DashClaw

The product is a fail-closed approval layer that intercepts destructive coding agent actions before they run and holds them for human approval with one-click decisions from anywhere. It solves the problem of unattended long-running agent workflows where native permission prompts cannot help because they require the user at the keyboard. It sells to developers, IT and operations teams in B2B organizations. It is positioned against native permission prompts that protect only at-keyboard sessions and observability tools that record after execution, by enforcing policy before execution and allowing approval later, delivered as self-hosted MIT-licensed software with a hosted trial option.

Key features

• Intercept destructive agent actions before execution
• Fail-closed approval hold for risky actions
• One-click approval from anywhere
• Policy enforcement at hook seam
• Risk scoring for agent actions
• Signed replayable audit ledger with Ed25519
• Cross-runtime unified policy enforcement
• Governance via Node and Python SDKs
• Governance via MCP server and REST API
• Support for Claude Code, Codex, Hermes hooks