This PR fixes CI build failures by addressing code formatting and linting issues that were causing the builds to fail.
**Changes made:**
1. **Rust formatting fixes in `cli/src/commands.rs`:**
- Removed unnecessary multi-line formatting for clipboard operations
- Applied consistent single-line formatting for return statements
- Fixed line length and formatting for the `test_wait_text_with_timeout` test function
2. **TypeScript fixes in `src/actions.ts`:**
- Fixed `waitForFunction` usage in the `handleWait` function by replacing the function parameter approach with a string-based implementation
- Properly escaped the text parameter using `JSON.stringify` to prevent potential injection issues
These changes ensure the code passes linting checks (clippy for Rust, ESLint for TypeScript) and formatting validation (rustfmt, prettier) that are enforced in the CI pipeline.
Fixes#751
* feat: add screenshot output config, clipboard CLI commands, and fix wait --text native path
## Summary
- Add `--screenshot-dir`, `--screenshot-quality`, and `--screenshot-format` CLI flags (with corresponding `AGENT_BROWSER_SCREENSHOT_DIR`, `AGENT_BROWSER_SCREENSHOT_QUALITY`, `AGENT_BROWSER_SCREENSHOT_FORMAT` env vars) so users can configure where and how screenshots are saved without specifying a full path every time
- Add `clipboard read`, `clipboard write <text>`, `clipboard copy`, and `clipboard paste` CLI commands, exposing the existing protocol-level clipboard handlers that were previously only accessible via JSON-RPC
- Fix `wait --text` in native mode: the CLI was emitting `selector: "text=..."` (a Playwright-style locator) which native's `querySelector` can't handle. Now emits a `text` field that correctly hits the native `wait_for_text` polling path
- Add native clipboard `copy` and `paste` support via CDP `Input.dispatchKeyEvent`, and a `write` operation to the Node.js handler
* fix: resolve CI failures in Rust formatting and TypeScript typecheck
Use string-based page.evaluate for clipboard writeText to avoid
referencing `navigator` in Node.js compilation context. Run cargo fmt
to fix formatting in commands.rs and screenshot.rs.
* fix: clipboard write captures full multi-word text
Use rest[1..].join(" ") instead of rest.get(1) so unquoted multi-word
input like `clipboard write hello world` sends the full string rather
than silently dropping everything after the first word.
* improvements
* fixes
* improvements
* improvements
Root cause: package.json `main` pointed at `dist/daemon.js` (the
internal daemon process), so programmatic consumers of the package
received the daemon module instead of a usable API. Additionally,
`BrowserManager.launch()` required IPC-only fields (`id`, `action`),
and `navigate()` existed only as a private function inside actions.ts.
Changes:
- Add src/index.ts as the public package entry point
- Add BrowserLaunchOptions type (Pick<LaunchCommand> minus id/action/engine)
to decouple the programmatic API from the IPC wire protocol
- Change launch() signature from LaunchCommand to BrowserLaunchOptions
- Add BrowserManager.navigate(url, options?) — consolidates domain check,
scoped-header setup, and page.goto() into one reusable method; auto-
recovers a new page when all pages have been closed (stale session)
- Add BrowserManager.getUrl() and getTitle() convenience methods
- Update package.json: main → ./dist/index.js, add types and exports["."]
- Add tests: navigate() with headers, waitUntil, allowedDomains blocking,
allowedDomains allow, non-http(s) scheme blocking, getUrl/getTitle, and
compile-time type assertions verifying the public entry exports the right
API surface (direct repro of #307)
Fixes#307
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix HTML retrieval by using browser.getLocator() for selector operations
This PR fixes an issue where HTML content retrieval was not working properly when using selectors.
**Problem:**
The `get html` command and other selector-based operations were failing because they were using `page.locator()` directly instead of the browser manager's locator method.
