@reticlehq/server 3.5.0 → 3.7.0

package/package.json

@@ -1,6 +1,7 @@
 {
   "name": "@reticlehq/server",
-  "version": "3.5.0",
+  "version": "3.7.0",
+  "mcpName": "io.github.reticlehq/reticle",
   "description": "Reticle CLI, bridge and MCP server: a coding agent drives your running web app from the inside and gets a verdict with evidence, for change verification, access control, accessibility, performance and SEO checks.",
   "type": "module",
   "main": "./dist/index.js",
@@ -19,15 +20,15 @@
     }
   },
   "dependencies": {
-    "@modelcontextprotocol/sdk": "^1.30.1",
+    "@modelcontextprotocol/sdk": "^1.31.0",
     "ws": "^8.21.3",
     "zod": "^3.24.1",
     "zod-to-json-schema": "^3.24.0",
-    "@reticlehq/browser": "3.5.0",
-    "@reticlehq/engine": "3.5.0",
-    "@reticlehq/core": "3.5.0",
+    "@reticlehq/browser": "3.7.0",
+    "@reticlehq/engine": "3.7.0",
     "open-verification": "1.0.0",
-    "@reticlehq/init": "3.5.0"
+    "@reticlehq/core": "3.7.0",
+    "@reticlehq/init": "3.7.0"
   },
   "devDependencies": {
     "@types/pngjs": "^6.0.5",
@@ -64,13 +65,9 @@
     "cli"
   ],
   "reticle": {
-    "changelog": "Reticle writes and tightens your regression flows for you: every drive is recorded automatically and replays with no model in the loop. The agent now sees nine tools instead of seventeen, `init` stops at a connected app and proving a flow is a separate first run, and Windows is supported. Upgrading from 2.x: see MIGRATION.md.",
+    "changelog": "Drive your localhost app from the Reticle platform's chat: the daemon of a linked project drives it on your machine, the chat shows it live and keeps it as a check with a replay. The Harness now runs only on the platform, on every plan, and plans, replays and learns each app. A page on SDK 3.1-3.4 is no longer version-skewed, a pressed key sends its keyCode, and a screenshot never carries another tab's pixels. Upgrading from 3.5: see MIGRATION.md.",
     "breakingChanges": [
-      "The default MCP surface is the merged nine. reticle_snapshot/query/inspect/state -> reticle_look { action }, reticle_network/console -> reticle_observe { action }, reticle_sessions/feedback -> reticle_session { action }, reticle_wait_for -> reticle_assert { action: \"wait\" }, reticle_act_sequence -> reticle_act { steps }. Old names redirect when CALLED, but an allowedTools allowlist or an MCP permission rule naming one refuses first -- update those by hand.",
-      "Twelve tools are unreachable on the default surface and have NO redirect: reticle_capabilities, reticle_clock, reticle_context, reticle_flow_replay, reticle_flow_save, reticle_intent, reticle_network_mock, reticle_record, reticle_run, reticle_screenshot, reticle_storage, reticle_visual_diff. The default surface ships no reticle_run hatch. Start the daemon with RETICLE_ADVERTISE_ALL_TOOLS=1 (read once at startup) and call them by name. Saved flows still replay via reticle_verify { action: \"flows\" }.",
-      "`reticle init` no longer drives: it ends at a connected session. --flow, --no-drive and --drive-model now exit non-zero, each naming its replacement. Prove a flow with reticle_verify { action: \"explore\", persona: \"<who does what>\" }, or npx @reticlehq/server verify <url> --explore --persona \"...\". init's exit code now means \"nothing connected\" rather than \"no verdict\". --files-only is unchanged.",
-      "@reticlehq/server now requires Node >=20.11 (was >=20.0.0).",
-      "@reticlehq/server, @reticlehq/init and @reticlehq/test declare license FSL-1.1-ALv2 instead of \"SEE LICENSE IN LICENSE\". The terms are identical; add the SPDX id to a licence allowlist."
+      "reticle_verify { action: \"explore\" } and reticle verify --explore no longer drive on a model key of your own. The Harness decides on the Reticle platform and needs a linked project (npx @reticlehq/server connect); every plan includes Harness credits. ANTHROPIC_API_KEY, OPENAI_API_KEY and JEV_API_KEY are no longer read for driving, RETICLE_HARNESS_DRIVER and RETICLE_HARNESS_MODEL are ignored, and the explore tool's `driver` argument is removed. A secret field's value comes from RETICLE_SECRET_<FIELD> on your machine."
