miku-live
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONCREDENTIALS_UNSAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill generates and executes commands across multiple environments, such as curl, JavaScript, and Go, to perform API management tasks.
- [CREDENTIALS_UNSAFE]: The skill utilizes access keys and secret keys (MIKU_LIVE_ACCESS_KEY, MIKU_LIVE_SECRET_KEY) for request signing and includes explicit instructions to suppress these secrets in all outputs.
- [DYNAMIC_EXECUTION]: The skill dynamically selects and runs code in various languages to interact with remote APIs, with built-in logic to handle missing compilers or runtimes.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user intents and parameters which could serve as an attack surface for indirect injection. It mitigates this via strict schema validation and parameter mapping.
- Ingestion points: User input via 'intent' and 'params_json' fields in SKILL.md.
- Boundary markers: The skill refers to references/schemas.md and references/interface-catalog.md to validate and delimit inputs.
- Capability inventory: Execution of HTTP methods (GET, POST, PUT, DELETE, PATCH) against Qiniu infrastructure via system scripts.
- Sanitization: Strict enforcement of parameter placement within host, path, query, and body structures as defined in the interface catalog.
Audit Metadata