Understand the Sonde Model

Sonde defines a clear model for building and evaluating AI-native CLI behavior in repeatable, machine-readable scenarios.

Purpose

Explain Sonde's deterministic model for developing and evaluating AI-native CLI behavior.

Inputs

CLI help/command behavior.
Scenario metadata captured in the Sondage manifest contract.
Runtime outputs from deterministic command execution.

Outputs

Structured observations (stdout, stderr, parsed JSON when available).
Determinism signals (interactive detection, non-zero exits).
Scoring-ready run results.

Key elements

State transitions

Model deterministic execution stages for each CLI scenario.

Observations

Capture stdout, stderr, parsed JSON payloads, and prompt interactions.

Scoring

Compute comparable totals from explicit weighted criteria. Sonde includes AI-native readiness scoring for manifest command support and required automation flags.

When to use

Use this model when you need deterministic test assertions across different CLIs, environments, or CI runs without relying on manual interpretation.

Edge cases

Deterministic checks degrade when interactive prompts are detected.
Equivalent flags across CLIs may map to different option names and must be captured in manifest options.