The Agent Skills Directory

[DATA_EXFILTRATION]: The skill manages the OPENCLAW_MACHINE_SECRET generated during onboarding. It transmits this secret to the vendor's API (api.unclawg.com) for verification and stores it in local identity files under .claude/agents/ for discovery by other skills. It also opens URLs in the browser containing authentication tokens for CLI-to-portal handoff.
[COMMAND_EXECUTION]: The skill performs multiple shell operations including curl for API interactions, jq for parsing responses, and file system commands for directory and file creation. It requires the uc_onboard binary to be present in the environment for execution.
[PROMPT_INJECTION]: The skill implements an 'Indirect Prompt Injection' surface (Category 8) by using a 'soul interview' process to capture user input and generate a SOUL.md instruction file that governs the agent's future persona and behavior.
Ingestion points: Collects user input regarding agent goals, audience demographics, and persona characteristics.
Boundary markers: The resulting soul draft includes explicit 'Non-Negotiable Rules' and 'Escalation Policies' to set operational constraints.
Capability inventory: The skill utilizes curl, shell execution, and delegates research tasks to a 'Task' subagent tool.
Sanitization: Content is JSON-encoded using python3 before being submitted to the platform API.

unclawg-internet