The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to execute curl for API interactions and media downloads. The implementation in runninghub.py and runninghub_app.py follows security best practices by passing arguments as lists and using the @filename flag for JSON payloads, which effectively prevents shell injection vulnerabilities.
[CREDENTIALS_UNSAFE]: The skill manages a RUNNINGHUB_API_KEY by reading from the environment or the platform's standard configuration file (~/.openclaw/openclaw.json). It provides a utility script for users to safely store their keys. Additionally, the build_capabilities.py script includes a _sanitize_default function that proactively strips authentication cookies and pre-signed URL signatures from configuration data to prevent accidental credential leakage.
[EXTERNAL_DOWNLOADS]: The skill downloads generated media files and application metadata from www.runninghub.cn. These operations are restricted to the vendor's official domain and are essential for the skill's primary function of delivering generated content to the user.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an intermediary, taking user prompts and passing them to the RunningHub API for media generation. It includes prompt enhancement logic to improve results. While it processes untrusted user input, it maintains clear boundaries by using structured API calls and predefined model routing, reducing the risk of the agent misinterpreting embedded instructions. Capability inventory: The skill utilizes subprocess calls (curl), writes to a specific local output directory (/tmp/openclaw/rh-output/), and performs network operations.
[SAFE]: The skill exhibits professional development practices, including robust error handling and clear documentation. It uses only standard system binaries and does not attempt to install untrusted third-party dependencies.

runninghub