Replace browser backend with agent-browser

.gitignore

@@ -1,5 +1,6 @@
 # Binary
-bin/
+/bin/
+/browser/bin/agent-browser-*
 cdp
 
 # Go

.goreleaser.yml

@@ -2,6 +2,10 @@ version: 2
 
 project_name: tap
 
+before:
+  hooks:
+    - scripts/prepare-agent-browser.sh
+
 builds:
   - main: ./cmd/tap
     binary: tap
@@ -14,6 +18,9 @@ builds:
     goarch:
       - amd64
       - arm64
+    ignore:
+      - goos: windows
+        goarch: arm64
     ldflags:
       - -s -w -X main.version={{.Version}}
 

AGENTS.md

@@ -2,11 +2,11 @@
 
 ## Project
 
-Go CLI and library for running JS scripts against websites (QuickJS + Chrome CDP fallback) and extracting clean content from URLs via go-defuddle.
+Go CLI and library for running JS scripts against websites (QuickJS + agent-browser fallback) and extracting clean content from URLs via go-defuddle.
 
 ## Stack
 
-Go 1.26+, urfave/cli v3, QuickJS (fastschema/qjs), chromedp, go-defuddle, mise.
+Go 1.26+, urfave/cli v3, QuickJS (fastschema/qjs), agent-browser, go-defuddle, mise.
 
 ## Commands
 
@@ -35,8 +35,8 @@ Emoji-prefixed Conventional Commits: `✨ feat:`, `🐛 fix:`, `♻️ refactor:
 
 ```
 tap.go / options.go   → Client API + functional options
-transport/            → Shared HTTP + CDP browser layer
-browser/              → Persistent sessions, tabs, network interception
+transport/            → Shared HTTP + agent-browser bridge
+browser/              → agent-browser adapter, binary install, pass-through types
 engine/               → QuickJS + browser fallback
 fetch/                → URL → clean content (go-defuddle)
 cmd/tap/              → CLI (site, fetch, sync, browser)

CHANGELOG.md

@@ -7,6 +7,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/).
 
 ## [Unreleased]
 
+### Changed
+
+- **Replaced browser backend with embedded agent-browser** — removed ~4,000 lines of hand-rolled chromedp/CDP code and switched to [agent-browser](https://github.com/vercel-labs/agent-browser) as the sole browser backend. Tap embeds the native agent-browser binary for supported release platforms and lets agent-browser manage Chrome/browser installation.
+- **Removed chromedp dependency** — `go.mod` no longer depends on `chromedp/chromedp`, `chromedp/cdproto`, or `chromedp/sysutil`.
+
+### Added
+
+- **New browser pass-through commands** — `tap browser set` (viewport, device, geo, offline, headers, credentials, media), `tap browser storage` (local/session), `tap browser state` (save/load/list/show/clear), `tap browser auth` (save/login/list/show/delete), `tap browser get` (text/html/value/attr/title/url/count/box/styles), `tap browser vitals`, `tap browser diff` (snapshot/screenshot/url).
+- **Global Lightpanda engine flag** — `--lightpanda` / `--lp` selects Lightpanda as the agent-browser engine for browser-backed commands.
+
 ## [0.4.8] - 2026-05-18
 
 ### Added

README.md

@@ -20,7 +20,7 @@ Upgrade later with:
 tap upgrade
 ```
 
-Browser features use Chrome by default. Check dependencies with `tap doctor`.
+Browser features use the embedded agent-browser backend. Run `tap doctor --install` to let agent-browser install browser dependencies.
 
 ## Quick start
 
@@ -54,7 +54,7 @@ tap fetch -b https://github.com/notifications
 
 ### Reuse your existing Chrome
 
-Chrome must already expose DevTools.
+agent-browser manages Chrome automatically. To attach to an existing Chrome with DevTools enabled:
 
 ```bash
 tap attach chrome
@@ -70,7 +70,6 @@ You can also attach explicitly:
 
