e2e 0.17.0 → 0.18.0

package/dist/run/worker/protocol.d.ts

@@ -12,7 +12,8 @@ import type { CliOverrides, PortAssignments, ResolvedTarget } from '../../config
 import type { AiTraceSnapshot } from '../../internal/ai-trace.ts';
 import type { DebugSnapshot } from '../../internal/debug.ts';
 import type { SerializedError } from '../../internal/errors.ts';
-import type { ResultRecord, RunError, SerialGroupRecord, WireResultRecord } from '../records.ts';
+import type { AttemptRecord, ResultRecord, RunError, SerialGroupRecord, SerialMemberRecord, WireResultRecord } from '../records.ts';
+import type { SerialAttemptRun, SerialAttemptStart } from '../serial.ts';
 import type { StepProgress } from '../steps.ts';
 /** One runnable pair on the wire; the worker resolves the test function. */
 export interface WirePair {
@@ -107,7 +108,15 @@ export interface ShutdownMessage {
 export interface TerminateMessage {
     readonly type: 'terminate';
 }
-export type MainToWorker = RunUnitMessage | InterruptMessage | ShutdownMessage | TerminateMessage;
+/**
+ * Asks the worker to answer with `pong` from its event loop. The scheduler
+ * sends it only once an attempt is past its test timeout, to tell a worker
+ * still tearing the attempt down from one whose event loop is blocked.
+ */
+export interface PingMessage {
+    readonly type: 'ping';
+}
+export type MainToWorker = RunUnitMessage | InterruptMessage | ShutdownMessage | TerminateMessage | PingMessage;
 export interface ReadyMessage {
     readonly type: 'ready';
 }
@@ -166,6 +175,35 @@ export interface NoticeMessage {
     readonly type: 'notice';
     readonly message: string;
 }
+/**
+ * An attempt's test timeout started: its `beforeEach` hooks and body run
+ * now. The worker enforces the timeout with its own timers, which a body
+ * that blocks the event loop never lets fire, so the scheduler keeps the
+ * deadline too; see `SchedulerWorker.watch`.
+ */
+export interface AttemptDeadlineMessage {
+    readonly type: 'attempt-deadline';
+    readonly testId: string;
+    readonly agent: string;
+    readonly repeat: number;
+    /** The attempt's index among the pair's attempts. */
+    readonly attempt: number;
+    /** The attempt's record id and start, which a watchdog kill reports it with. */
+    readonly attemptId: string;
+    readonly startedAt: string;
+    /** The test's resolved `timeout`. */
+    readonly timeoutMs: number;
+    /** How long past the timeout the worker may go without answering a `ping`: the cleanup budget. */
+    readonly graceMs: number;
+}
+/** The attempt the last `attempt-deadline` announced has ended, verdict and cleanup included. */
+export interface AttemptEndMessage {
+    readonly type: 'attempt-end';
+}
+/** The answer to `ping`. */
+export interface PongMessage {
+    readonly type: 'pong';
+}
 export interface ResultMessage {
     readonly type: 'result';
     readonly result: WireResultRecord;
@@ -174,6 +212,48 @@ export interface SerialGroupMessage {
     readonly type: 'serial-group';
     readonly group: SerialGroupRecord;
 }
+/**
+ * An attempt of an ordinary or setup pair begins, its realm and `beforeAll`
+ * hooks included. With `attempt`, it tells a crash during an attempt from
+ * one between attempts (an `afterAll` after the last one): only the first
+ * is charged a new attempt, with this index.
+ */
+export interface AttemptStartMessage {
+    readonly type: 'attempt-start';
+    readonly testId: string;
+    readonly agent: string;
+    readonly repeat: number;
+    readonly index: number;
+}
+/**
+ * One finished attempt of an ordinary or setup pair. A pair's result waits
+ * for its last attempt, so each attempt also goes out as it ends: a worker
+ * that dies during a retry must not take the attempts before it along. The
+ * result still carries every attempt; the scheduler keeps these only until
+ * it arrives.
+ */
+export interface AttemptMessage {
+    readonly type: 'attempt';
+    readonly testId: string;
+    readonly agent: string;
+    readonly repeat: number;
+    readonly attempt: AttemptRecord;
+}
+/** One member that finished in the serial group attempt running now; see `AttemptMessage`. */
+export interface SerialMemberMessage {
+    readonly type: 'serial-member';
+    /** The group's report id (`SerialGroupRecord.id`). */
+    readonly groupId: string;
+    readonly attempt: SerialAttemptStart;
+    readonly member: SerialMemberRecord;
+}
+/** One finished serial group attempt; see `AttemptMessage`. */
+export interface SerialAttemptMessage {
+    readonly type: 'serial-attempt';
+    /** The group's report id (`SerialGroupRecord.id`). */
+    readonly groupId: string;
+    readonly run: SerialAttemptRun;
+}
 export interface UnitDoneMessage {
     readonly type: 'unit-done';
     readonly unitId: string;
@@ -208,6 +288,6 @@ export interface RunAbortMessage {
     readonly type: 'run-abort';
     readonly error: SerializedError;
 }
-export type WorkerToMain = ReadyMessage | PairStartMessage | ProgressMessage | OutputMessage | NoticeMessage | ResultMessage | SerialGroupMessage | UnitDoneMessage | ShutdownDoneMessage | FatalMessage | RunAbortMessage;
+export type WorkerToMain = ReadyMessage | PairStartMessage | ProgressMessage | AttemptDeadlineMessage | AttemptEndMessage | PongMessage | OutputMessage | NoticeMessage | ResultMessage | SerialGroupMessage | AttemptStartMessage | AttemptMessage | SerialMemberMessage | SerialAttemptMessage | UnitDoneMessage | ShutdownDoneMessage | FatalMessage | RunAbortMessage;
 /** Reattaches the scheduler's resolved target to a wire result. */
 export declare function decodeResult(wire: WireResultRecord, target: ResolvedTarget): ResultRecord;

