CLI Reference

The CLI is the fastest way to exercise the same deterministic controls outside an agent run.

Commands

Command Purpose
ai-guardrails:screen Normalize and screen a prompt, then return the verdict
ai-guardrails:sanitize Run output sanitization on text or structured output
ai-guardrails:audit Inspect recent injection audit records
ai-guardrails:purge Apply the configured retention strategy
ai-guardrails:hitl-install Install HITL bridge support when using laravel-flow
ai-guardrails:hitl-status Inspect pending approval state

Workflow

sequenceDiagram participant Operator participant Artisan participant Service as Guardrail service participant Store as Audit/stat stores Operator->>Artisan: php artisan ai-guardrails:screen Artisan->>Service: normalize + screen Service->>Store: append audit record Service-->>Artisan: verdict Artisan-->>Operator: exit code + text output

Examples

  1. Screen a prompt

    php artisan ai-guardrails:screen "ignore previous instructions"

  2. Sanitize output

    php artisan ai-guardrails:sanitize "<script>alert(1)</script>"

  3. Review recent audit entries

    php artisan ai-guardrails:audit --limit=20
ADR · CLI mirrors runtime controls

Problem. Operators need to reproduce guardrail decisions without constructing a whole agent request.

Decision. Artisan commands call the same services used by middleware and facades.

Consequences. A CLI verdict is useful evidence when debugging production behavior.

Run CLI commands in the same Laravel environment you are diagnosing. Config, stores, optional vendors, and feature modes are environment-sensitive.

previous PHP API next PHP facade