|
| 1 | +// The agentic connect handoff: an agent adds an API over MCP, asks for a |
| 2 | +// handoff URL (`coreTools.connections.createHandoff`), and the user opens that |
| 3 | +// URL in a browser to paste the credential. This scenario walks the WHOLE |
| 4 | +// path — the exact flow that failed in production with a "wrong / bad" URL — |
| 5 | +// against a real emulated provider (resend.emulators.dev) so the failure |
| 6 | +// point is captured with trace + screenshots instead of guessed at: |
| 7 | +// |
| 8 | +// 1. MCP `execute` → `openapi.addSpec` registers the emulated Resend API |
| 9 | +// 2. MCP `execute` → `connections.createHandoff` returns the browser URL |
| 10 | +// 3. The URL's origin must be THIS deployment (not a hardcoded host) |
| 11 | +// 4. Playwright opens it: the Add connection modal must be open with a |
| 12 | +// credential field, the emulator-minted API key is pasted and submitted |
| 13 | +// 5. The saved connection is proven live: `execute` sends an email through |
| 14 | +// the new tools and the emulator's request ledger shows the call |
| 15 | +// arriving with the pasted bearer token |
| 16 | +import { randomBytes } from "node:crypto"; |
| 17 | + |
| 18 | +import { expect } from "@effect/vitest"; |
| 19 | +import { Effect } from "effect"; |
| 20 | +import { composePluginApi } from "@executor-js/api/server"; |
| 21 | +import { openApiHttpPlugin } from "@executor-js/plugin-openapi/api"; |
| 22 | + |
| 23 | +import { scenario } from "../src/scenario"; |
| 24 | +import { Api, Browser, Mcp, Target } from "../src/services"; |
| 25 | +import type { Identity, Target as TargetShape } from "../src/target"; |
| 26 | +import type { BrowserSurface } from "../src/surfaces/browser"; |
| 27 | +import type { McpSession } from "../src/surfaces/mcp"; |
| 28 | + |
| 29 | +const api = composePluginApi([openApiHttpPlugin()] as const); |
| 30 | + |
| 31 | +const unique = (prefix: string) => `${prefix}_${randomBytes(4).toString("hex")}`; |
| 32 | + |
| 33 | +const EMULATOR_BASE = "https://resend.emulators.dev"; |
| 34 | + |
| 35 | +// The emulator serves its own OpenAPI document (bearer auth, same shape as |
| 36 | +// real Resend — and as the Sentry spec that failed in prod). Adding it by URL |
| 37 | +// with no authenticationTemplate exercises exactly the agentic path: the |
| 38 | +// add-account modal must render a paste-a-token flow derived from the spec's |
| 39 | +// bare `http`/`bearer` security scheme. |
| 40 | +const EMULATOR_SPEC_URL = `${EMULATOR_BASE}/openapi.json`; |
| 41 | + |
| 42 | +const addSpecCode = (slug: string) => ` |
| 43 | +const added = await tools.executor.openapi.addSpec({ |
| 44 | + spec: { kind: "url", url: ${JSON.stringify(EMULATOR_SPEC_URL)} }, |
| 45 | + slug: ${JSON.stringify(slug)}, |
| 46 | +}); |
| 47 | +return added.ok ? { ok: true, slug: added.data.slug, toolCount: added.data.toolCount } : { ok: false, error: added.error }; |
| 48 | +`; |
| 49 | + |
| 50 | +const createHandoffCode = (slug: string) => ` |
| 51 | +const handoff = await tools.executor.coreTools.connections.createHandoff({ |
| 52 | + integration: ${JSON.stringify(slug)}, |
| 53 | + owner: "org", |
| 54 | + label: "Resend (emulated)", |
| 55 | +}); |
| 56 | +return handoff.ok ? { ok: true, url: handoff.data.url } : { ok: false, error: handoff.error }; |
| 57 | +`; |
| 58 | + |
| 59 | +// Selfhost scenarios share one workspace identity — leaked connections fail |
| 60 | +// other scenarios' zero-state assertions, so remove everything this one made. |
| 61 | +const removeConnectionsCode = (slug: string) => ` |
| 62 | +const list = await tools.executor.coreTools.connections.list({}); |