     ]
   },
   "publishConfig": {

package/dist/command/cli/demo/demo-page.d.ts

@@ -1,78 +0,0 @@
-/**
- * The page the tutorial drives when the user has no app of their own yet.
- *
- * `reticle tutorial` used to print four steps and stop. Printing is not teaching: the reader is told
- * that a verdict is the point, and then watches nothing produce one. So the tour now RUNS — against a
- * page served from here, because the alternative (drive whatever is in front of the user) is how a
- * first run ends on a login wall, and a tutorial that fails on step one teaches that Reticle does not
- * work.
- *
- * It carries the real SDK, not a stub. A demo instrumented by something other than the thing being
- * demonstrated proves nothing about the thing being demonstrated.
- *
- * WHY THERE IS NO IMPORT MAP WRITTEN DOWN HERE. The SDK is ESM with bare specifiers, the shipped
- * server has no bundler, and the obvious fix — a hand-written map of specifier to path — is the exact
- * defect this repository keeps paying for: two lists that must agree, where the one nobody updated is
- * the one that breaks. A dependency added to the SDK would leave this map short by one, the page
- * would fail to load a module, and the tour would report that the user's setup is broken.
- *
- * So the map is DISCOVERED. Start at the SDK entry, scan the files beside it for bare specifiers,
- * resolve each one, and repeat. It cannot be short by one, because nothing here knows the names.
- *
- * The walk follows imports out of OUR packages only; a third-party one is mounted whole and not
- * read. That is a deliberate line, not an oversight. Reading zod means walking several thousand
- * files to learn something that has not changed in this repository's lifetime, and it measurably
- * cost about a minute — on the one command whose whole job is a good first impression. The drift
- * this guards against is our own SDK gaining a dependency, which is the thing that actually moves;
- * a third-party leaf that grows a bare runtime import would fail loudly on the page instead, which
- * is a worse message but not a silent one.
- */
-/** The signal the demo app fires when it saves — the consequence the tour declares in advance. */
-export declare const DEMO_SIGNAL = "demo:saved";
-/** Controls the tour points at. Named here so the page and the run cannot disagree about them. */
-export declare const DemoTestId: {
-    readonly SAVE: "demo-save";
-    readonly NAME: "demo-name";
-    readonly STATUS: "demo-status";
-};
-export interface SdkMount {
-    /** URL prefix the page fetches under. */
-    prefix: string;
-    /** Directory on disk it serves from. */
-    dir: string;
-}
-export interface SdkGraph {
-    /** Bare specifier to the URL the page should load it from. */
-    imports: Record<string, string>;
-    mounts: SdkMount[];
-    /** URL of the SDK entry module itself. */
-    entry: string;
-}
-/**
- * Resolve the SDK and everything it reaches, as URLs a page can fetch.
- *
- * `resolve` is injected so a test can describe a dependency graph without one existing on disk —
- * and because the real one is `import.meta.resolve`, which answers against THIS module's location
- * and so cannot be pointed at a fixture.
- */
-export declare function sdkGraph(resolve?: (spec: string) => string): SdkGraph;
-/**
- * The demo app, and the one line that makes it verifiable.
- *
- * Saving fires a SIGNAL as well as changing the text. Both are here on purpose: the text is what a
- * person sees, the signal is what the app itself says happened, and the tour waits on the signal
- * because that is the difference it exists to teach — a rendered word can be a coincidence, and an
- * app declaring its own state cannot.
- */
-export declare function demoPageHtml(graph: SdkGraph, token: string, bridgeUrl: string): string;
-/**
- * Serve the demo on a port the OS picks.
- *
- * Port 0 rather than a chosen number: this starts while a daemon, a dev server and possibly another
- * copy of this tour are already running, and a tutorial whose first act is a port collision has
- * taught the wrong lesson about the product.
- */
-export declare function serveDemoPage(html: string, graph: SdkGraph): Promise<{
-    url: string;
-    close: () => Promise<void>;
-}>;

package/dist/command/cli/demo/demo-page.js

@@ -1,241 +0,0 @@
-/**
- * The page the tutorial drives when the user has no app of their own yet.