 ```bash
 tap attach chrome --browser-url http://127.0.0.1:9222
-tap attach chrome --port-file ~/Library/Application\ Support/Google/Chrome/DevToolsActivePort
 ```
 
 ### Browser workflow
@@ -79,7 +78,7 @@ tap attach chrome --port-file ~/Library/Application\ Support/Google/Chrome/DevTo
 tap browser open https://news.ycombinator.com
 tap browser open https://github.com --new-tab
 tap browser tabs
-tap browser switch tab-2
+tap browser switch t2
 tap browser screenshot --output github.png
 tap browser status
 ```
@@ -152,9 +151,9 @@ These show up on the relevant commands instead of only in global help:
 | `--wait-selector` | Wait for a CSS selector |
 | `--wait-js` | Wait for a JS expression |
 | `--timeout` | Set execution timeout |
-| `--browser-url` | One-shot DevTools override |
+| `--browser-url` | One-shot agent-browser/DevTools connection override |
 | `--profile-dir` | One-shot profile override |
-| `--lightpanda`, `--lp` | Use Lightpanda instead of Chrome |
+| `--lightpanda`, `--lp` | Use Lightpanda as the browser engine |
 
 Compatibility aliases still work:
 - `--ws-url` -> `--browser-url`
@@ -171,16 +170,18 @@ tap browser snapshot
 tap browser forms
 tap browser cookies ...
 tap browser network ...
+tap browser set ...
+tap browser storage ...
+tap browser state ...
+tap browser auth ...
+tap browser get ...
+tap browser vitals
+tap browser diff ...
 ```
 
-## Lightpanda
+## Browser backend
 
-| Backend | Platforms | Best for |
-| --- | --- | --- |
-| Chrome | macOS, Linux, Windows | Full browser automation, auth, network interception |
-| Lightpanda | macOS, Linux | Fast headless rendering without auth-heavy flows |
-
-Install or update Lightpanda with:
+Tap embeds agent-browser as the single browser backend. Chrome is the default engine; pass `--lightpanda`/`--lp` to use Lightpanda for fast browser-backed rendering. agent-browser manages browser installation for full automation, auth, screenshots, and network workflows. Install browser dependencies with:
 
 ```bash
 tap doctor --install

browser/agentbrowser.go

@@ -0,0 +1,140 @@
+package browser
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"os/exec"
+	"slices"
+	"strings"
+)
+
+const DefaultAgentBrowserSession = "default"
+
+type AgentBrowser struct {
+	Path        string
+	SessionName string
+	ProfileDir  string
+	Headed      bool
+	Attached    bool
+	Engine      string
+}
+
+type OpenOpts struct {
+	Headed     bool
+	Headers    map[string]string
+	InitScript string
+}
+
+type ExecResult struct {
+	Stdout json.RawMessage
+	Stderr string
+}
+
+func NewAgentBrowser(path string) (*AgentBrowser, error) {
+	if path == "" {
+		resolved, err := ResolveAgentBrowserPath()
+		if err != nil {
+			return nil, err
+		}
+		path = resolved
+	}
+	return &AgentBrowser{Path: path, SessionName: DefaultAgentBrowserSession}, nil
+}
+
+func (a *AgentBrowser) Exec(ctx context.Context, args ...string) (json.RawMessage, string, error) {
+	return a.exec(ctx, nil, args...)
+}
+
+func (a *AgentBrowser) exec(ctx context.Context, stdin []byte, args ...string) (json.RawMessage, string, error) {
+	cmdArgs := a.commandArgs(args...)
+	cmd := exec.CommandContext(ctx, a.Path, cmdArgs...)
+	if stdin != nil {
+		cmd.Stdin = bytes.NewReader(stdin)
+	}
+	var stdout, stderr bytes.Buffer
+	cmd.Stdout = &stdout
+	cmd.Stderr = &stderr
+	if err := cmd.Run(); err != nil {
+		return nil, stderr.String(), fmt.Errorf("agent-browser %s: %w: %s", strings.Join(cmdArgs, " "), err, strings.TrimSpace(stderr.String()))
+	}
+	return json.RawMessage(bytes.TrimSpace(stdout.Bytes())), stderr.String(), nil
+}
+
+func (a *AgentBrowser) commandArgs(args ...string) []string {
+	out := make([]string, 0, len(args)+8)
+	out = append(out, args...)
+	if !slices.Contains(out, "--json") {
+		out = append(out, "--json")
+	}
+	if !a.Attached && a.SessionName != "" && !hasFlag(out, "--session-name") {
+		out = append(out, "--session-name", a.SessionName)
+	}
+	if a.ProfileDir != "" && !hasFlag(out, "--profile") {
+		out = append(out, "--profile", a.ProfileDir)
+	}
+	if a.Headed && !slices.Contains(out, "--headed") {
+		out = append(out, "--headed")
+	}
+	if a.Engine != "" && !hasFlag(out, "--engine") {
+		out = append(out, "--engine", a.Engine)
+	}
+	return out
+}
+
+func hasFlag(args []string, flag string) bool {
+	for _, arg := range args {
+		if arg == flag || strings.HasPrefix(arg, flag+"=") {
+			return true
+		}
+	}
+	return false
+}
+
+func (a *AgentBrowser) Open(ctx context.Context, url string, opts OpenOpts) error {
+	args := []string{"open", url}
+	if opts.Headed {
+		args = append(args, "--headed")
+	}
+	if opts.InitScript != "" {
+		args = append(args, "--init-script", opts.InitScript)
+	}
+	for name, value := range opts.Headers {
+		args = append(args, "--headers", name+": "+value)
+	}
+	_, _, err := a.Exec(ctx, args...)
+	return err
+}
+
+func (a *AgentBrowser) Eval(ctx context.Context, js string) (any, error) {
+	out, stderr, err := a.exec(ctx, []byte(js), "eval", "--stdin")
+	if err != nil {
+		return nil, err
+	}
+	var envelope AgentBrowserEnvelope[map[string]any]
+	if err := json.Unmarshal(out, &envelope); err == nil && envelope.Success {
+		return envelope.Data["result"], nil
+	}
+	var value any
+	if err := json.Unmarshal(out, &value); err != nil {
+		return nil, fmt.Errorf("parse eval JSON: %w: %s", err, stderr)
+	}
+	return value, nil
+}
+
+func (a *AgentBrowser) GetHTML(ctx context.Context) (string, error) {
+	value, err := a.Eval(ctx, "document.documentElement.outerHTML")
+	if err != nil {
+		return "", err
+	}
+	if s, ok := value.(string); ok {
+		return s, nil
+	}
+	return "", fmt.Errorf("agent-browser html result is %T, not string", value)
+}
+
+func (a *AgentBrowser) Close(ctx context.Context) error {
+	_, _, err := a.Exec(ctx, "close")
+	return err
+}

