diff options
| author | Aman Kumar <145339995+amankumarsingh77@users.noreply.github.com> | 2026-07-25 00:23:05 +0530 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2026-07-25 00:23:05 +0530 |
| commit | 6dd79f7c3bafacba54acb2840f8655fcb05318bf (patch) | |
| tree | 1570768785ae836717d3dbb2ff42fe0a13f19585 /src/profile/bind.ts | |
| parent | 0fc1ece6fa4a2f24e87eb4a66eb828a9928e30f7 (diff) | |
feat: pin the agent to a chosen browser profile (#16)
* feat: pin the agent to a chosen browser profile (#15)
Harness tabs were created with a bare Target.createTarget, which places
them in Chrome's defaultBrowserContextId — and that value follows window
focus. Which profile the agent acted as therefore depended on which
browser window the user last clicked, so the same task could run as a
work account on one run and a personal account on the next.
/browser-profile lists the connected browser's profiles as "Name (email)"
and pins one to ~/.pi/agent/browser-harness.json, so the choice outlives
the session. The same picker appears on first browser_setup when nothing
is pinned; with no interactive UI, or on cancel, setup continues with a
note and the previous behavior.
Chrome offers no direct route to another profile: Target.createTarget
rejects a foreign browserContextId, and openerId does not inherit the
opener's context. The window is opened through the browser's own command
line (--profile-directory, handed to the running browser by Chromium's
ProcessSingleton) and identified by a unique file:// sentinel page, since
a bare about:blank#token argument is discarded. Further tabs come from
window.open evaluated with userGesture: true — the only CDP-reachable way
to place a tab in a non-default browser context — correlated by a unique
token so concurrent isolated tabs stay distinct. A failed handshake is
reported, never silently redirected to another profile.
All four tab-creation sites now funnel through cdp/target-factory, so
open_urls and the isolated tabs behind web_search / read_page cannot run
with a different profile's cookies than the visible tabs.
Also fixed along the way: snap Chromium is discovered, Windows uses
%LOCALAPPDATA%, browsers launched with an explicit --user-data-dir are
found, the running browser is chosen by ranking candidates against the
connected user-data-dir rather than taking the first process found, and
an installed-but-closed browser no longer reads as running.
CI runs the profile suites and a real-browser end-to-end test on ubuntu,
macOS, and Windows — profile discovery is the one part of the harness
whose behavior genuinely differs per OS.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix: single-flight harness window creation
Isolated tabs run concurrently (browser_read_page is registered
unserialized), so several could each find "no harness window" and create
one — two blank windows before this feature, two LAUNCHED profile windows
after it. ensureHarnessWindow now shares one in-flight creation per
client.
Sharing introduces a second hazard: freshlyCreated is a claim on the seed
tab, and if every waiter received it, they would all drive the same tab.
Only the caller that started the creation keeps the claim; the rest open
their own tab beside it.
Covered by test/manual/profile-concurrency-test.ts (five concurrent
isolated tabs, pinned and unpinned: distinct targets, one window, correct
profile), wired into CI on Linux only — the coordination is pure JS and
identical across platforms.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix: tilde expansion differs by purpose on Windows
CI on windows-latest caught this: join("~", "x") yields "~\x" there, and
expandTilde only handled "~/".
The two uses are not the same, so they no longer share a helper. The
agent dir stays byte-compatible with pi's own expandTildePath ("~/" only,
every platform) — being more lenient would put the pin file somewhere pi
never looks. $CHROME_USER_DATA_DIR is ours to interpret, so it accepts
"~\" as well, which is what a Windows user would type.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Diffstat (limited to 'src/profile/bind.ts')
| -rw-r--r-- | src/profile/bind.ts | 138 |
1 files changed, 138 insertions, 0 deletions
diff --git a/src/profile/bind.ts b/src/profile/bind.ts new file mode 100644 index 0000000..aeb59d3 --- /dev/null +++ b/src/profile/bind.ts @@ -0,0 +1,138 @@ +/** + * Binding the harness session to a pinned profile. + * + * Sequence, all of it verified against a live browser: + * 1. resolve the browser binary and any explicit --user-data-dir + * 2. launch a window into the profile, marked with a file:// sentinel + * 3. poll Target.getTargets for the sentinel page — its browserContextId IS + * the profile's context, which is otherwise undiscoverable (CDP exposes no + * profile identity) + * 4. adopt that page as the harness window, blank it, drop the sentinel file + * + * If the sentinel never appears the bind FAILS LOUDLY. Falling through to a + * normal createTarget would put the agent in the focus-derived default profile + * — the exact bug this