- *
- * `reticle tutorial` used to print four steps and stop. Printing is not teaching: the reader is told
- * that a verdict is the point, and then watches nothing produce one. So the tour now RUNS — against a
- * page served from here, because the alternative (drive whatever is in front of the user) is how a
- * first run ends on a login wall, and a tutorial that fails on step one teaches that Reticle does not
- * work.
- *
- * It carries the real SDK, not a stub. A demo instrumented by something other than the thing being
- * demonstrated proves nothing about the thing being demonstrated.
- *
- * WHY THERE IS NO IMPORT MAP WRITTEN DOWN HERE. The SDK is ESM with bare specifiers, the shipped
- * server has no bundler, and the obvious fix — a hand-written map of specifier to path — is the exact
- * defect this repository keeps paying for: two lists that must agree, where the one nobody updated is
- * the one that breaks. A dependency added to the SDK would leave this map short by one, the page
- * would fail to load a module, and the tour would report that the user's setup is broken.
- *
- * So the map is DISCOVERED. Start at the SDK entry, scan the files beside it for bare specifiers,
- * resolve each one, and repeat. It cannot be short by one, because nothing here knows the names.
- *
- * The walk follows imports out of OUR packages only; a third-party one is mounted whole and not
- * read. That is a deliberate line, not an oversight. Reading zod means walking several thousand
- * files to learn something that has not changed in this repository's lifetime, and it measurably
- * cost about a minute — on the one command whose whole job is a good first impression. The drift
- * this guards against is our own SDK gaining a dependency, which is the thing that actually moves;
- * a third-party leaf that grows a bare runtime import would fail loudly on the page instead, which
- * is a worse message but not a silent one.
- */
-import { fileURLToPath } from 'node:url';
-import { createServer } from 'node:http';
-import { dirname, join, relative, extname, sep } from 'node:path';
-import { existsSync, readFileSync, readdirSync, statSync } from 'node:fs';
-/** Where the tour starts. Everything else on the page is reached from here. */
-const SDK_ENTRY = '@reticlehq/browser';
-/** The signal the demo app fires when it saves — the consequence the tour declares in advance. */
-export const DEMO_SIGNAL = 'demo:saved';
-/** Controls the tour points at. Named here so the page and the run cannot disagree about them. */
-export const DemoTestId = {
-    SAVE: 'demo-save',
-    NAME: 'demo-name',
-    STATUS: 'demo-status',
-};
-/** What the status line says before anything has been proved. */
-const STATUS_BEFORE = 'nothing saved yet';
-/** Bare specifiers this cannot serve and must not try to: the page is a browser. */
-const NOT_SERVABLE = /^node:/;
-/**
- * The packages whose imports are followed. See the note at the top for why the walk stops here.
- *
- * `open-verification` is listed by NAME because it is ours without being in the scope: it left the
- * org to stand on its own as the specification, and matching on `@reticlehq/` alone would silently
- * demote it to an unscanned third-party leaf. Today its only dependency is zod, which core already
- * pulls in, so nothing would have broken yet — which is exactly the kind of gap that is cheap now
- * and expensive on the day the spec grows a second dependency.
- */
-const OURS = /^(@reticlehq\/|open-verification$)/;
-const JS_CONTENT_TYPE = 'text/javascript; charset=utf-8';
-const HTML_CONTENT_TYPE = 'text/html; charset=utf-8';
-/**
- * A compiled package's own imports, read off disk.
- *
- * Test files are skipped: `dist` carries compiled specs, and those import `vitest` and `node:fs`,
- * which are not part of what a page loads and would send the walk looking for packages that only
- * exist in this repository.
- */
-function bareSpecifiersIn(dir) {
-    const found = new Set();
-    const walk = (at) => {
-        for (const entry of readdirSync(at, { withFileTypes: true })) {
-            const name = entry.name;
-            const path = join(at, name);
-            if (entry.isDirectory()) {
-                if ('node_modules' !== name)
-                    walk(path);
-                continue;
-            }
-            if (!name.endsWith('.js') || name.endsWith('.test.js'))
-                continue;
-            // `from` as a keyword, not the tail of a string: core's `z.discriminatedUnion('from', [...])`
-            // read as an import of everything up to the next quote. A specifier never holds whitespace.
