Vibe Coding DiscoverDashClaw
View on GitHubRemote approvals, policy checks, and execution evidence for unattended AI agents.
Self-hosted governance layer that intercepts AI agent tool calls before they run, routing destructive or expensive actions to remote human approvals and recording signed decision evidence. Integrates via Claude Code/Codex/Hermes hooks, an OpenClaw gateway, an MCP server, and Node/Python SDKs.
Use Cases
Approve or deny destructive agent tool calls remotelyRun coding agents unattended without unconditional authorityBlock rm -rf, DROP TABLE, or force-push before executionMaintain a signed audit ledger of agent decisionsShare one policy across Claude Code, Codex, Hermes, and MCPScan agent actions for prompt-injection overridesIssue scoped standing grants for repeated actionsCalibrate interruption rate from approve/deny verdictsOversee CI or background agent fleetsSelf-host an agent governance and approval instance
Built With
- Language
- TypeScript
- Frameworks
- Next.js · Node.js · Model Context Protocol · LangChain · AutoGen · CrewAI · Claude Code · Codex · Hermes · OpenClaw · Python
Tags
agent-governance · approvals · human-in-the-loop · policy-engine · mcp · mcp-server · hooks · claude-code · codex · audit-ledger · agent-runtime · prompt-injection · unattended-agents · self-hosted · security · ai-ops