package/dist/run/worker/session.js

@@ -4,6 +4,7 @@
  * in-process) drive this class, so unit execution, interrupt handling, and
  * run-error forwarding exist exactly once.
  */
+import inspector from 'node:inspector';
 import { DebugTrace } from '../../internal/debug.js';
 import { TargetExecutor } from '../execute.js';
 import { encodeResult } from '../records.js';
@@ -66,6 +67,10 @@ export class TargetWorker {
                             this.countFailure();
                     },
                     onSerialGroup: (group) => this.host.emit({ type: 'serial-group', group }),
+                    onAttemptStart: (pair, index) => this.host.emit({ type: 'attempt-start', testId: pair.test.id, agent: pair.agent, repeat: pair.repeat, index }),
+                    onAttempt: (pair, attempt) => this.host.emit({ type: 'attempt', testId: pair.test.id, agent: pair.agent, repeat: pair.repeat, attempt }),
+                    onSerialMember: (groupId, attempt, member) => this.host.emit({ type: 'serial-member', groupId, attempt, member }),
+                    onSerialAttempt: (groupId, run) => this.host.emit({ type: 'serial-attempt', groupId, run }),
                     onPairStart: (pair) => {
                         this.inFlight = { testId: pair.test.id, agent: pair.agent, repeat: pair.repeat };
                         this.host.emit({
@@ -78,6 +83,23 @@ export class TargetWorker {
                             serialId: pair.test.serialId,
                         });
                     },
+                    onAttemptDeadline: (pair, attempt) => {
+                        // A worker paused at a breakpoint looks blocked; with a debugger attached nothing watches its attempts.
+                        if (inspector.url() !== undefined)
+                            return;
+                        this.host.emit({
+                            type: 'attempt-deadline',
+                            testId: pair.test.id,
+                            agent: pair.agent,
+                            repeat: pair.repeat,
+                            attempt: attempt.index,
+                            attemptId: attempt.id,
+                            startedAt: attempt.startedAt,
+                            timeoutMs: pair.options.timeout,
+                            graceMs: deps.config.cleanupTimeout,
+                        });
+                    },
+                    onAttemptEnd: () => this.host.emit({ type: 'attempt-end' }),
                     onProgress: (pair, progress) => this.host.emit({ type: 'progress', testId: pair.test.id, agent: pair.agent, repeat: pair.repeat, progress }),
                     onNotice: (message) => this.host.emit({ type: 'notice', message }),
                     onRunAbort: (runError) => {
@@ -113,6 +135,12 @@ export class TargetWorker {
             case 'terminate':
                 this.terminate();
                 return;
+            case 'ping':
+                // Answered now, not queued behind the running unit: the answer is the proof the event loop turns.
+                this.host.emit({ type: 'pong' });
+                return;
+            default:
+                message;
         }
     }
     /**

