rysweet/amplihack-rs/e2e-outside-in-test-generator

E2E Outside-In Test Generator

Purpose

Generate runnable end-to-end tests from the user's point of view. The bundled helper creates a starter outside-in plan or seed scenario; the agent then adapts that output to the repository's existing test framework and validates it with normal test commands.

Supported App Types

Workflow

1. Detect the app type from package manifests, routes, binaries, docs, and existing tests, or run the helper.
2. Identify 3-8 critical user journeys and expected outcomes.
3. Generate tests in the repository's existing e2e/test location, or e2e/ if there is no convention.
4. Prefer real user entry points over mocked internals.
5. Run the relevant existing test command and fix failures caused by the new tests.
6. Report any product bugs found separately from test harness problems.

Output Requirements

• Tests must be runnable by existing tooling.
• Generated assertions must validate observable behavior, not implementation details.
• Do not add a new test framework unless the repository has no suitable e2e path.
• Do not silently skip journeys; document unsupported journeys and why.

tmux in tests and sandboxes: pass -S <private socket> (or use env -u TMUX tmux -S <private socket>) on every tmux command. When $TMUX is set, as it is inside any tmux pane, tmux uses that socket and ignores TMUX_TMPDIR, so TMUX_TMPDIR=… tmux kill-server kills the user's own server. Never run kill-server or kill-session -a without -S/-L; a PreToolUse hook denies them.

Native Helper

# Inspect detected app type and journey plan
amplifier-bundle/skills/e2e-outside-in-test-generator/scripts/generate-e2e.sh plan
 
# Generate a starter scenario/spec
amplifier-bundle/skills/e2e-outside-in-test-generator/scripts/generate-e2e.sh generate --type web --output e2e
amplifier-bundle/skills/e2e-outside-in-test-generator/scripts/generate-e2e.sh generate --type cli --output e2e

The helper intentionally creates starter artifacts only. The agent must adapt the scenarios to real project commands, routes, and assertions before claiming completion.