**Changes:**
- Updated `handleContent()` to use `browser.getLocator()` instead of `page.locator()` for HTML retrieval with selectors
- Applied the same fix to other affected functions: `handleCount()`, `handleBoundingBox()`, `handleInnerText()`, `handleInnerHtml()`, and `handleSetValue()`
- Ensures consistent locator handling across all selector-based operations
**Implementation Details:**
The fix replaces direct `page.locator(command.selector)` calls with `browser.getLocator(command.selector)` to ensure proper element selection and interaction through the browser manager's abstraction layer.
Fixes#735
* Fix remaining page.locator() calls to use browser.getLocator()
Apply the same fix to all remaining functions that were using
page.locator(command.selector) directly instead of going through
browser.getLocator(): handleWheel, handleHighlight, handleClear,
handleSelectAll, handleDispatch, handleNth, handleMultiSelect,
and handleDiffScreenshot.
---------
Co-authored-by: ctate <366502+ctate@users.noreply.github.com>
* feat: Add browserless as a hosted option + boolean env-parsing utility
* Add ensureDomainFilter, sanitizeExistingPage and move parseBooleanParam
* Add docs in relevant places, fix utils, rename of API env var
* Update readme
* Fix env variable name in readme
* Cleanup session stop urls when errors happen
* Fix browserlessStopUrl not being assigned in happy path
* feat: add idle timeout to daemon to prevent orphaned Chrome processes
The daemon persists indefinitely after browser sessions are used,
leaving orphaned Chromium processes consuming memory and CPU.
Add a configurable idle timeout (default 15 minutes) that shuts down
the daemon when no commands arrive. Resets on every incoming command,
so active sessions are unaffected.
Set AGENT_BROWSER_IDLE_TIMEOUT_MS=0 to disable (preserves old behavior).
Fixes#721
* fix: save session state before shutdown to prevent silent data loss
The shutdown() function (used by idle timeout, SIGINT, SIGTERM, SIGHUP)
previously closed the browser without saving state, unlike the explicit
`close` command which calls saveStateToFile(). This meant idle timeouts
silently destroyed cookies, localStorage, and login sessions.
Now shutdown() mirrors the close command's auto-save behavior: it calls
saveStateToFile() before manager.close(), preserving session state to
disk. This makes idle timeout functionally equivalent to an explicit
close — users returning after an idle shutdown get their state restored.
Addresses review feedback on #722 by @ctate.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Max Stoel <maxalerator@hotmail.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The client (connection.rs) and native daemon (native/daemon.rs) used
different get_port_for_session() implementations on Windows:
- Client: i32, .chars(), djb2 — (hash << 5) - hash + c
- Daemon: i64, .bytes(), Java hashCode — hash * 31 + b
For session name "default", client computes port 50838 while the
daemon binds on 51174, causing a 5-second timeout and startup failure.
Fix: align native/daemon.rs to use the identical djb2 algorithm from
connection.rs (i32, chars, djb2), so both sides agree on the port.
Unix is unaffected (uses Unix domain sockets, no port hashing).
Tests: add port hash regression tests to all three implementations
(native/daemon.rs, connection.rs, daemon.ts) to prevent future drift.
Fixes#705
The handleGetText function now retrieves text using innerText, falling back to textContent if innerText is not available. This change enhances the accuracy of text extraction from elements.
Co-authored-by: Honglei Wu <honglei.wu@shopee.com>
When launched with --extension or --profile, launchPersistentContext()
is used which sets isPersistentContext=true but leaves this.browser as
null. The guard in newTab() checked !this.browser, causing a false
"Browser not launched" error even though the browser was running.
Replace !this.browser with !this.isLaunched(), which already accounts
for both launch paths (browser !== null || isPersistentContext).
Also improve the error message in newWindow() to clarify that it is
not supported in persistent context mode, since it requires a Browser
object to create a new context.
