The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/switch-model.py script executes the system command openclaw gateway restart to apply changes after modifying configuration files.
[COMMAND_EXECUTION]: The troubleshooting guide in references/troubleshooting-model-config.md provides instructions to generate a Python script (/tmp/cap.py) and execute it via nohup to run a local echo server for intercepting outgoing network traffic.
[DATA_EXFILTRATION]: The skill has a high degree of access to sensitive data as its primary purpose is reading and writing openclaw.json configuration files, which contain plaintext API keys and authentication tokens. The troubleshooting instructions also guide the user/agent to capture and display these headers, creating an exposure risk.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and acting upon the contents of existing configuration files.
Ingestion points: Reads from ~/.openclaw/openclaw.json and several mirror paths during the discovery phase.
Boundary markers: None; the skill relies on standard JSON parsing and does not use specific markers to distinguish between configuration data and potentially malicious instructions embedded in fields like model names or IDs.
Capability inventory: File system write access (config modification), local command execution (openclaw CLI), and network probing capabilities (curl).
Sanitization: Uses standard json.load and json.dump for file processing, which prevents syntax-based injection but does not validate the semantic content of the configuration data.

openclaw-model-switch