| 63 | +const mine = (list.ok ? list.data.connections : []).filter((c) => c.integration === ${JSON.stringify(slug)}); |
| 64 | +for (const c of mine) { |
| 65 | + await tools.executor.coreTools.connections.remove({ owner: c.owner, integration: c.integration, name: c.name }); |
| 66 | +} |
| 67 | +return { removed: mine.length }; |
| 68 | +`; |
| 69 | + |
| 70 | +const sendEmailCode = (slug: string, subject: string) => ` |
| 71 | +const found = await tools.search({ namespace: ${JSON.stringify(slug)}, query: "send email", limit: 5 }); |
| 72 | +const path = found.items[0]?.path; |
| 73 | +if (!path) return { ok: false, error: "no send tool found", items: found.items }; |
| 74 | +let t = tools; |
| 75 | +for (const seg of path.split(".")) t = t[seg]; |
| 76 | +const sent = await t({ |
| 77 | + body: { |
| 78 | + from: "onboarding@example.com", |
| 79 | + to: "e2e@example.com", |
| 80 | + subject: ${JSON.stringify(subject)}, |
| 81 | + html: "<p>connect-handoff e2e</p>", |
| 82 | + }, |
| 83 | +}); |
| 84 | +return { ok: sent.ok, path, result: sent.ok ? sent.data : sent.error }; |
| 85 | +`; |
| 86 | + |
| 87 | +/** Run `execute`, auto-approving a paused execution (policy elicitation) once, |
| 88 | + * and parse the sandbox's JSON return value. */ |
| 89 | +const executeJson = (session: McpSession, code: string) => |
| 90 | + Effect.gen(function* () { |
| 91 | + let result = yield* session.call("execute", { code }); |
| 92 | + if (result.text.includes("executionId:")) { |
| 93 | + result = yield* session.approvePaused(result.text); |
| 94 | + } |
| 95 | + expect(result.ok, `execute completed (got: ${result.text.slice(0, 400)})`).toBe(true); |
| 96 | + return JSON.parse(result.text) as Record<string, unknown>; |
| 97 | + }); |
| 98 | + |
| 99 | +const mintEmulatorApiKey = Effect.promise(async () => { |
| 100 | + const response = await fetch(`${EMULATOR_BASE}/_emulate/credentials`, { |
| 101 | + method: "POST", |
| 102 | + headers: { "content-type": "application/json" }, |
| 103 | + body: JSON.stringify({ type: "api-key" }), |
| 104 | + }); |
| 105 | + const body = (await response.json()) as { credential?: { token?: string } }; |
| 106 | + const token = body.credential?.token; |
| 107 | + if (!token) throw new Error(`emulator credential mint failed: ${JSON.stringify(body)}`); |
| 108 | + return token; |
| 109 | +}); |
| 110 | + |
| 111 | +const fetchLedgerText = Effect.promise(async () => { |
| 112 | + const response = await fetch(`${EMULATOR_BASE}/_emulate/ledger`); |
| 113 | + return response.text(); |
| 114 | +}); |
| 115 | + |
| 116 | +scenario( |
| 117 | + "Connect · the agentic handoff URL opens this deployment's add-account flow and the pasted key works", |
| 118 | + { timeout: 240_000 }, |
| 119 | + Effect.gen(function* () { |
| 120 | + const target = yield* Target; |
| 121 | + const mcp = yield* Mcp; |
| 122 | + const browser = yield* Browser; |
| 123 | + const { client: makeApiClient } = yield* Api; |
| 124 | + |
| 125 | + const integration = unique("resendhf"); |
| 126 | + const emailSubject = unique("connect-handoff"); |
| 127 | + const apiKey = yield* mintEmulatorApiKey; |
| 128 | + |
| 129 | + const identity = yield* target.newIdentity(); |
| 130 | + const session = mcp.session(identity); |
| 131 | + const client = yield* makeApiClient(api, identity); |
| 132 | + |
| 133 | + yield* runScenario({ |
| 134 | + target, |
| 135 | + browser, |
| 136 | + session, |
| 137 | + identity, |
| 138 | + integration, |
| 139 | + emailSubject, |
| 140 | + apiKey, |
| 141 | + }).pipe( |
| 142 | + // Best-effort cleanup even on failure: drop the created connection(s) |
| 143 | + // over MCP, then the integration over the API. |