Fixes#411
* fix: isolate getEncryptionKey tests from local filesystem
Tests for getEncryptionKey() failed on machines where
~/.agent-browser/.encryption-key existed, because the file-based
fallback was not mocked out. Mock node:fs to isolate both env var
and key file paths, and add missing tests for the file fallback.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: clean up fs mock naming in encryption tests
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: hyunjinee <leehj0110@kakao.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Remove BROWSERBASE_PROJECT_ID requirement
The Browserbase API no longer requires a project ID to create sessions —
it is inferred from the API key. Remove the env var requirement from both
the TypeScript daemon and Rust CLI, and update docs accordingly.
* Remove unnecessary Content-Type header since no body is sent
Browserbase has no DELETE endpoint for sessions. The correct API is
POST /v1/sessions/:id with body { status: "REQUEST_RELEASE" }. The old
DELETE call returned an error that was silently swallowed, causing every
session to leak until the 30-min idle timeout.
Fixed in both Node.js (src/browser.ts) and native Rust
(cli/src/native/providers.rs) paths.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The connectViaCDP and connectToBrowserbase methods hardcoded
context.setDefaultTimeout(10000), ignoring the AGENT_BROWSER_DEFAULT_TIMEOUT
env var. This made page.goto time out after 10s on CDP connections even when
the env var was set to a higher value. Now both paths use getDefaultTimeout()
like all other connection modes.
Fixes#703
* fix: sanitize lone Unicode surrogates in snapshot and response serialization (#635)
Pages with emoji/special characters can contain lone surrogates (e.g. \uD800
without a matching \uDC00-\uDFFF), causing serde_json to fail with
"unexpected end of hex escape" when parsing the JSON response.
- Add sanitizeSurrogates() to replace lone surrogates with U+FFFD in
ariaSnapshot output
- Add sanitizeJsonSurrogates() safety net in serializeResponse for other
response fields (page.title, page.content, etc.)
- Add tests for both sanitization paths
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: remove trivial "no surrogates unchanged" test
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* ci: retry flaky Rust test
* refactor: remove unnecessary sanitizeSurrogates from snapshot.ts
Chromium's ariaSnapshot() converts lone surrogates to literal text
(e.g. the 6-char string "\ud800"), not actual surrogate code points.
The real fix is sanitizeJsonSurrogates() in protocol.ts which handles
eval and other response paths where actual surrogates appear.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: use toWellFormed() instead of regex for lone surrogate sanitization
Upgrade tsconfig target/lib from ES2022 to ES2024 and replace the
manual regex-based surrogate sanitization with String.prototype.toWellFormed().
This is simpler, more readable, and relies on the standard API.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: remove trivial no-surrogate test
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: hyunjinee <leehj0110@kakao.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Use 127.0.0.1 instead of localhost when constructing CDP URL from port
number, since Chrome only binds to IPv4. This prevents connection
failures on systems like Ubuntu 24.04 where localhost resolves to ::1.
Co-authored-by: hyunjinee <leehj0110@kakao.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Make KERNEL_API_KEY optional for external credential injection
When running inside environments with external credential injection
(e.g. Vercel Sandbox credentials brokering), the KERNEL_API_KEY env
var can be omitted. The network layer injects the Authorization header
on outbound requests to api.onkernel.com, so the API key never needs
to exist inside the sandbox.
If KERNEL_API_KEY is set, it's used as before. If not, requests are
sent without an Authorization header, allowing external injection.
Without either, the Kernel API returns 401.
Made-with: Cursor
* Make KERNEL_API_KEY optional in native Rust daemon too
Applies the same change to the native Rust connect_kernel() function
so both the Node.js and native code paths support external credential
injection.
Made-with: Cursor
* Address review feedback: fix type errors, cargo fmt, always send cleanup DELETE
- Fix kernelApiKey assignment: use ?? null for undefined -> null
- Fix closeKernelSession signature: accept string | undefined
- Always send DELETE on cleanup even without local API key (external
injection covers it)
- Run cargo fmt on Rust code
Made-with: Cursor
Per POSIX, kill(pid, 0) returns EPERM when the process exists but the
caller lacks permission to signal it, and ESRCH when it does not exist.
