git diffs, logs, and support transcripts are ingested via --state-file, --state, or piped input in SKILL.md.\n
- Boundary markers: The tool enforces a structural boundary by separating state (content to be judged) from questions (judgments to be made).\n
- Capability inventory: The skill executes the jev-cli binary to perform evaluations and supports routing logic based on the structured output.\n
- Sanitization: The documentation explicitly warns against shell command injection when using the -s flag and recommends using files or pipes to avoid shell expansion of untrusted data.\n- [COMMAND_EXECUTION]: The skill relies on the execution of the jev-cli binary and demonstrates the use of standard system utilities such as jq, bc, and git for data processing and logic branching.\n- [EXTERNAL_DOWNLOADS]: The skill requires the jev-cli binary to be installed on the system path, as specified in the metadata requirements of SKILL.md.
Audit Metadata