Safe Action

Before any destructive or irreversible action, run a safety pre-flight — check risks, reversibility, and timing.

Before installing or enabling this skill: (1) Verify the endpoints (think.agentutil.net, undo.agentutil.net, context.agentutil.net) and the operator's privacy/data-retention policy; ask how paid checks are authenticated and billed. (2) Never include secrets, credential strings, or large sensitive documents in the 'description' field — the instructions do not sanitize inputs. (3) Test on non-production actions first to confirm what is sent and returned. (4) Consider restricting autonomous invocation (or require manual confirmation) until you are confident the service will not log or retain sensitive metadata. (5) Ask the publisher for an explicit data-handling contract (what fields are stored, for how long, and how payments are handled) before using this in high-stakes workflows.

Type: OpenClaw Skill Name: safe-action Version: 1.0.1 The 'safe-action' skill (SKILL.md) instructs the agent to send metadata about high-stakes operations—such as database deletions, infrastructure changes, and deployments—to external APIs at agentutil.net. While presented as a safety pre-flight check, this pattern encourages the exfiltration of operational context and action descriptions to a third-party service. The reliance on external endpoints for security decision-making and the mention of a crypto-based payment mechanism (USDC on Base) represent a non-standard and potentially risky architectural choice for a safety utility.