The Agent Skills Directory

[DATA_EXPOSURE]: The skill is designed to read and modify configuration files located in the user's home directory (e.g., ~/.openclaw/openclaw.json and ~/.kimi_openclaw/openclaw.json), which are documented in references/openclaw_architecture.md to contain sensitive information such as apiKey.
[COMMAND_EXECUTION]: The skill uses the subprocess module in scripts/openclaw_config.py to execute system commands. It attempts to restart the openclaw-gateway service by calling systemctl restart openclaw-gateway, systemctl --user restart openclaw-gateway, and the openclaw CLI tool.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external JSON configuration files and model definitions to perform its operations.
Ingestion points: Configuration files are loaded via load_config in scripts/openclaw_config.py and model definitions are loaded via load_model_json in scripts/add_model.py.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between configuration data and administrative instructions.
Capability inventory: The skill has the capability to write to the file system (saving configs and creating/pruning backups in config-backups/) and to restart system services via subprocess calls.
Sanitization: The skill uses standard JSON parsing (json.load) but lacks secondary validation or sanitization of the content of these files before they influence the agent's logic or are written back to disk.

openclaw