Sonde: AI-native CLI manifest and tooling.
Capture how your CLI should behave in one machine-readable manifest, validate it with deterministic checks, and publish scored reports. Agents, CI, and other integrations can target the same contract instead of reverse-engineering every binary.
Quick start
Adopt the manifest and publish comparable reports.
npx @sonde-sh/sonde generate cli-good --json
npx @sonde-sh/sonde score cli-good --json
npx @sonde-sh/sonde publish cli-good --jsonWhat good looks like
A strong manifest is explicit about command behavior, machine-readable output, and non-interactive mode. That is what lets LLM tooling use a CLI reliably across environments.
{
"name": "my-cli",
"version": "1.4.0",
"commands": [
{
"name": "scan",
"description": "Run repository scan",
"supportsJson": true,
"nonInteractive": true
}
],
"output": {
"schemaVersion": "1.0.0",
"formats": ["json"]
}
}Adoption creates leverage
Better LLM interoperability
Fewer retries and fewer hidden assumptions when external tools invoke your CLI.
Safer releases
Manifest checks surface behavior drift before integrations break in downstream repos.
Clear integration contract
Integrators consume known capabilities instead of reverse engineering prompt/output behavior.
Shared quality bar
Contributors validate against one standard, not one-off scripts.
How Sonde helps you implement it
Use Sonde to go from initial manifest draft to verified compatibility in local dev and CI.
1Run validation while implementing commands
2Generate report artifacts for CI
3Publish results in the report explorer
Implementation flow
sonde generate <cli> --json
sonde run <cli> --json
sonde score <cli> --jsonIntegration guide