silent-background-task-execution
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill provides specific instructions for agents to execute shell commands (e.g.,
npm run dev,docker build,python -m http.server) using aCommandLineparameter. The protocol encourages running these tasks in the background ('silently'), which may reduce user visibility into ongoing agent activities. - [INDIRECT_PROMPT_INJECTION]: The skill facilitates the execution of commands provided in the
CommandLinefield, creating a vulnerability surface if the agent populates this field using untrusted external data. - Ingestion points:
CommandLineproperty in tool invocation examples withinSKILL.md. - Boundary markers: None mentioned or enforced.
- Capability inventory: Arbitrary shell command execution capabilities (e.g.,
npm,docker,python,cargo). - Sanitization: No evidence of input validation, escaping, or sanitization for the generated command strings.
- [CREDENTIALS_UNSAFE]: The documentation contains hardcoded local Windows file paths (e.g.,
c:/Users/ASUS/Documents/Newfolder/Antigravity/Major AI Skills) within the example JSON payloads. This reveals internal directory structures and the workstation username.
Audit Metadata