A process supervisor that force-stops runaway OpenClaw agents when they ignore halt commands
An OpenClaw agent executed 515 tool calls after receiving a stop command. Context compression silently drops safety instructions, enabling completely uncontrolled agent behavior. There is no reliable way to halt an agent that has gone rogue. The stop button in the UI sends a signal the agent can ignore. This tool implements a kill switch that operates below the agent layer, forcibly terminating processes, revoking API tokens, and blocking network access within milliseconds of activation regardless of what the agent is doing.
Demand Breakdown
Social Proof 4 sources
Gap Assessment
2 tools exist (SkillFortify, SecureClaw tool-boundary module) but gaps remain: Pre-execution verification only, cannot stop a running agent, no kill switch, no rate limiting, no runtime enforcement; Per-action approval, no global kill switch, no rate limiting across sessions, no safety instruction persistence.
Features3 agent-ready prompts
Competitive LandscapeFREE
| Product | Does | Missing |
|---|---|---|
| SkillFortify | Formal verification for AI agent skills, validates tool-boundary contracts before execution | Pre-execution verification only, cannot stop a running agent, no kill switch, no rate limiting, no runtime enforcement |
| SecureClaw tool-boundary module | External enforcement layer evaluating every tool action returning allow/block/require_approval | Per-action approval, no global kill switch, no rate limiting across sessions, no safety instruction persistence |
Sign in to unlock full access.