-            for (const match of readFileSync(path, 'utf8').matchAll(/(?<![\w$.'"])from\s*'([^.'\s][^'\s]*)'/g)) {
-                const spec = match[1];
-                if (spec !== undefined && !NOT_SERVABLE.test(spec))
-                    found.add(spec);
-            }
-        }
-    };
-    walk(dir);
-    return found;
-}
-/** The directory holding a resolved entry's `package.json` — what has to be mounted whole. */
-function packageRootOf(entryPath) {
-    let at = dirname(entryPath);
-    for (;;) {
-        if (existsSync(join(at, 'package.json')))
-            return at;
-        const up = dirname(at);
-        if (up === at)
-            return undefined;
-        at = up;
-    }
-}
-/**
- * Resolve the SDK and everything it reaches, as URLs a page can fetch.
- *
- * `resolve` is injected so a test can describe a dependency graph without one existing on disk —
- * and because the real one is `import.meta.resolve`, which answers against THIS module's location
- * and so cannot be pointed at a fixture.
- */
-export function sdkGraph(resolve = defaultResolve) {
-    const imports = {};
-    const mounts = [];
-    const roots = new Map();
-    const queue = [SDK_ENTRY];
-    const seen = new Set();
-    let entry = '';
-    while (queue.length > 0) {
-        const spec = queue.shift();
-        if (spec === undefined || seen.has(spec))
-            continue;
-        seen.add(spec);
-        const entryPath = resolve(spec);
-        const root = packageRootOf(entryPath);
-        if (root === undefined) {
-            throw new Error(`the demo page cannot serve ${spec}: it resolves outside any package`);
-        }
-        let prefix = roots.get(root);
-        if (prefix === undefined) {
-            prefix = `/sdk/${String(mounts.length)}/`;
-            roots.set(root, prefix);
-            mounts.push({ prefix, dir: root });
-        }
-        const url = prefix + relative(root, entryPath).split(sep).join('/');
-        if (SDK_ENTRY === spec)
-            entry = url;
-        else
-            imports[spec] = url;
-        if (OURS.test(spec)) {
-            for (const next of bareSpecifiersIn(dirname(entryPath)))
-                queue.push(next);
-        }
-    }
-    return { imports, mounts, entry };
-}
-function defaultResolve(spec) {
-    return fileURLToPath(import.meta.resolve(spec));
-}
-/**
- * The demo app, and the one line that makes it verifiable.
- *
- * Saving fires a SIGNAL as well as changing the text. Both are here on purpose: the text is what a
- * person sees, the signal is what the app itself says happened, and the tour waits on the signal
- * because that is the difference it exists to teach — a rendered word can be a coincidence, and an
- * app declaring its own state cannot.
- */
-export function demoPageHtml(graph, token, bridgeUrl) {
-    const importMap = JSON.stringify({ imports: graph.imports });
-    const connect = JSON.stringify({ session: 'demo', url: bridgeUrl, token });
-    return `<!doctype html>
-<html lang="en"><head><meta charset="utf-8"><title>Reticle demo</title>
-<script type="importmap">${importMap}</script>
-<style>
- body{font:16px/1.5 system-ui,sans-serif;margin:0;padding:2.5rem 1.5rem;max-width:34rem}
- h1{font-size:1.3rem;margin:0 0 .25rem}
- p.lede{color:#555;margin:0 0 2rem}
- label{display:block;font-weight:600;margin:0 0 .4rem}
- input{font:inherit;padding:.5rem .6rem;width:100%;box-sizing:border-box;border:1px solid #bbb;border-radius:6px}
- button{font:inherit;margin-top:1rem;padding:.55rem 1.1rem;border:0;border-radius:6px;background:#111;color:#fff;cursor:pointer}
- p.status{margin-top:1.5rem;padding:.75rem;background:#f4f4f5;border-radius:6px}
-</style></head>
-<body>
- <h1>Reticle demo</h1>
- <p class="lede">A tiny app, instrumented with the real SDK. The tutorial is about to drive it.</p>
- <label for="${DemoTestId.NAME}">Display name</label>
- <input id="${DemoTestId.NAME}" data-testid="${DemoTestId.NAME}" value="Ada">
- <button data-testid="${DemoTestId.SAVE}">Save</button>
- <p class="status" data-testid="${DemoTestId.STATUS}">${STATUS_BEFORE}</p>
-<script type="module">
- import { reticle } from ${JSON.stringify(graph.entry)};
- reticle.connect(${connect});
- const status = document.querySelector('[data-testid="${DemoTestId.STATUS}"]');
- document.querySelector('[data-testid="${DemoTestId.SAVE}"]').addEventListener('click', () => {
-   const name = document.querySelector('[data-testid="${DemoTestId.NAME}"]').value;
-   status.textContent = 'saved ' + name;
-   reticle.signal(${JSON.stringify(DEMO_SIGNAL)}, { name });
- });
-</script>
-</body></html>`;
-}
-/**
- * Serve the demo on a port the OS picks.