package/dist/telemetry/environment.d.ts

@@ -42,7 +42,8 @@ export declare function fleetName(env: NodeJS.ProcessEnv): string | null;
  * The id the environment states for itself, standing in for a machine that
  * has no preferences file: a fleet by its name, else the CI vendor when `CI`
  * is set. Undefined on a person's machine, where the random per-machine id
- * is the unit; a vendor marker alone, without `CI`, does not make one a runner.
+ * is the unit: a vendor marker without `CI` is a shell on a runner, and `CI`
+ * with a coding agent but no vendor is the agent's shell on a laptop.
  */
 export declare function statedIdentity(env: NodeJS.ProcessEnv): string | undefined;
 export declare function collectEnvironment(options: {

package/dist/telemetry/environment.js

@@ -30,6 +30,10 @@ const CI_VENDORS = [
     ['SEMAPHORE', 'semaphore'],
     ['APPVEYOR', 'appveyor'],
     ['WOODPECKER', 'woodpecker'],
+    ['EAS_BUILD', 'eas'],
+    ['BITRISE_IO', 'bitrise'],
+    ['CM_BUILD_ID', 'codemagic'],
+    ['CI_XCODE_CLOUD', 'xcode-cloud'],
 ];
 /**
  * Environment variables coding agents set in the shells they spawn. The
@@ -99,13 +103,19 @@ export function fleetName(env) {
  * The id the environment states for itself, standing in for a machine that
  * has no preferences file: a fleet by its name, else the CI vendor when `CI`
  * is set. Undefined on a person's machine, where the random per-machine id
- * is the unit; a vendor marker alone, without `CI`, does not make one a runner.
+ * is the unit: a vendor marker without `CI` is a shell on a runner, and `CI`
+ * with a coding agent but no vendor is the agent's shell on a laptop.
  */
 export function statedIdentity(env) {
     const fleet = fleetName(env);
     if (fleet !== null)
         return `fleet:${fleet}`;
-    return isCiMode(env) ? `ci:${ciName(env) ?? 'unknown'}` : undefined;
+    if (!isCiMode(env))
+        return undefined;
+    const vendor = firstMarker(env, CI_VENDORS);
+    if (vendor === undefined && firstMarker(env, CODING_AGENTS) !== undefined)
+        return undefined;
+    return `ci:${vendor ?? 'unknown'}`;
 }
 /** The runtime executing this process: Bun and Deno each announce themselves in `process.versions`. */
 function runtime(versions) {

package/dist/telemetry/project.d.ts

@@ -8,10 +8,9 @@
  * commit, which every clone shares and no one can produce without the
  * repository itself; a commit hash carries no name, path, or remote. A
  * shallow clone has no root to offer: its boundary commits pose as roots and
- * move with every fetch, so it is treated like a directory outside git.
- * There, the project root path is hashed together with the machine's local
- * salt, so the id is stable on that machine and unrecoverable anywhere else.
- * In CI there is no salt, so a runner without a full history has no id: its
- * working directory is not a project.
+ * move with every fetch, so it has no id, like a directory outside git: a
+ * path is the same project on one machine only, so it cannot be told apart
+ * from a scratch directory nobody returns to.
  */
-export declare function anonymousProjectId(projectRoot: string, salt: string | undefined): Promise<string | undefined>;
+/** The hashed root commit of the repository at `projectRoot`; undefined outside git and in a shallow clone. */
+export declare function anonymousProjectId(projectRoot: string): Promise<string | undefined>;

package/dist/telemetry/project.js

@@ -8,11 +8,9 @@
  * commit, which every clone shares and no one can produce without the
  * repository itself; a commit hash carries no name, path, or remote. A
  * shallow clone has no root to offer: its boundary commits pose as roots and
- * move with every fetch, so it is treated like a directory outside git.
- * There, the project root path is hashed together with the machine's local
- * salt, so the id is stable on that machine and unrecoverable anywhere else.
- * In CI there is no salt, so a runner without a full history has no id: its
- * working directory is not a project.
+ * move with every fetch, so it has no id, like a directory outside git: a
+ * path is the same project on one machine only, so it cannot be told apart
+ * from a scratch directory nobody returns to.
  */
 import { COMMIT_HASH, git } from '../internal/git.js';
 import { sha256Hex } from '../internal/ids.js';
@@ -31,11 +29,8 @@ async function rootCommit(cwd) {
         .filter((line) => COMMIT_HASH.test(line))
         .toSorted()[0];
 }
-export async function anonymousProjectId(projectRoot, salt) {
+/** The hashed root commit of the repository at `projectRoot`; undefined outside git and in a shallow clone. */
+export async function anonymousProjectId(projectRoot) {
     const commit = await rootCommit(projectRoot);
-    if (commit !== undefined)
-        return sha256Hex(`git\n${commit}`);
-    if (salt === undefined)
-        return undefined;
-    return sha256Hex(`path\n${salt}\n${projectRoot}`);
+    return commit === undefined ? undefined : sha256Hex(`git\n${commit}`);
 }