feature exists to remove — so the caller gets an + * actionable error instead. + * + * The resolved context id is session-scoped state, never persisted: Chrome + * mints new context ids on every browser run. + */ + +import type { BrowserClient } from "../client"; +import { type CdpError, cdpError } from "../cdp/errors"; +import { type Result, err, ok } from "../util/result"; +import { detectRunningBrowser, resolveBrowserExecutable } from "./browser-process"; +import { openProfileWindow } from "./launch"; +import type { ProfilePin } from "./store"; + +/** The launch delegates through ProcessSingleton, so allow for a cold profile. */ +const SEED_DEADLINE_MS = 15_000; +const SEED_POLL_MS = 500; + +type PageTarget = { + readonly targetId: string; + readonly type: string; + readonly url: string; + readonly browserContextId?: string; +}; + +const pageTargets = async (client: BrowserClient): Promise<ReadonlyArray<PageTarget>> => { + const r = await client.session().callBrowser("Target.getTargets"); + if (!r.success) return []; + const data = r.data as { targetInfos: ReadonlyArray<PageTarget> }; + return data.targetInfos.filter((t) => t.type === "page"); +}; + +const manualFallbackHint = (pin: ProfilePin): string => + `open a window for "${pin.label}" yourself (browser profile menu → ${pin.profileDir}), then retry. ` + + `Run /browser-profile to pick a different profile, or clear the selection to use whichever window is focused.`; + +/** + * Open (and adopt) a harness window inside the pinned profile. Returns the + * seed tab's target id. + */ +export const seedProfileWindow = async ( + client: BrowserClient, + pin: ProfilePin, +): Promise<Result<string, CdpError>> => { + // A pin belongs to one user-data-dir. If the harness is now attached to a + // different browser, silently seeding the "same" profile name would be wrong. + const connectedDir = client.userDataDir(); + if (connectedDir && connectedDir !== pin.userDataDir) { + return err(cdpError( + "discovery_failed", + `the selected profile "${pin.label}" belongs to ${pin.userDataDir}, but the harness is connected to ${connectedDir}. ` + + `Run /browser-profile to choose a profile in the browser you're using.`, + )); + } + + // Identify the browser by the dir we're connected to. With several Chromium + // browsers running, launching into the wrong one would open a window the + // harness cannot see — in a browser the user never pointed us at. + const targetUserDataDir = connectedDir ?? pin.userDataDir; + const running = await detectRunningBrowser(targetUserDataDir); + const exePath = await resolveBrowserExecutable(running); + if (!exePath) { + return err(cdpError( + "discovery_failed", + `could not locate the browser executable needed to open "${pin.label}" — ${manualFallbackHint(pin)}`, + )); + } + + const launched = await openProfileWindow({ + exePath, + profileDir: pin.profileDir, + // Forward the user-data-dir ONLY when the running browser advertises one. + // ProcessSingleton keys on this path: passing the value the target process + // itself reports guarantees delegation, whereas supplying a default dir we + // reconstructed could differ by case or normalisation and start a SECOND + // browser instead. + ...(running.explicitUserDataDir ? { explicitUserDataDir: running.explicitUserDataDir } : {}), + }); + if (!launched.success) { + return err(cdpError("discovery_failed", `${launched.error} — ${manualFallbackHint(pin)}`)); + } + + const { sentinelUrl, cleanup } = launched.data; + const deadline = Date.now() + SEED_DEADLINE_MS; + let seed: PageTarget | undefined; + while (Date.now() < deadline && !seed) { + await new Promise((r) => setTimeout(r, SEED_POLL_MS)); + seed = (await pageTargets(client)).find((t) => t.url === sentinelUrl); + } + + if (!seed) { + await cleanup(); + return err(cdpError( + "timeout", + `couldn't open a window in "${pin.label}" automatically — ${manualFallbackHint(pin)}`, + )); + } + + // The sentinel's context is the profile's context — the only way to learn it. + if (seed.browserContextId) client.setProfileContextId(seed.browserContextId); + + const ownership = client.ownership(); + ownership.setHarnessWindow(seed.targetId); + ownership.add(seed.targetId); + const win = await client.session().callBrowser("Browser.getWindowForTarget", { targetId: seed.targetId }); + if (win.success) { + const windowId = (win.data as { windowId?: number }).windowId; + if (typeof windowId === "number") ownership.setHarnessWindowId(windowId); + } + + // Blank the handshake page so the seed tab behaves like any other fresh tab, + // then remove the sentinel file. Best-effort: a failure here costs nothing + // beyond the placeholder page staying visible. + const attached = await client.session().callBrowser("Target.attachToTarget", { + targetId: seed.targetId, + flatten: true, + }); + if (attached.success) { + const sessionId = (attached.data as { sessionId: string }).sessionId; + await client.session().callOnTarget("Page.navigate", { url: "about:blank" }, sessionId); + } + await cleanup(); + + return ok(seed.targetId); +}; |