The daemon liveness checks in both the Rust CLI and TypeScript daemon
treated any kill failure as "not running", which is incorrect when
running inside a macOS sandbox that restricts signal delivery to
(target self). This caused the CLI to delete the real daemon's socket
and PID files, then spawn a duplicate daemon.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix: resolve unnamed element refs matching multiple elements (#500)
When a page has one unnamed button among several named buttons,
clicking its ref fails with "matched N elements" because the
locator `getByRole('button')` matches all buttons on the page.
Normalize unnamed interactive elements to `name: ""` so the
selector becomes `getByRole('button', { name: "", exact: true })`
which matches only buttons with empty accessible names.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: remove dead code branch in buildSelector
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: make RefMap.name required string, remove dead code branches
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: hyunjinee <leehj0110@kakao.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* add security hardening features
- Add authentication vault (`auth save/login/list/show/delete`) so credentials are stored locally and never exposed to the LLM (fixes Snyk W007)
- Add `--content-boundaries` flag to wrap page-sourced output in structural markers, helping LLMs distinguish tool output from untrusted page content (fixes Snyk W011)
- Add `--allowed-domains` flag to restrict browser navigation to trusted domains
- Add `--action-policy` for static allow/deny gating of action categories, with opt-in `--confirm-actions`/`--confirm-interactive` for orchestrator or human-in-the-loop confirmation
- Add `--max-output` flag to truncate large page outputs, preventing context flooding
- New docs page at /security, updated README, SKILL.md, CLI help text, and templates
* fixes
* fixes
* fixes
* fixes
* fixes
* fixes
* fixes
* docs
* add --selector flag to scroll command
The `scroll` command uses `window.scrollBy()`, which has no effect on apps
that use custom scrollable containers (e.g. a nested div with overflow-y: auto).
The backend `handleScroll` already supports a `selector` parameter, but the CLI
never exposed it. This adds `-s` / `--selector` to the `scroll` command so users
can target a specific scrollable element:
agent-browser scroll down 500 --selector "div.scroll-container"
Also fixes the backend to apply `direction`/`amount` when a selector is present
(previously those fields were only used in the no-selector branch).
Closes#501
* fixes
* feat: add --download-path option
Adds a `--download-path` flag (and `AGENT_BROWSER_DOWNLOAD_PATH` env / `downloadPath` config key) to set a default download directory for browser downloads.
Without this, Playwright stores downloads in a temp directory that is deleted when the browser closes. The new option passes through to Playwright's `downloadsPath` on `launch()` and `launchPersistentContext()`.
Fixes#507
* improvements
* fixes
* fixes
- Add AGENT_BROWSER_DEFAULT_TIMEOUT env var to override Playwright's
default 60s timeout (CDP/recording 10s timeouts unaffected)
- Add backpressure-aware safeWrite() that waits for drain when socket
buffer is full, preventing data loss under load
- Serialize command execution per socket via queue to prevent concurrent
writes that cause buffer contention
These daemon-side fixes complement #329 (CLI-side EAGAIN retry) by
addressing the root causes: Playwright operations that outlast the
CLI's IPC timeout, and concurrent socket.write() calls that overflow
the kernel buffer.
Tested with heavy React app (1000+ DOM nodes) — 10 consecutive
snapshot commands complete without os error 35/11.
Refs #322
Co-authored-by: shohu <shohu@users.noreply.github.com>
Adds `keyboard type` and `keyboard insertText` subcommands that
operate on the currently focused element without requiring a selector.
Essential for contenteditable editors (Lexical, ProseMirror, CodeMirror,
Monaco) where `type <selector>` doesn't trigger the editor's internal
event pipeline (beforeinput/DOM mutation).
- `keyboard type <text>` — page.keyboard.type() with real keystrokes
- `keyboard insertText <text>` — page.keyboard.insertText()
Note: `keyboard press` intentionally omitted — the existing top-level
`press` command already operates on current focus.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Fixes#519. Playwright defaults `colorScheme` to `light` on all new contexts, overriding the browser/OS dark mode setting. This is especially disruptive in CDP mode, where every reconnection resets the scheme. The `set media dark` command also didn't persist its choice to new tabs or pages.
