The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses host-level shell commands and the openclaw CLI to perform administrative tasks such as directory creation, file backup, and service management. Remote operations are conducted over SSH.
[CREDENTIALS_UNSAFE]: The skill manages SSH credentials (passwords and private keys) for remote server access. It includes strict directives to ensure these credentials are never persisted in logs or configuration files and are not repeated in agent responses.
[DATA_EXFILTRATION]: Network activity is restricted to SSH connections to remote host machines as defined by the user. No unauthorized data exfiltration patterns were detected.
[PROMPT_INJECTION]: The skill facilitates the creation of agents that process untrusted data from messaging channels and local knowledge files. Evidence chain: 1. Ingestion points: Incoming messages from messaging channels and knowledge directory files. 2. Boundary markers: Mandates 'Security Hardening' prompt blocks in generated workspace files. 3. Capability inventory: Agents can be configured with tools like read, write, or exec. 4. Sanitization: Provides templates for refusal and identity anchoring to mitigate risks from malicious inputs.

openclaw-agent-builder