package/dist/telemetry/store.d.ts

@@ -2,9 +2,11 @@
  * The user-level telemetry preferences: `~/.config/e2e/telemetry.json`
  * (`$XDG_CONFIG_HOME/e2e` when set, `%APPDATA%\e2e` on Windows). The file
  * holds the opt-out, the random id every event from this machine is
- * attributed to, the salt that keeps a hashed path unrecoverable, and which
- * notice was shown when. Nothing in it names the person or the machine: the
- * id and the salt are random bytes, generated here and never derived.
+ * attributed to, and which notice was shown when. Nothing in it names the
+ * person or the machine: the id is random bytes, generated here and never
+ * derived. A complete file also holds a random salt that nothing is hashed
+ * with anymore: older versions hashed paths with it and replace the id of a
+ * file without one, and a machine often has several versions installed.
  *
  * The store fails closed. `open` completes the file in one write when the id
  * or the salt is missing, and when that write cannot happen there is no
@@ -38,8 +40,6 @@ export declare class TelemetryStore {
     get enabled(): boolean;
     /** The random per-machine id every event from this machine is attributed to. */
     get anonymousId(): string;
-    /** The local salt that keeps a hashed project path unrecoverable; never leaves the machine. */
-    get pathSalt(): string;
     /** Whole days since the ids were generated; undefined for a file from before the time was recorded. */
     ageDays(now?: number): number | undefined;
     /** Saves the choice; false when the file could not be written and the choice stayed in memory. */

package/dist/telemetry/store.js

@@ -2,9 +2,11 @@
  * The user-level telemetry preferences: `~/.config/e2e/telemetry.json`
  * (`$XDG_CONFIG_HOME/e2e` when set, `%APPDATA%\e2e` on Windows). The file
  * holds the opt-out, the random id every event from this machine is
- * attributed to, the salt that keeps a hashed path unrecoverable, and which
- * notice was shown when. Nothing in it names the person or the machine: the
- * id and the salt are random bytes, generated here and never derived.
+ * attributed to, and which notice was shown when. Nothing in it names the
+ * person or the machine: the id is random bytes, generated here and never
+ * derived. A complete file also holds a random salt that nothing is hashed
+ * with anymore: older versions hashed paths with it and replace the id of a
+ * file without one, and a machine often has several versions installed.
  *
  * The store fails closed. `open` completes the file in one write when the id
  * or the salt is missing, and when that write cannot happen there is no
@@ -113,10 +115,6 @@ export class TelemetryStore {
     get anonymousId() {
         return this.preferences.anonymousId;
     }
-    /** The local salt that keeps a hashed project path unrecoverable; never leaves the machine. */
-    get pathSalt() {
-        return this.preferences.salt;
-    }
     /** Whole days since the ids were generated; undefined for a file from before the time was recorded. */
     ageDays(now = Date.now()) {
         const created = this.preferences.createdAt;

package/dist/telemetry/telemetry.d.ts

@@ -43,7 +43,7 @@ export interface TelemetryOptions {
     readonly fetch?: typeof fetch;
     /** Where the notice and the debug output go; stderr when absent. */
     readonly write?: (text: string) => void;
-    /** Resolves the anonymous project id; git and the salted path when absent. */
+    /** Resolves the anonymous project id; the hashed root commit when absent. */
     readonly projectId?: typeof anonymousProjectId;
 }
 export declare class Telemetry {
@@ -114,8 +114,7 @@ export declare class Telemetry {
     record(event: TelemetryEvent): void;
     /**
      * Git is asked for the project id now, while the command runs, so the
-     * flush at the end waits on the network alone. A fleet or CI has no
-     * store, so a path outside git has nothing to salt it and yields no id.
+     * flush at the end waits on the network alone.
      */
     private lookUpProject;
     /** The session event, first in the batch; absent for an invocation that never reached a command. */

