Reach a 100 Sonde Score

Purpose

Define a practical and repeatable process to push a CLI toward a 100/100 Sonde score.

Inputs

• A target CLI with stable --help output.
• Support (or equivalent) for machine-readable and non-interactive execution.
• Sonde CLI installed or available via npx.

Outputs

• Stable manifest and run results.
• Score output at or near 100, with actionable deltas if below 100.

Reproducible workflow

1) Generate a fresh manifest

sonde generate <cli> --json

2) Validate deterministic run behavior

sonde run <cli> --json

Review these fields in result:

• ok should be true
• exitCode should be 0
• interactiveDetected should be false
• usedPreferences should include JSON and non-interactive flags when available

3) Validate manifest contract exposure

sonde manifest --json

Use this output as the reference for what target CLIs should expose via their own manifest command.

4) Score with repeat runs

sonde score <cli> --json

Target:

{
  "ok": true,
  "apiVersion": "1.0.0",
  "command": "score",
  "result": {
    "manifestVersion": "1.0.0",
    "generatedAt": "2026-03-05T00:00:00.000Z",
    "total": 100
  }
}

What usually prevents 100/100

• CLI output changes across runs.
• Prompt-like output appears in stdout/stderr.
• Non-zero exits in baseline or repeat run.
• Missing machine-readable output support.
• Missing AI-native capabilities (manifest, non-interactive, dry-run, field selection).

Improving score systematically

• Prefer stable output formats (--json, --format json, or equivalent).
• Add explicit non-interactive mode (--non-interactive, --yes, etc.).
• Add a manifest command and expose required capability flags in help output.
• Add --dry-run and --fields (or equivalent) for safer agent execution.
• Keep --help and command metadata deterministic.
• Avoid timestamps/random values in default output paths checked by Sonde.

Versioning and compatibility

• Ensure manifest version remains on supported major version 1 while upgrading Sonde.
• Use manifestVersion to enforce compatibility and generatedAt to track report freshness.

Edge cases

• run and score currently execute manifest.cli.binary --help; ensure that path is deterministic first.
• Passing <cli> to run/score does not override manifest.cli.binary.
• AI-native capability gaps are score penalties with remediation notes (not hard failures).

See also

• CLI Reference
• AI-Native CLI Requirements
• Quickstart
• Troubleshooting