jianying-harness

Pass

Audited by Gen Agent Trust Hub on Oct 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is designed to interact with a local Rust CLI named jianying. It defines standard operations for job management including job run, job list, job show, and job audit, as well as Model Context Protocol (MCP) services. These executions are strictly scoped to the tool's primary purpose and include explicit guidance on using help flags for argument discovery rather than guessing.
  • [INDIRECT_PROMPT_INJECTION]: The skill manages a potential attack surface where malicious instructions could be embedded in data processed by the agent.
  • Ingestion points: The agent processes job configuration files following the jianying-job/v2 schema and media file metadata such as duration, resolution, and stream facts.
  • Boundary markers: Rules mandate that all input facts must be verified against probe outputs or hashes. The skill also enforces the use of isolated output directories to prevent accidental or malicious modification of original source assets.
  • Capability inventory: Capabilities are limited to subprocess calls to the jianying CLI for file analysis and project manipulation, with a refusal to generate arbitrary scripts (e.g., Python).
  • Sanitization: The skill requires strict validation for schemas, time ranges, and file hashes, instructing the agent to reject mismatched or unknown data structures with UNVERIFIED or BLOCKED statuses.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 13, 2026, 10:38 PM