Show HN: Multicorn Shield – Open-source permissions and approvals for AI agents

I'm a backend engineer on Atlassian's Rovo Agents team. A few weeks ago, OpenAI acquired the OpenClaw project and I started thinking about what happens when agents get broad access to your data with no permission layer. Six days later, Summer Yue (Director of Alignment at Meta) posted about her OpenClaw agent deleting 200+ emails while ignoring her stop commands. The root cause was context window compaction dropping her safety instruction. The agent kept working. It just lost the part where it was supposed to ask first. Shield is a plugin that hooks into OpenClaw's tool system and intercepts every action before it executes. Permissions are enforced outside the model's context window, so they can't be compressed away. You set what each agent can access (read/write/execute per service), and anything outside those boundaries gets blocked or routed through an approval workflow with time-limited grants. TypeScript, open source, MIT licensed. The plugin, dashboard, and docs are all live. Happy to answer questions about the architecture or how the OpenClaw Plugin API integration works.