hermes-agent/apps/desktop/electron/link-title-wall.ts
teknium1 3d0bdce9f9 fix(desktop): prove the sign-in wall from curl's arrival URL and title too
Port the reporter's (@tianyu-liu) url_effective half: curl reports where it
landed via --write-out on a tail buffer that survives the body budget, and
isAuthWall decides on arrival URL -> sign-in title -> markup, so the Apps
Script ServiceLogin shape (no markup id) is a wall too and a wall's own title
is never used as the link label. Tests trimmed to two invariants.

Co-authored-by: tianyu-liu <[email protected]>
2026-09-20 19:37:57 -07:00

128 lines
5 KiB
TypeScript
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

// The ladder between the curl tier and the hidden-Chromium tier of link-title
// resolution: curl (tier 1) → hidden BrowserWindow (tier 2).
//
// The title partition is cookieless (`session.fromPartition('hermes:link-titles',
// { cache: false })`), so a Google Workspace link answers tier 1 with Google's
// sign-in page — and tier 2 used to be reached *exactly* when tier 1 produced no
// usable title, i.e. for every sign-in wall. Loading that wall in the real
// hidden Chromium makes it ask the OS authenticator for a passkey: a native
// credential dialog on the user's desktop for what is only a link title.
//
// Measured shapes of the wall (curl, the app's own flags):
// drive.google.com/drive/folders/<id> -> accounts.google.com/v3/signin/identifier?… (body: id="identifierId")
// docs.google.com/document/d/<id>/edit -> stays on docs.google.com, no <title>, body names accounts.google.com/ServiceLogin
// script.google.com/a/<d>/macros/s/<id>/exec -> www.google.com/a/<d>/ServiceLogin?… (no marker in the body at all)
// The last shape is why markup alone is not enough: the hosts whose cookieless
// answer can only ever be that wall are named too. A published Doc or Site still
// gets its title from the curl tier, which is what keeps these links fetchable.
const TITLE_MAX_CHARS = 240
// Strips known error/captcha titles (e.g. "GetYourGuide – Error", "Just a
// moment...") so they don't get cached as the resolved title. An error title is
// '' here, which is also the one case the renderer tier is still allowed to try.
const TITLE_ERROR_RE =
/\b(access denied|attention required|captcha|error|forbidden|just a moment|request blocked|too many requests)\b/i
function usableTitle(value: string): string {
return value && !TITLE_ERROR_RE.test(value) ? value : ''
}
/** Sign-in markup — the tell for a wall whose redirect curl already followed. */
const AUTH_WALL_BODY_RE =
/accounts\.google\.com(?:\/|&#47;)(?:ServiceLogin|signin)|id="identifierId"|id="gaia_loginform"/i
/** Hosts whose only cookieless answer is a sign-in page. */
const RENDERER_WALL_HOSTS = new Set([
'accounts.google.com',
'console.cloud.google.com',
'docs.google.com',
'drive.google.com',
'script.google.com',
'sheets.google.com',
'slides.google.com',
'sites.google.com'
])
function hostOf(rawUrl: string): string {
try {
return new URL(rawUrl).hostname.toLowerCase()
} catch {
return ''
}
}
/** Arrival URL or title of a sign-in page — the shapes that carry no markup id. */
const AUTH_WALL_URL_RE = /^accounts\.google\.com$|\/ServiceLogin\b|\/signin\b|\/o\/oauth2\//i
const AUTH_WALL_TITLE_RE = /\bsign[ -]?in\b/i
/**
* Curl's proof that the page it landed on is a sign-in wall, decided on the
* arrival URL → the title → the markup, in that order of reliability: an Apps
* Script `/a/<domain>/ServiceLogin` wall carries none of the markup ids.
*/
export function isAuthWall(input: { body: string; effectiveUrl: string; title: string }): boolean {
let arrival: URL | null = null
try {
arrival = input.effectiveUrl ? new URL(input.effectiveUrl) : null
} catch {
arrival = null
}
if (arrival && (AUTH_WALL_URL_RE.test(arrival.hostname) || AUTH_WALL_URL_RE.test(arrival.pathname))) {
return true
}
if (AUTH_WALL_TITLE_RE.test(input.title || '')) {
return true
}
return AUTH_WALL_BODY_RE.test(input.body || '')
}
/**
* May this answer still escalate to the tier-2 hidden Chromium?
*
* `title` is tier 1's title after `usableTitle` — '' for both "nothing found"
* and "found an error/captcha title". `authWall` is tier 1's proof that the page
* it landed on is a sign-in wall.
*/
export function needsRendererFallback(input: { authWall: boolean; title: string; url: string }): boolean {
// Tier 1 resolved a usable title — nothing left to escalate for.
if (input.title) {
return false
}
// A proven sign-in wall, or a host that can only answer with one: tier 2 would
// load the same signed-out page in a real browser and raise the OS passkey
// dialog. The link keeps its host/path label instead of a document title.
if (input.authWall || RENDERER_WALL_HOSTS.has(hostOf(input.url))) {
return false
}
return true
}
/**
* Tier 1 → tier 2, with the sign-in wall never reaching the renderer. Both tiers
* are injected so the ladder is provable without booting Electron: main.ts owns
* the I/O, this owns the decision.
*/
export async function resolveLinkTitle(input: {
curl: () => Promise<{ authWall: boolean; title: string }>
renderer: () => Promise<string>
url: string
}): Promise<string> {
const tier1 = await input.curl().catch(() => ({ authWall: false, title: '' }))
// A wall's own title ("Sign in - Google Accounts") is not the document's.
const title = tier1.authWall ? '' : usableTitle((tier1.title || '').slice(0, TITLE_MAX_CHARS))
if (!needsRendererFallback({ authWall: tier1.authWall, title, url: input.url })) {
return title
}
const rendered = await input.renderer().catch(() => '')
return usableTitle((rendered || '').slice(0, TITLE_MAX_CHARS))
}