package/dist/telemetry/telemetry.js

@@ -187,12 +187,10 @@ export class Telemetry {
     }
     /**
      * Git is asked for the project id now, while the command runs, so the
-     * flush at the end waits on the network alone. A fleet or CI has no
-     * store, so a path outside git has nothing to salt it and yields no id.
+     * flush at the end waits on the network alone.
      */
     lookUpProject() {
-        const store = this.statedId === undefined ? this.store() : undefined;
-        this.project ??= this.projectId(this.cwd, store?.pathSalt);
+        this.project ??= this.projectId(this.cwd);
     }
     /** The session event, first in the batch; absent for an invocation that never reached a command. */
     sessionEvent() {

package/dist/types.d.ts

@@ -265,6 +265,7 @@ export interface SwipeOptions {
     momentum?: Momentum;
     from?: never;
     to?: never;
+    duration?: never;
 }
 /** A swipe along a path between two viewport points: a touch swipe on a device, a pointer drag on a document platform. */
 export interface SwipePathOptions {
@@ -272,6 +273,8 @@ export interface SwipePathOptions {
     from: Point;
     /** Where it lifts. */
     to: Point;
+    /** Gesture duration in milliseconds; omitted = the engine's default. */
+    duration?: number;
     direction?: never;
     momentum?: never;
 }
@@ -1028,9 +1031,9 @@ export interface StoredArtifact {
     readonly path: string;
     /**
      * How much of the file the runner masked, as the report records it. A
-     * `download` is `incomplete` unless the runner rewrote it as text after a
-     * secret fill; a store that exports only what the runner vouches for reads
-     * this rather than the kind.
+     * `download` is `incomplete` unless the runner ran it as text through the
+     * session's secret values; a store that exports only what the runner
+     * vouches for reads this rather than the kind.
      */
     readonly redaction: 'complete' | 'not-required' | 'incomplete';
     readonly runId: string;

package/docs/agent-steps.mdx

@@ -1,6 +1,7 @@
 ---
 title: How agent steps work
 description: What the model sees, what it may do, how a step ends, and what it costs.
+icon: "list-check"
 ---
 
 An agent step is one `agent.act`, `agent.assert`, `agent.waitFor`, or

package/docs/agents.mdx

@@ -1,6 +1,7 @@
 ---
 title: Agents and personas
 description: Configure several agents in one suite, pin tests to them, and run one flow as each persona.
+icon: "users"
 ---
 
 Define named agents in `agents` to give tests different models, instructions,

package/docs/api-testing.mdx

@@ -1,6 +1,7 @@
 ---
-title: Testing APIs
+title: API
 description: Check your API with fetch and value matchers, in the same files and run as UI tests.
+icon: "server"
 ---
 
 API tests run in the same files and the same run as UI tests. A test that

package/docs/authentication.mdx

@@ -1,6 +1,7 @@
 ---
 title: Signing in
 description: Sign in once, reuse the session across tests, and pass secrets without exposing them to the model.
+icon: "key"
 ---
 
 Sign in once in a setup test and save a named session. Other tests restore
@@ -164,7 +165,7 @@ runner keeps it away from test code and the model altogether.
 
 A `Secret` has no `.value`. Three sinks accept one: the two below in a
 test, and an engine option that declares it, as
-[`basicAuth`](/starting-your-app#protected-previews) does:
+[`basicAuth`](/web#protected-previews) does:
 
 <Tabs>
   <Tab title="Deterministic">
@@ -279,7 +280,7 @@ account, a username with the secrets that sign it in, while a secret is one
 value, so `credentials.admin` and `secrets.admin` can both exist.
 
 In `e2e.config.ts`, `secrets.get()` hands an engine option a secret it needs
-before any test runs, such as a [basic-auth password](/starting-your-app#protected-previews).
+before any test runs, such as a [basic-auth password](/web#protected-previews).
 The config load checks the name, and the value is redacted once the engine
 resolves it, together with the base64 `Authorization` credential it becomes.
 The protection is text only: screenshots and the model's pixels are kept, so

package/docs/browser.mdx

@@ -1,6 +1,7 @@
 ---
 title: Working with the browser
 description: Mock an API, handle a dialog, download a file, reach into an iframe, and pick a browser.
+icon: "window-maximize"
 ---
 
 `@e2e-dev/web` runs tests in Chromium, Firefox, and WebKit. It is built on
@@ -144,7 +145,7 @@ await browser.setCookies([{ url: 'http://localhost:3000', name: 'tz', value: 'UT
 A cookie URL that does not parse, or uses a scheme other than `http:` or
 `https:` (`about:blank` included), is `POLICY_DENIED`; a domain cookie is checked as the origin
 it would be sent to. If any entry is refused, no cookies are set. To resize
-the page, see [Testing viewports](/viewports).
+the page, see [Viewports](/web#viewports).
 
