openclaw-model-switch
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/switch-model.pyscript executes theopenclaw gateway restartcommand to apply configuration changes. This is a standard administrative task within the OpenClaw ecosystem. - [EXTERNAL_DOWNLOADS]: Instructions in
SKILL.mdusecurlto probe external API endpoints to verify model availability. These requests target user-defined URLs and utilize placeholders for authentication tokens. - [DYNAMIC_EXECUTION]: The troubleshooting guide in
references/troubleshooting-model-config.mdprovides a diagnostic recipe that creates and executes a temporary Python script (/tmp/cap.py) to function as a local echo server for debugging authentication issues. - [INDIRECT_PROMPT_INJECTION]: The skill interacts with external API responses and local configuration files. Ingestion points include
openclaw.jsonand gateway responses. Capabilities include file writing and subprocess execution inswitch-model.py. Data is parsed as structured JSON, providing basic boundary enforcement, and no evidence of unsafe interpolation into executable commands was found.
Audit Metadata