AI-Native CLI Requirements

Purpose

Define a clear, testable baseline for AI-native CLI design so teams can implement it and verify it with Sonde.

To be considered AI-native ready in Sonde, a CLI should expose:

manifest command
- Returns the CLI's Sondage manifest contract in machine-readable form.
Machine-readable output mode
- --json (preferred) or equivalent (--output json, --format json).
Non-interactive execution mode
- --non-interactive (preferred) or equivalent (--no-input, --yes, --assume-yes).
Safe planning mode for mutating actions
- --dry-run (preferred) or equivalent preview/plan mode.
Response-size control
- --fields (preferred) or equivalent selective output projection.

Make JSON output deterministic and stable across repeated runs.
Keep prompts out of automation paths when non-interactive mode is active.
Keep command/flag help output structured and consistent for parser tooling.
Reject unsafe inputs early (for example control chars, path traversal, malformed IDs).
Prefer full-payload machine inputs and explicit schemas over ambiguous flag combinations.

Sonde uses score-only compliance semantics.
Missing capabilities reduce score and appear in report metrics[*].details notes.
Sonde does not hard-fail solely because a required AI-native capability is missing.

sonde generate <cli> --json
sonde run <cli> --json
sonde score <cli> --json

Look for the AI-native readiness metric in score output and resolve each MISS: note.