Gatekeeper adds consent layer for agents

Sharp idea, and the timing is right: a lot of agent stacks still treat plugin trust as a policy note instead of a hard boundary.

• –The repo calls out a real gap in frameworks that let plugins run with the same OS privileges as the host process.
• –The demo keeps the UX simple: the agent proposes an action, the user approves or rejects it, and the decision is logged.
• –The strongest part of the pitch is concrete control over irreversible actions like file deletes, messages, and API calls.
• –It is clearly a demo, not production infrastructure, so the big question is whether this consent loop can stay usable at scale.
• –The idea lands well for developers building agent tooling, because it shifts “trust” from advice into an actual interaction model.