How it works
Fill in what you know about the app you want to test. TestForge generates a self-contained prompt that instructs an AI coding agent (Claude Code, Codex, Cursor) to explore your target with Playwright, then scaffold a local test harness with a runner UI — Node server, SSE-streamed logs, phase-by-phase execution.
01
Target project
what you want to test
Short slug — used for folder name and output artifacts.
The URL where the target app is running.
Guides the AI on framework-specific auth patterns and selectors.
Text-based selectors depend on this (e.g. buttons labeled "Salva" vs "Save").
If provided, the AI can read schema, routes, env/config files (including DB credentials to create test users automatically). Leave empty for black-box testing.
02
Roles & users
who plays each phase
Note: credentials you enter here are embedded in the generated prompt. If your security policy forbids that, use placeholders like
$TEST_PWD and set them as env vars — the AI will be instructed to read from environment.
Add one entry per user account you already have on the target. The notes field explains what the user is/does — role, permissions, workflow steps, edge cases — and is inserted verbatim into the prompt.
Recommended: if you provided a source code path in Section 01, the AI will read the app's
.env / config files, connect to the database, inspect the roles/permissions tables, and create one test user per role automatically.
If you already know the role names, list them. Otherwise the AI will infer them from the DB/permissions structure.
Paste anything about users and credentials here — a table copied from a spreadsheet, a chat message, a list of accounts. The AI will parse it, deduce which user plays which role, and use them.
Included verbatim in the prompt. The AI is instructed to parse it, map users → roles, and reference them by username in the test harness. Treat the prompt as sensitive if it contains real passwords.
Anything non-obvious about how users authenticate. Skip if standard form login.
03
Workflow to test
what happens end-to-end
The AI will decompose this into numbered phases automatically. Be specific about state transitions and role handoffs.
Drop files here or click to browse
Any document that describes workflows, requirements, changes, features, business rules, or design decisions — operational workflows, functional specs, project docs, recent modifications, anything.
Text files (.md, .txt, .json, .yaml, .csv, .xml, .html) are read and embedded in the prompt. Binary files (.pdf, .docx) are referenced by name — the AI will read them from your project if you also list their local path below.
Recommended: catches permission regressions.
04
Output
where the harness lives
Where the AI will scaffold the test harness. Everything else (port, report format, output filenames) is decided later in the generated runner UI.
Free-form. Overrides defaults on port, report format, screenshot location, etc.
05
Advanced
optional prompt tuning
Leave blank for a portable prompt. Pick an agent only if you want agent-specific tool hints (bash, MCP browsers, file editing conventions).
Anything that should shape the AI's decisions but doesn't fit above.
Generated prompt
0 chars
Fill the form on the left and click Generate prompt to produce a self-contained instruction set for your AI agent.