is-cli-agentic

The benchmark for agent-friendly CLIs

Built for humans.
Ready for agents?

We put developer CLIs in a clean sandbox and let an agent take the wheel. See what works, where it stalls, and the evidence behind every score.

evaluation / walkthroughIllustrative demo
prisma01 / 04
Example command: npm install --global prisma
Install the CLI in an isolated environment.CLI → evidence
Chain of evidence01 Command02 Capture03 FindingEvery claim has a source.
A clean start. A real task. An inspectable result.

THE PUBLIC CATALOG

Find your next CLI.

Compare the interface. Inspect the evidence.
CLI 11 in the catalogProtocol checksAgent outcomeReport
auth0saas
0/10 measured
Not run
awscloud
0/10 measured
Not run
firebasesaas
0/10 measured
Not run
ghvcs
0/10 measured
Not run
neondata
0/10 measured
Not run
prismadata
0/10 measured
Not run
sentry-cliobs
0/10 measured
Not run
stripesaas
0/10 measured
Not run
supabasecloud
0/10 measured
Not run
vercelcloud
0/10 measured
Not run
wranglercloud
0/10 measured
Not run

Protocol checks describe sampled commands. Agent outcomes apply to a specific task. These are experimental measurements, not overall CLI ratings.

Read the methodology ↗

Don't see a CLI? Request one.

No setup. No login. We’ll review it and run a Floor eval.

We’ll review it and run a Floor eval from the operator console.