@verist/cli

CLI for Verist workflows — replay, diff, and inspect AI decisions.

Install

bun add @verist/cli

Commands

verist init

Scaffold a working project with no API keys needed.

verist init

Creates verist.config.ts (or .mjs for Node) with a parse-contact step and sample input.

verist capture

Run a step against input files and save baselines.

verist capture --step extract --input "inputs/*.json"
verist capture --step extract --input "inputs/*.json" --label "v2-prompt"
verist capture --step extract --input "inputs/*.json" --sample 10 --seed 42

| Option | Description | | ------------------ | -------------------------------------------- | | --step <name> | Step name to execute (required) | | --input <glob> | Glob pattern for input JSON files (required) | | --workflow <id> | Workflow identifier (defaults to step name) | | --version <ver> | Workflow version (defaults to "0.0.0") | | --label <text> | Human-readable label for the baseline | | --no-commands | Skip capturing commands | | --sample <n> | Randomly sample n inputs from the glob | | --seed <n> | Seed for deterministic sampling (default: 0) | | --meta <key=val> | Attach metadata (repeatable) |

verist diff

Recompute baselines and show diffs (exploratory).

verist diff --step extract
verist diff --baseline .verist/baselines/extract/001.json
verist diff --step extract --format json

| Option | Description | | ------------------- | -------------------------------------------- | | --step <name> | Step name to recompute | | --baseline <path> | Path to specific baseline file or directory | | --workflow <id> | Workflow identifier for auto-resolution | | --version <ver> | Workflow version for auto-resolution | | --label <name> | Filter by metadata label | | --format <mode> | Output format: text, json, or markdown | | --meta <key=val> | Filter baselines by metadata (repeatable) |

verist replay

Inspect baselines and verify hash integrity.

verist replay --step extract
verist replay --step extract --verify
verist replay --baseline .verist/baselines/ --label "v2-prompt"

Use --verify to recompute hashes and check that stored content matches.

| Option | Description | | ------------------- | ----------------------------------------- | | --step <name> | Filter by step name | | --baseline <path> | Specific file or directory | | --label <name> | Filter by metadata label | | --workflow <id> | Workflow identifier | | --version <ver> | Workflow version | | --verify | Recompute hashes and check integrity | | --meta <key=val> | Filter baselines by metadata (repeatable) |

verist test

Recompute baselines and fail on regressions (CI mode).

Exit codes: 0 = clean, 1 = regressions detected, 2 = infrastructure failure.

Schema violations always trigger exit 1, independent of --no-fail-on-diff.

verist test --step extract
verist test --step extract --no-fail-on-diff
verist test --step extract --format json

| Option | Description | | ---------------------------- | -------------------------------------------- | | --step <name> | Step name to recompute | | --baseline <path> | Path to specific baseline file or directory | | --workflow <id> | Workflow identifier for auto-resolution | | --version <ver> | Workflow version for auto-resolution | | --label <name> | Filter by metadata label | | --format <mode> | Output format: text, json, or markdown | | --meta <key=val> | Filter baselines by metadata (repeatable) | | --no-fail-on-diff | Exit 0 even when value diffs are detected | | --no-fail-on-commands-diff | Ignore command diffs for exit code |

Global Options

• --debug — show full error details
• --quiet — suppress non-essential output

License

Apache-2.0