- *
- * Port 0 rather than a chosen number: this starts while a daemon, a dev server and possibly another
- * copy of this tour are already running, and a tutorial whose first act is a port collision has
- * taught the wrong lesson about the product.
- */
-export function serveDemoPage(html, graph) {
-    const server = createServer((req, res) => {
-        const path = (req.url ?? '/').split('?')[0] ?? '/';
-        if ('/' === path) {
-            res.writeHead(200, { 'content-type': HTML_CONTENT_TYPE });
-            res.end(html);
-            return;
-        }
-        for (const mount of graph.mounts) {
-            if (!path.startsWith(mount.prefix))
-                continue;
-            const file = join(mount.dir, path.slice(mount.prefix.length));
-            // A request that climbed out of the mount is not served, whatever it points at.
-            if (!file.startsWith(mount.dir + sep) || !existsSync(file) || statSync(file).isDirectory()) {
-                break;
-            }
-            res.writeHead(200, {
-                'content-type': '.js' === extname(file) ? JS_CONTENT_TYPE : 'application/octet-stream',
-            });
-            res.end(readFileSync(file));
-            return;
-        }
-        res.writeHead(404);
-        res.end('not found');
-    });
-    return new Promise((resolve, reject) => {
-        server.once('error', reject);
-        server.listen(0, '127.0.0.1', () => {
-            const address = server.address();
-            if (null === address || 'string' === typeof address) {
-                reject(new Error('the demo page could not be served: no port was assigned'));
-                return;
-            }
-            resolve({
-                url: `http://127.0.0.1:${String(address.port)}/`,
-                close: () => new Promise((done) => {
-                    server.close(() => {
-                        done();
-                    });
-                }),
-            });
-        });
-    });
-}

package/dist/command/cli/demo/demo-run.d.ts

@@ -1,38 +0,0 @@
-/**
- * The tutorial, run rather than printed.
- *
- * `reticle tutorial` described four steps and produced nothing. The last of those steps says that a
- * drive without a verdict has no result — and the tour itself ended without one, which is the lesson
- * taught by counter-example. So the same four steps now execute, against the demo page, and every
- * line the reader sees is a thing that actually happened.
- *
- * The steps are NOT re-listed here. They come from `tutorialScript`, the same source the printed tour
- * reads, because two descriptions of one sequence drift and the one nobody updated is the one
- * somebody follows.
- *
- * Tools are called in-process rather than over the daemon's MCP transport. This command starts the
- * daemon it drives, so it already holds the session; going back out over a socket to reach a server
- * inside this process would add a transport that can fail to a tour whose job is to not fail. The
- * funnel telemetry is unaffected either way — `runTool` is the chokepoint that reports it, and both
- * routes pass through it.
- */
-export interface DemoTourOptions {
-    port: number;
-    headless: boolean;
-    /** Every line the reader sees, in order. Injected so a test reads the tour instead of a terminal. */
-    say: (line: string) => void;
-    now?: () => number;
-}
-/** A tour either reached a verdict or it did not; the exit code says which. */
-export interface DemoTourResult {
-    code: number;
-    verified: string | undefined;
-}
-/**
- * Run the tour, and report whether it ended at a proof.
- *
- * A step that cannot be taken stops the tour and says which one. Carrying on past a failed step
- * would end at a verdict taken against a page the earlier steps never reached — which is precisely
- * the false green the fourth step is there to warn about.
- */
-export declare function runDemoTour(options: DemoTourOptions): Promise<DemoTourResult>;

package/dist/command/cli/demo/demo-run.js

@@ -1,138 +0,0 @@
-/**
- * The tutorial, run rather than printed.
