# Boundary & Risk Card

Project: promptfoo/promptfoo

## Doramagic Trial Decision

Current decision: ready for pre-publication recommendation checks. First use should still start with least privilege, a temporary directory, and rollback.

## What The User Can Do Now

- Read the Human Manual first to understand purpose and main workflows.
- Copy the Prompt Preview for a pre-install trial. This checks interaction feel, not real execution.
- Test the official Quick Start command in an isolated environment before using a primary machine.

## What Not To Do Yet

- Do not treat Prompt Preview output as an actual project run.
- Do not treat metadata-only validation as sandbox install validation.
- Do not write unverified capabilities as supported, tested, or safe to install.
- Do not provide production data, private files, real secrets, or primary configuration directories on first use.

## Pre-install Checklist

- Host AI match: claude, chatgpt
- Official install entry state: official entry found
- Verification location: temporary directory, temporary host, or container required
- Rollback readiness: required
- API keys, network access, file writes, or host configuration changes: treat as high risk until confirmed
- Install command, actual output, and failure logs: must be recorded

## Current Blockers

- No blockers.

## Project-specific Pitfalls

- Installation risk requires verification (medium): Upgrade or migration may change expected behavior: 0.121.8
- Installation risk requires verification (medium): Upgrade or migration may change expected behavior: code-scan-action: 0.1.6
- Configuration risk requires verification (medium): May increase setup, validation, or first-run risk for the user.
- Configuration risk requires verification (medium): Upgrade or migration may change expected behavior: 0.121.15
- Configuration risk requires verification (medium): Developers may misconfigure credentials, environment, or host setup: Per-test-case `repeat` option to control how many times individual tests run

## Risk And Permission Notes

- no_demo: medium

## Evidence Gaps

- No structured evidence gaps found.