The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation file references/agent-creation.md and SKILL.md contain a specific hardcoded Telegram bot token (8794745677:AAGusS8oiSXX8NK1F-CzLKqsGYWQGmDpvw8) within usage examples.
[COMMAND_EXECUTION]: The provided Python scripts (scripts/create_agent.py and scripts/add_binding.py) are designed to modify the ~/.openclaw/openclaw.json configuration file, which governs the core behavior and security settings of the OpenClaw gateway.
[DATA_EXFILTRATION]: The skill scripts access and read the contents of the ~/.openclaw/openclaw.json file, exposing sensitive configuration data and bot credentials to the execution environment.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by instructing users to create agent identity files (SOUL.md, AGENTS.md) using templates from the assets/ directory. These files define the agent's core personality and instructions but lack boundary markers or sanitization, potentially allowing external configuration data to influence agent behavior.

openclaw-config