- Add `--color-scheme <dark|light|no-preference>` flag, config key (`colorScheme`), and env var (`AGENT_BROWSER_COLOR_SCHEME`)
- Store the preference in `BrowserManager` and automatically apply it to all new contexts (via Playwright's context option) and all new pages (via `page.emulateMedia` in `setupPageTracking`)
- `set media dark/light` now also persists its choice for subsequent pages and tabs
`state load` always fails with "Cannot load state while browser is running" even when no browser is running, making the command completely unusable (#526).
The daemon's auto-launch logic starts a browser before `state_load` gets to handle the command. This adds `state_load` to the exclusion list alongside `launch` and `close`, so `handleStateLoad` can perform its own launch with the state file.
* feat: Enable capture of profiling data
Adding a new set of commands:
```
agent-browser profiler start
agent-browser profiler stop trace.json
```
With this, agents can start a profiling trace, perform a set of actions, and then extract the profiling data for analysis.
**Note:** I was originally going to call it `agent-browser profile` but I realized that might cause confusion with the `--profile` flag
CDP supports a couple commands for starting/stopping a trace.
When a trace is running, it emits events that need to be picked up.
We store these locally in the daemon until the trace is completed.
When the final event is received, we dump all of them into an output file.
That file can be loaded directly into chrome devtools or another analysis tool to visualize what happened during the agentic run.
Added some basic rust tests for parsing the commands (since they have some optional / required args)
TS daemon adds ~6 tests to make sure the profiling lifecycle (including saving the output file) works as intended
* add docs
* fixes
* fixes
---------
Co-authored-by: Chris Tate <chris@ctate.dev>
Fixes#491
When `--start-maximized` or `--window-size` is passed as a browser arg, Playwright's default viewport (1280x720) overrides the browser's own window sizing, making those flags have no effect on the page content.
This change auto-detects those args and sets `viewport: null` so Playwright defers to the browser's window size. Explicit viewport values still take priority.
Also allows `viewport: null` in the launch protocol for agents that want to disable viewport emulation directly.
## Summary
- Return a `ParseError` when `--headers` receives invalid JSON instead of silently dropping the headers and proceeding
- Reject `frame` commands that provide no `selector`, `name`, or `url` (previously returned `{ switched: true }` without doing anything)
- Add missing mutual exclusion check for `--cdp` + `--extension` (extensions require a local browser, not a CDP connection)
## Summary
- Fix `allowFileAccess` being silently stripped from launch commands by adding it to the Zod schema in `protocol.ts` (the `--allow-file-access` CLI flag was not reaching the browser)
- Fix `trace stop` requiring a path argument despite help text documenting it as optional -- now works with or without a path
- Fix `addscript`/`addstyle` silently succeeding when neither `content` nor `url` is provided -- now returns a validation error
- Replace hardcoded ANSI escape code with `color::error_indicator()` in `main.rs` to respect `NO_COLOR`
- Fix double-parse pattern and add descriptive expect messages in `commands.rs`
- Fix incomplete string escaping in `snapshot.ts` `buildSelector` (use `JSON.stringify` instead of manual quote escaping)
- Simplify redundant ternary in `snapshot.ts` cursor-interactive role assignment
- Sync docs changelog with CHANGELOG.md (v0.8.1 through v0.10.0)
Three fixes to eliminate duplicate entries:
1. Skip elements that only inherit cursor:pointer from a parent
(the parent element is captured instead)
2. Broaden dedup by extracting all quoted text from the ARIA tree,
not just ref names
3. Add accepted cursor elements to the dedup set to prevent
multiple DOM elements with the same text from duplicating
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Rebased and fixed implementation of PR #184 features on current main:
Session persistence:
