The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to search and process local files (including logs and configuration files) using tools like ripgrep and ls. This creates a standard indirect prompt injection surface where malicious instructions embedded in local files could attempt to influence the agent's output during the evidence-gathering phase.
Ingestion points: File search results (rg), directory listings (ls), and log files (logs/).
Boundary markers: The skill employs a strict response framework with dedicated headers (e.g., 'Claims to Verify', 'Evidence') to separate verification metadata from the final answer.
Capability inventory: File system inspection (ls, stat), text searching (rg), and platform status checks (openclaw gateway status). No write operations or network tools are utilized.
Sanitization: No explicit instruction-based sanitization is present; however, the structured 'Verified/Inferred/Unknown' classification system naturally encourages the agent to treat data as evidence rather than instructions.
[COMMAND_EXECUTION]: The skill generates and uses standard system commands such as ls, stat, and ripgrep (rg) for local environment inspection. This behavior is consistent with the skill's primary purpose of system state verification and does not involve the execution of untrusted remote code.

truth-first