Skill Detail

Gate risky agent actions behind approval checkpoints with Latch

Use Latch to put an MCP policy and approval layer between agents and tools so risky calls pause for review while safe calls continue automatically.

Security & VerificationMCP
Security & Verification MCP Security Reviewed
⭐ 8 GitHub stars
INSTALL WITH ANY AGENT
npx skills add agentskillexchange/skills --skill gate-risky-agent-actions-behind-approval-checkpoints-with-latch Copy
Works best when you want a reusable capability, not another fragile one-off prompt.
At a glance
Tools required
Docker, Latch CLI, an upstream MCP server to wrap
Install & setup
Clone the repository and start Latch with docker compose, create an account and API key, then run the Latch CLI in front of an upstream MCP server using the documented –upstream and –upstream-command flags.
Author
Latch
Publisher
Organization
Last updated
Apr 20, 2026
Quick brief

Use Latch when an agent already has MCP tool access and you need a concrete control point that classifies actions as allowed, approval-required, or denied before the upstream tool call executes. The upstream project is explicit here: Latch is an MCP proxy with a policy engine, approval workflow, audit log, and action classes such as READ, EXECUTE, SEND, and TRANSFER_VALUE.

How it works

What this skill actually does

Invoke this instead of using the product normally when the real task is not “run an MCP server” but “wrap an MCP server with review gates for risky actions”. The operator workflow is specific: launch Latch, point it at an upstream MCP server, define policy, inspect blocked or approval-bound actions, then resume or deny the exact tool call.

The scope boundary is clear and prevents this from being a plain product card. This is not publishing Latch as a generic platform or dashboard. It is the narrowly bounded skill of enforcing human approval checkpoints and policy decisions on MCP-mediated agent actions.