- *
- * `reticle tutorial` described four steps and produced nothing. The last of those steps says that a
- * drive without a verdict has no result — and the tour itself ended without one, which is the lesson
- * taught by counter-example. So the same four steps now execute, against the demo page, and every
- * line the reader sees is a thing that actually happened.
- *
- * The steps are NOT re-listed here. They come from `tutorialScript`, the same source the printed tour
- * reads, because two descriptions of one sequence drift and the one nobody updated is the one
- * somebody follows.
- *
- * Tools are called in-process rather than over the daemon's MCP transport. This command starts the
- * daemon it drives, so it already holds the session; going back out over a socket to reach a server
- * inside this process would add a transport that can fail to a tour whose job is to not fail. The
- * funnel telemetry is unaffected either way — `runTool` is the chokepoint that reports it, and both
- * routes pass through it.
- */
-import { ReticleTool } from '@reticlehq/core';
-// TOOLS comes through the package barrel rather than from `surface/tools/tools.js` directly. That
-// aggregator composes the tool arrays out of five directories, and anything importing it back is how
-// this package accumulated cycles; the barrel sits above both it and this file, which is the
-// direction that does not close a loop.
-import { start, TOOLS } from '../../../index.js';
-import { runTool } from '../../../surface/tools/invoke-tool.js';
-import { buildVerifyDeps } from '../cli-verify.js';
-import { readOrCreatePairingTokenSync, defaultPairingTokenDir, } from '../../../portal/bridge/pairing-token.js';
-import { tutorialScript, TutorialAudience, demoPlan } from '../tutorial.js';
-import { sdkGraph, demoPageHtml, serveDemoPage, DEMO_SIGNAL } from './demo-page.js';
-/** How long to wait for the demo page to dial the bridge before giving up on it. */
-const SESSION_TIMEOUT_MS = 30_000;
-const SESSION_POLL_MS = 250;
-/** Only `verified: "yes"` is a pass — see the contract every verdict tool here is held to. */
-const PROVED = 'yes';
-function toolNamed(name) {
-    const found = TOOLS.find((t) => t.name === name);
-    if (found === undefined) {
-        throw new Error(`the tutorial cannot run: ${name} is not in the tool registry`);
-    }
-    return found;
-}
-function asRecord(value) {
-    return 'object' === typeof value && null !== value ? value : {};
-}
-function asText(value) {
-    return 'string' === typeof value ? value : undefined;
-}
-/** The controls `reticle_query` found, in the shape the safety rule reads them. */
-function controlsIn(result) {
-    const raw = asRecord(result)['elements'];
-    if (!Array.isArray(raw))
-        return [];
-    const out = [];
-    for (const item of raw) {
-        const el = asRecord(item);
-        const ref = asText(el['ref']);
-        const name = asText(el['name']);
-        if (ref === undefined || name === undefined)
-            continue;
-        const role = asText(el['role']);
-        out.push(role === undefined ? { ref, name } : { ref, name, role });
-    }
-    return out;
-}
-async function waitForSession(running, now) {
-    const deadline = now() + SESSION_TIMEOUT_MS;
-    while (now() < deadline) {
-        if (running.bridge.sessions.count() > 0)
-            return true;
-        await new Promise((resolve) => setTimeout(resolve, SESSION_POLL_MS));
-    }
-    return running.bridge.sessions.count() > 0;
-}
-/**
- * Run the tour, and report whether it ended at a proof.
- *
- * A step that cannot be taken stops the tour and says which one. Carrying on past a failed step
- * would end at a verdict taken against a page the earlier steps never reached — which is precisely
- * the false green the fourth step is there to warn about.
