e2e-test

name	e2e-test
description	Run agent-browser based E2E tests against the worktree-isolated stack brought up by dev-up. Reads .env.dispatch for URLs and uses --session <project>-<slot> so each worktree has its own browser context (cookies, storage, history). Scenarios are bash files in .e2e-scenarios/ written dynamically by the child Claude per ticket. Run `install` once to install agent-browser globally.
argument-hint	[install|run <scenario-name>|snapshot|teardown]

e2e-test

A thin wrapper around agent-browser that pairs with dev-up to run E2E tests against worktree-isolated stacks.

Subcommands

Subcommand	Description
install	npm i -g agent-browser && agent-browser install. Run once per machine.
run <scenario-name>	Source .env.dispatch, export AGENT_BROWSER_SESSION=<project>-<slot> and RESULTS_DIR=.e2e-results/<scenario-name>/, then execute .e2e-scenarios/<scenario-name>.sh. scenario-name is a single token without / or ..
snapshot [url]	Take an accessibility snapshot (optionally navigate to a URL first).
teardown	Close the agent-browser session for this worktree.

How to invoke

bash <skill-dir>/scripts/install.sh
bash <skill-dir>/scripts/run.sh login-flow
bash <skill-dir>/scripts/snapshot.sh "http://localhost:$VITE_PORT/dashboard"
bash <skill-dir>/scripts/teardown.sh

Scenario file template

Scenarios live in <worktree-root>/.e2e-scenarios/<name>.sh. They are written dynamically by the child Claude per ticket. The skill is project-agnostic.

Environment available to the scenario:

• AGENT_BROWSER_SESSION = <project>-<slot>
• SLOT, PROJECT, WORKTREE_ROOT
• RESULTS_DIR = <worktree-root>/.e2e-results/<scenario-name>/ (auto-mkdir'd)
• All *_PORT env vars from .env.dispatch

Example .e2e-scenarios/login-flow.sh:

#!/usr/bin/env bash
set -euo pipefail

agent-browser --session "$AGENT_BROWSER_SESSION" open "http://localhost:$VITE_PORT/login"
agent-browser --session "$AGENT_BROWSER_SESSION" snapshot --json > "$RESULTS_DIR/01-login.json"
agent-browser --session "$AGENT_BROWSER_SESSION" fill --ref e3 "[e-mail]"
agent-browser --session "$AGENT_BROWSER_SESSION" fill --ref e4 "password"
agent-browser --session "$AGENT_BROWSER_SESSION" click --ref e5
agent-browser --session "$AGENT_BROWSER_SESSION" wait --url-contains "/dashboard" --timeout 10
agent-browser --session "$AGENT_BROWSER_SESSION" screenshot --path "$RESULTS_DIR/02-dashboard.png"

The child Claude then writes $RESULTS_DIR/report.md summarizing what happened.

Parallel session isolation

Each worktree gets its own browser context via --session <project>-<slot>. Cookies, localStorage, history, and tabs are fully isolated between sessions, so N worktrees can run E2E tests concurrently without interfering.

Failure modes

Situation	Behavior
.env.dispatch missing	Exit 1, hint to run dev-up up first.
agent-browser not installed	Exit 1, hint to run install.
Scenario name contains / or .	Exit 2, scenario names are single tokens.
Scenario file missing	Exit 1, hint with the resolved path.