The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check_config.sh is vulnerable to shell command injection.
The variable CONFIG_PATH is assigned directly from user-supplied input ($1) without sanitization.
This variable is then interpolated into a shell command: jq -r ".$CONFIG_PATH" ~/.openclaw/openclaw.json.
An attacker can escape the string and execute arbitrary commands by providing input such as foo" | command #.
[CREDENTIALS_UNSAFE]: The skill instructions in SKILL.md explicitly encourage the unsafe practice of hardcoding sensitive credentials.
In "案例 3: pass: 格式的 apiKey 不被支持", the documentation states that API keys "必须硬编码真实 key" (must be hardcoded with the real key) because external references are not supported.
It provides a visual example of hardcoding an OpenAI API key (sk-f873092ea177b75b...). Hardcoding secrets in configuration files is a high-risk practice that leads to credential exposure.
[EXTERNAL_DOWNLOADS]: The skill references and fetches documentation from external sources to guide the agent's actions.
It uses web_fetch to retrieve content from docs.openclaw.ai (e.g., /channels/telegram, /gateway/configuration-reference).
[DATA_EXPOSURE]: The skill facilitates the reading and modification of ~/.openclaw/openclaw.json.
This is a sensitive system configuration file that likely contains bot tokens and API keys. While the primary purpose is legitimate configuration management, the lack of input validation in the helper script poses a risk to this data.

openclaw-config-helper