browser/agentbrowser_test.go

@@ -0,0 +1,55 @@
+package browser
+
+import (
+	"slices"
+	"testing"
+)
+
+func TestAgentBrowserCommandArgs(t *testing.T) {
+	ab := &AgentBrowser{Path: "agent-browser", SessionName: "dev", ProfileDir: "/tmp/profile", Headed: true}
+	args := ab.commandArgs("open", "https://example.com")
+	for _, want := range []string{"open", "https://example.com", "--json", "--session-name", "dev", "--profile", "/tmp/profile", "--headed"} {
+		if !slices.Contains(args, want) {
+			t.Fatalf("args %v missing %q", args, want)
+		}
+	}
+}
+
+func TestAgentBrowserCommandArgsIncludesEngine(t *testing.T) {
+	ab := &AgentBrowser{Path: "agent-browser", SessionName: "dev", Engine: "lightpanda"}
+	args := ab.commandArgs("open", "https://example.com")
+	for _, want := range []string{"--engine", "lightpanda"} {
+		if !slices.Contains(args, want) {
+			t.Fatalf("args %v missing %q", args, want)
+		}
+	}
+}
+
+func TestAgentBrowserCommandArgsAttachedSkipsSession(t *testing.T) {
+	ab := &AgentBrowser{Path: "agent-browser", SessionName: "dev", Attached: true}
+	args := ab.commandArgs("get", "url")
+	if slices.Contains(args, "--session-name") {
+		t.Fatalf("attached args included session name: %v", args)
+	}
+}
+
+func TestAgentBrowserCommandArgsPreservesExplicitJSONAndSession(t *testing.T) {
+	ab := &AgentBrowser{Path: "agent-browser", SessionName: "dev"}
+	args := ab.commandArgs("session", "--json", "--session-name", "other")
+	if got := count(args, "--json"); got != 1 {
+		t.Fatalf("--json count = %d, want 1 in %v", got, args)
+	}
+	if got := count(args, "--session-name"); got != 1 {
+		t.Fatalf("--session-name count = %d, want 1 in %v", got, args)
+	}
+}
+
+func count(values []string, target string) int {
+	var n int
+	for _, value := range values {
+		if value == target {
+			n++
+		}
+	}
+	return n
+}