Features

CLI-first design

Every workflow runs through the blop CLI:

Command	Behavior
blop test	Discover specs, run them, write reports, exit non-zero on any failure.
blop run	Same as test.
blop watch	Run, then re-run on file change.
blop list	Load specs and print discovered test names. No browser launched.
blop init	Scaffold a starter spec.
blop skills	Manage bundled SKILL.md files — list, view, check, improve.

See the CLI reference for every flag and option.

Intent-based test DSL

Three exports cover every spec file. Imported from "blop":

• agentTest(name, handler) — register a test inside a describe block.
• describe(name, fn) — create a named test scope. Nest as deep as you like.
• defineAgentTest(test) — object-form test definition for generated specs.

Inside a test handler, the agent API is just two methods:

• agent.goto(url) — queue a page navigation.
• agent.goal(text) — describe a verification step in natural language.

No beforeEach/afterEach, no assertion functions in the DSL itself — assertions are delegated to the agent’s browser tools. See Writing tests.

Real Playwright browser

The agent uses Chromium, Camoufox, Firefox, or WebKit through controlled native tools — no script execution, no shell escapes. Pick a browser with --browser or in config. Camoufox requires Node.js 22 or newer; local runs need a provisioned binary, while --containerized builds and reuses a separate warm Camoufox server container.

$ blop test --browser firefox --headed

Camoufox setup

Camoufox JS is experimental and must be installed explicitly. Blop pins the Playwright Core version used by the audited integration:

pnpm add -D camoufox-js@0.11.1 playwright-core@1.61.1
pnpm exec camoufox-js fetch
pnpm exec camoufox-js version
blop test --browser camoufox

Camoufox can run locally or in a separate warm container with --containerized. It does not provide Chromium CDP frame streaming, so Blop falls back to normal Playwright screenshots. The local upstream fetch command follows a mutable prerelease; reproducible CI should provision a verified binary directory and set CAMOUFOX_INSTALL_DIR. The final benchmark used 150.0.2-alpha.26.

Camoufox’s optional dependency tree includes differently licensed packages, including ua-parser-js@2.0.10 under AGPL-3.0-or-later. Review distribution and service-use obligations before enabling it outside local evaluation.

Browser tools catalog

The agent has a curated set of Playwright-backed tools across navigation, page inspection, mouse, keyboard, forms, assertions, data extraction, diagnostics, batched execution, and lifecycle. You don’t call them directly — the agent picks the right tool for your goal — but knowing the catalog helps you write actionable goals. See Browser tools.

Reporters & output

Every run writes a structured artifact bundle to .blop/:

File	Format	When
results.json	JSON	Always
events.jsonl	JSONL	Always
report.xml	JUnit XML	--reporter junit or all
screenshots/*	PNG	Captured during the run

Choose a reporter with --reporter: basic, json, junit, or all (default). See Reporters & output.

Provider-agnostic

OpenAI, Anthropic, Google, Groq, xAI, OpenRouter, Mistral, Cerebras, NVIDIA via Khadim. Configure via CLI flags, env vars, or blop.config.ts. See Agent providers.

Configuration

blop.config.ts (or .mts, .js, .mjs) with full TypeScript types. Every option is optional; the runner has sensible defaults. CLI flags always override config values. See Configuration.

Skills

Bundled SKILL.md files teach the agent domain-specific patterns. Manage them with blop skills list, view, check, and improve. See Skills.

CI-native

Designed CI-first. JUnit XML, JSON results, and a JSONL event stream any pipeline can pick up. GitHub Actions metadata auto-captured. See CI integration.

Keyless testing

For unit testing the runner or CI smoke checks of the framework itself, pass a mock agentStream to runBlopTests — no API keys required, no network calls. See Writing tests → Keyless testing.