- --session-name flag and AGENT_BROWSER_SESSION_NAME env var auto-save/restore
cookies and localStorage across browser restarts
- State files stored in ~/.agent-browser/sessions/ with owner-only permissions
- AES-256-GCM encryption via AGENT_BROWSER_ENCRYPTION_KEY env var
- Auto-expiration of old state files (AGENT_BROWSER_STATE_EXPIRE_DAYS, default 30)
State management commands:
- state list: list saved state files with metadata
- state show <file>: display state summary (cookies, origins, domains)
- state rename <old> <new>: rename state files
- state clear [name] [--all]: clear saved states
- state clean --older-than <days>: delete expired states
New --new-tab flag for click command:
- Opens link href in a new tab instead of navigating the current tab
Security hardening:
- Session name validation prevents path traversal (CLI + daemon)
- safeHeaderMerge prevents prototype pollution in header merging
- WebSocket stream server binds to 127.0.0.1 only
- State files written with 0o600 permissions
Fixes applied over the original PR:
- Use color.rs module instead of hardcoded ANSI escape codes
- Align CLI output field names with daemon response format
- Add CLI-level --session-name validation (not just daemon-side)
- Avoid adding "DOM" to tsconfig.json lib (use proper typing in evaluate)
- Keep version at 0.9.3 (matches current main)
- Centralize session name validation in daemon.ts helper
- Update all documentation (README, SKILL.md, docs site, --help output)
Co-authored-by: Chris Tate <chris@ctate.dev>
The `-C`/`--cursor` flag was added to the CLI parser and snapshot
implementation in #374, but the Zod schema in protocol.ts was not
updated. This caused the `cursor` field to be silently stripped
during command validation, so cursor-interactive element detection
never ran.
Fixes#434
Update `setupContextTracking` in `BrowserManager` to auto-switch `activePageIndex` to newly opened tabs and invalidate the CDP session accordingly. This mirrors what `newTab()` and `newWindow()` already do for explicitly created tabs, and aligns CLI behavior with how real browsers focus newly opened tabs.
Fixes#384
Summary
- The `--exact` flag on `find role`, `find label`, and `find placeholder` was accepted by the CLI but silently dropped by the server. The Zod validation schema, TypeScript types, and action handlers all lacked the `exact` field, so it was stripped before reaching Playwright's `getByRole`, `getByLabel`, and `getByPlaceholder` calls.
- Added `exact` to the schema, types, and handler for all three locators so the flag is forwarded to Playwright as intended.
- Added tests confirming `exact: true` survives protocol parsing for `getbyrole`, `getbylabel`, and `getbyplaceholder`.
Fixes#402
* feat: add --allow-file-access flag for file:// URL support
Adds the ability to open and interact with local files using file:// URLs.
This enables use cases like viewing local PDFs, testing local HTML files,
and allowing JavaScript to access other local files via XHR.
The flag adds Chromium's --allow-file-access-from-files and --allow-file-access
launch arguments. Only supported in Chromium browsers.
Fixes#345
* fix: add cli_allow_file_access tracking to prevent spurious warning
When --allow-file-access is set via AGENT_BROWSER_ALLOW_FILE_ACCESS env var
(not CLI), don't warn about the flag being ignored when daemon is already running.
* fix: only warn about ignored flags when explicitly passed via CLI
The warning about launch-time options being ignored (when daemon is
already running) was incorrectly shown when options were set via
environment variables like AGENT_BROWSER_EXECUTABLE_PATH, even when
no CLI flag was passed.
Now the warning only appears when flags are explicitly passed on the
command line, not when values come solely from environment variables.
Fixes#372
* feat: add cursor-interactive element detection in snapshots
Add -C/--cursor flag to snapshot command that detects clickable elements
that don't have proper ARIA roles but are interactive based on:
- cursor: pointer CSS style
- onclick attribute/handler
- tabindex attribute
This helps with modern web apps that use custom divs/spans as buttons.
Fixes#366
* fix: add cursor option to getSnapshot type signature
Chris Tate