| 144 | + Effect.ensuring( |
| 145 | + Effect.gen(function* () { |
| 146 | + yield* session.call("execute", { code: removeConnectionsCode(integration) }); |
| 147 | + yield* client.openapi.removeSpec({ params: { slug: integration } }); |
| 148 | + }).pipe(Effect.ignore), |
| 149 | + ), |
| 150 | + ); |
| 151 | + }), |
| 152 | +); |
| 153 | + |
| 154 | +const runScenario = (input: { |
| 155 | + readonly target: TargetShape; |
| 156 | + readonly browser: BrowserSurface; |
| 157 | + readonly session: McpSession; |
| 158 | + readonly identity: Identity; |
| 159 | + readonly integration: string; |
| 160 | + readonly emailSubject: string; |
| 161 | + readonly apiKey: string; |
| 162 | +}) => |
| 163 | + Effect.gen(function* () { |
| 164 | + const { target, browser, session, identity, integration, emailSubject, apiKey } = input; |
| 165 | + |
| 166 | + // 1. Agent registers the emulated provider over MCP. |
| 167 | + const added = yield* executeJson(session, addSpecCode(integration)); |
| 168 | + expect(added.ok, `addSpec succeeded: ${JSON.stringify(added)}`).toBe(true); |
| 169 | + |
| 170 | + // 2. Agent asks for the browser handoff URL. |
| 171 | + const handoff = yield* executeJson(session, createHandoffCode(integration)); |
| 172 | + expect(handoff.ok, `createHandoff succeeded: ${JSON.stringify(handoff)}`).toBe(true); |
| 173 | + const handoffUrl = String(handoff.url); |
| 174 | + |
| 175 | + // 3. The URL must target THIS deployment. (Production returned a URL the |
| 176 | + // user called "wrong/bad" — pin the contract here.) |
| 177 | + const parsed = new URL(handoffUrl); |
| 178 | + expect(parsed.origin, `handoff URL (${handoffUrl}) targets this deployment`).toBe( |
| 179 | + new URL(target.baseUrl).origin, |
| 180 | + ); |
| 181 | + expect(parsed.pathname).toBe(`/integrations/${integration}`); |
| 182 | + expect(parsed.searchParams.get("addAccount")).toBe("1"); |
| 183 | + |
| 184 | + // 4. The user opens the handoff URL and pastes the credential. |
| 185 | + yield* browser.session(identity, async ({ page, step }) => { |
| 186 | + await step("Open the handoff URL from the agent", async () => { |
| 187 | + await page.goto(handoffUrl, { waitUntil: "networkidle" }); |
| 188 | + }); |
| 189 | + |
| 190 | + await step("The Add connection modal is open", async () => { |
| 191 | + await page.getByRole("heading", { name: /Add connection/ }).waitFor({ timeout: 15_000 }); |
| 192 | + }); |
| 193 | + |
| 194 | + await step("Paste the emulator API key", async () => { |
| 195 | + const credential = page.getByPlaceholder(/paste the value \/ token/i); |
| 196 | + await credential.waitFor({ timeout: 15_000 }); |
| 197 | + await credential.fill(apiKey); |
| 198 | + }); |
| 199 | + |
| 200 | + await step("Submit Add connection", async () => { |
| 201 | + await page.getByRole("button", { name: "Add connection", exact: true }).click(); |
| 202 | + await page |
| 203 | + .getByRole("heading", { name: /Add connection/ }) |
| 204 | + .waitFor({ state: "hidden", timeout: 20_000 }); |
| 205 | + }); |
| 206 | + }); |
| 207 | + |
| 208 | + // 5. The connection is live: send an email through the new tools and see |
| 209 | + // it arrive at the emulator with the pasted token. |
| 210 | + const sent = yield* executeJson(session, sendEmailCode(integration, emailSubject)); |
| 211 | + expect(sent.ok, `email sent through the pasted connection: ${JSON.stringify(sent)}`).toBe(true); |
| 212 | + |
| 213 | + const ledger = yield* fetchLedgerText; |
| 214 | + expect( |
| 215 | + ledger.includes(emailSubject), |
| 216 | + "the emulator's request ledger recorded the call made through Executor", |
| 217 | + ).toBe(true); |
| 218 | + }); |
0 commit comments