 ## Pick a browser
 
@@ -183,7 +184,7 @@ bunx @e2e-dev/web install webkit --with-deps
 
 `connect` attaches to a remote Chromium over CDP instead of launching
 one; a [browser provider](#hosted-browsers) does the same and also leases
-and releases the browser. [Testing viewports](/viewports) covers sizes.
+and releases the browser. [Viewports](/web#viewports) covers sizes.
 
 ## Hosted browsers
 
@@ -197,7 +198,7 @@ when the run no longer uses it. The engine knows no vendor: for
 provider, and for any other service a provider is a few dozen lines in your
 project. Add `viewport: null` when
 the service shows the window in a live view: the page then
-[fills the window](/viewports#fill-the-window) instead of a fixed 1280 by
+[fills the window](/web#fill-the-window) instead of a fixed 1280 by
 720 area of it.
 
 ```ts
@@ -285,7 +286,7 @@ same browser.
 A provider against a typical session API has three requests: create a
 session, poll it until it exposes a DevTools endpoint, stop it.
 
-```ts title="browser-provider.ts"
+```ts title="browser-provider.ts" expandable
 import type { BrowserProvider } from '@e2e-dev/web';
 
 /**
@@ -470,7 +471,7 @@ root declared in HTML (`<template shadowrootmode="closed">`) is out of reach.
   <Card title="Web engine reference" icon="masks-theater" href="/reference/web">
     Every `browser` method, its timeout, and what it throws.
   </Card>
-  <Card title="Starting your app" icon="play" href="/starting-your-app">
-    Let the runner start the dev server and reach protected previews.
+  <Card title="Web" icon="play" href="/web">
+    Let the runner start the dev server, reach protected previews, and test viewports.
   </Card>
 </CardGroup>

package/docs/bug-bash.mdx

@@ -1,6 +1,7 @@
 ---
 title: Bug bashes
 description: Run many explorations at once, then prove each bug with a failing test before anyone reads about it.
+icon: "bug"
 ---
 
 A bug bash runs several `e2e explore` sessions at once, each on one area
@@ -50,7 +51,7 @@ itself: `npx e2e guide bug-bash`.
   `reuseExisting: true` on the target's `command`, or declare the URL with
   port `0` so each run starts its own copy. `reuseExisting` does not apply
   when `CI` is set; use port `0` there. See
-  [Starting your app](/starting-your-app).
+  [Starting your app](/web#starting-your-app).
 - When the target's `app.command` brings up a stack of its own, such as a
   database from `docker compose` that it removes on exit, start the stack once
   and explore with a config that points at the running app and declares no

package/docs/cache.mdx

@@ -1,6 +1,7 @@
 ---
-title: Caching agent steps
+title: Caching
 description: Replay verified agent actions without model calls, and let the agent take over when the app changes.
+icon: "box-archive"
 ---
 
 The replay cache records the actions in an `agent.act()` step after a later

package/docs/ci.mdx

@@ -1,13 +1,14 @@
 ---
 title: Continuous integration
 description: Run tests on pull requests, upload reports, and configure model credentials in CI.
+icon: "rotate"
 ---
 
 Run deterministic and agent tests in the same CI job. This workflow assumes
-your config includes a [command](/starting-your-app) to start the app and
+your config includes a [command](/web#starting-your-app) to start the app and
 uses AI Gateway for model calls:
 
-```yaml title=".github/workflows/e2e.yml"
+```yaml title=".github/workflows/e2e.yml" expandable
 name: e2e
 
 on:

package/docs/coding-agents.mdx

@@ -1,6 +1,7 @@
 ---
 title: Coding agents
 description: Give a coding agent the skill, a live app session, and reports it can use to write and debug tests.
+icon: "robot"
 ---
 
 `e2e init` installs a skill for writing and running tests, plus MCP
@@ -168,7 +169,7 @@ model call in the flow.
 Do not edit recordings by hand. A failed read-write run removes implicated
 entries so a later verified run can replace them. A failure where no model
 answered implicates nothing and removes none. See
-[Caching agent steps](/cache).
+[Caching](/cache).
 
 <CardGroup cols={2}>
   <Card title="CLI reference" icon="terminal" href="/reference/cli">