cli-anything-agent-native-software
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone the core framework from the HKUDS/CLI-Anything GitHub repository and install the 'cli-anything-hub' package via pip.
- [COMMAND_EXECUTION]: The framework utilizes
subprocess.runto interact with external GUI applications, enabling agents to execute operations like image editing, 3D rendering, and vector graphics processing. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection as it ingests untrusted data (media files) and produces structured JSON output for agent consumption.
- Ingestion points: External files provided via
--inputparameters and metadata processed by CLI commands. - Boundary markers: Delimiters or 'ignore embedded instructions' warnings are absent in the provided implementation examples.
- Capability inventory: The skill has capabilities for file system writes (configurations, trajectories, script generation), network operations (package installation), and shell command execution.
- Sanitization: Relies on standard CLI parameter validation, such as
click.Path(exists=True), which does not inspect the content of the files for embedded instructions. - [DYNAMIC_EXECUTION]: The skill includes functionality to dynamically generate Python scripts (
setup.py,cli.py), configuration files (config.json), and markdown documentation (SKILL.md) based on internal templates to facilitate the creation of new software interfaces.
Audit Metadata