- */
-export async function runDemoTour(options) {
-    const now = options.now ?? (() => Date.now());
-    const steps = new Map(tutorialScript(TutorialAudience.HUMAN).map((s) => [s.id, s]));
-    const announce = (id) => {
-        const step = steps.get(id);
-        if (step === undefined)
-            return;
-        options.say('');
-        options.say(step.say);
-        options.say(`  why: ${step.why}`);
-    };
-    const graph = sdkGraph();
-    const token = readOrCreatePairingTokenSync(defaultPairingTokenDir()) ?? '';
-    const bridgeUrl = `ws://localhost:${String(options.port)}/reticle`;
-    const page = await serveDemoPage(demoPageHtml(graph, token, bridgeUrl), graph);
-    let running;
-    try {
-        running = await start({ port: options.port, driveUrl: page.url, headless: options.headless });
-        const deps = buildVerifyDeps(running, process.cwd(), now);
-        const call = async (name, args) => await runTool(toolNamed(name), deps, args);
-        announce('connect');
-        if (!(await waitForSession(running, now))) {
-            options.say('  the demo page never dialled the bridge, so there is nothing to show.');
-            return { code: 1, verified: undefined };
-        }
-        options.say(`  one session connected: the demo app at ${page.url}`);
-        announce('look');
-        const snapshot = asRecord(await call(ReticleTool.SNAPSHOT, { mode: 'interactive' }));
-        for (const line of (asText(snapshot['tree']) ?? '').split('\n'))
-            options.say(`  ${line}`);
-        announce('declare');
-        const found = controlsIn(await call(ReticleTool.QUERY, { role: 'button' }));
-        const plan = demoPlan(found);
-        if (!plan.ok) {
-            options.say(`  ${plan.because}`);
-            return { code: 1, verified: undefined };
-        }
-        options.say(`  "${plan.name}" is safe to press. Saving should make the app say so itself:`);
-        options.say(`  reticle_act_and_wait { ref: "${plan.ref}", action: "click", until: {`);
-        options.say(`    kind: "signal", name: "${DEMO_SIGNAL}" } }`);
-        announce('verdict');
-        const verdict = asRecord(await call(ReticleTool.ACT_AND_WAIT, {
-            ref: plan.ref,
-            action: 'click',
-            until: { kind: 'signal', name: DEMO_SIGNAL },
-        }));
-        const verified = asText(verdict['verified']);
-        const because = asText(verdict['verifiedReason']) ?? asText(verdict['failureReason']);
-        options.say(`  verified: ${verified ?? 'unknown'}`);
-        if (because !== undefined)
-            options.say(`  because: ${because}`);
-        return { code: PROVED === verified ? 0 : 1, verified };
-    }
-    finally {
-        await page.close();
-        await running?.close?.();
-    }
-}

package/dist/features/harness/driver.d.ts

@@ -1,50 +0,0 @@
-/**
- * The model binding: the one file in the harness that knows there is a network.
- *
- * Everything about the loop lives in harness.ts against the `ModelDriver` interface, so this file is
- * a translation layer and nothing else. That split is not tidiness — it is what lets the loop be
- * tested against a scripted model with no API key, on every run of the unit gate.
- *
- * It speaks the Messages API over `fetch` rather than through a vendor SDK. The loop is ours (the
- * harness owns the step budget and has to record every call even when a run is cut short), so an SDK
- * would contribute one HTTP POST and a dependency — and this package's dependency list is short on
- * purpose.
- */
-import type { ModelDriver } from './harness.js';
-/**
- * The default driver model.
- *
- * Deliberately not the largest model available. The harness is not reasoning about business logic; it
- * is exploring an interface and stating consequences, and the verdict is produced by the engine
- * reading what it drove. If finding defects required a frontier model HERE, the engine would not be
- * doing its job — so the model tier is also a standing test of the design.
- */
-export declare const DEFAULT_HARNESS_MODEL = "claude-sonnet-5";
-export declare const DEFAULT_HARNESS_BASE_URL = "https://api.anthropic.com";
-/** The one call this file makes. Injected so a test can drive the translation with no network. */
-export type HarnessFetch = (url: string, init: {
-    method: string;
-    headers: Record<string, string>;
-    body: string;
-}) => Promise<{
-    ok: boolean;
-    status: number;
-    text(): Promise<string>;
-}>;
-export interface HarnessDriverOptions {
-    apiKey: string;
-    model?: string;
-    baseUrl?: string;
-    /** Injected for tests. Defaults to the platform `fetch`. */
-    fetch?: HarnessFetch;
-}
-/**
- * Read the harness's configuration out of the environment.
- *
- * Returns `undefined` when there is no key, which is the ONLY thing that decides whether the harness
- * is available. Stated as a read rather than a throw because "you have not configured this" is a
- * routine answer a caller offers a way out of, not a failure.
- */
-export declare function harnessOptionsFromEnv(env: Record<string, string | undefined>): HarnessDriverOptions | undefined;
-/** Build a driver backed by the Messages API. */
-export declare function harnessDriver(options: HarnessDriverOptions): ModelDriver;