Latch

Latch is a software tool that evaluates public GitHub repositories for readiness to be operated by AI coding agents. It runs a command-line readout against a given repository and produces a score on a 600-point scale broken down across six categories including machine orientation and reproducible setup. The service is aimed at developers who want to know whether an autonomous AI agent can read, build, and safely modify a given codebase.

To get started, users run the command 'latch grip github.com/owner/repo' in their terminal. The tool reads the repository the way an AI agent must, top to bottom, and reports a score per signal plus the precise fixes that raise it. This process provides a readiness readout rather than a general vibe check.

The tool scores six specific signals on a 0–100 scale, including machine orientation, reproducible setup, code navigability, verification surface, agent safety, and context economy. Each signal receives a one-line readout that does not round up, along with specific fixes to improve the score. These categories ensure the assessment covers how well an agent can grasp the repo and work within it.

Latch operates on a 600-point scale where each of the six signals is scored individually from 0 to 100. The final output includes receipts for every signal, providing a detailed breakdown rather than a single aggregate number. This granular scoring helps developers understand exactly which areas need improvement for AI agent readiness.

The service is currently free while in beta, allowing developers to test the tool without cost. Users can run free-tier runs on repositories like 'sindresorhus/p-limit' or 'trpc/trpc' to see the tool in action. This pricing model encourages widespread adoption during the early development phase.

The tool specifically targets developers who need to verify if an autonomous AI agent can safely read, build, and modify a codebase. It is designed for those concerned about secret hygiene, prompt-injection surfaces, and clear 'don't touch' boundaries within their repositories. The focus is on ensuring agent safety and context economy for automated workflows.

Latch provides a one-line assessment for each signal along with specific fixes that would raise the score. For example, it might identify missing pinned versions or lockfiles under the reproducible setup category. This feedback loop allows developers to close the gap between current state and AI-ready standards.

The tool includes a specific signal called 'context economy' which checks if the signal an agent needs fits its effective context window. It also evaluates 'verification surface' to ensure agents can check their work against ground truth using runnable tests and type-checking. These features help prevent blind editing and ensure deterministic builds.

You can reach Latch by e-mail at hi@latch.fyi.

The official website of Latch is latch.fyi.

Latch has no reviews on Trustburn yet. Anyone who has used it can write the first review on its Trustburn page.