mirror of
https://github.com/harivansh-afk/clanker-agent.git
synced 2026-04-21 12:00:19 +00:00
Merge pull request #260 from getcompanion-ai/codex/deterministic-hosting
deterministic hosting
This commit is contained in:
Hari
GitHub
commit
63e33460f6
2 changed files with 689 additions and 608 deletions
|
|
@ -6,68 +6,81 @@ import { join, resolve } from "node:path";
|
||||||
import type { AgentTool } from "@mariozechner/pi-agent-core";
|
import type { AgentTool } from "@mariozechner/pi-agent-core";
|
||||||
import { type Static, Type } from "@sinclair/typebox";
|
import { type Static, Type } from "@sinclair/typebox";
|
||||||
import { getAgentDir } from "../../config.js";
|
import { getAgentDir } from "../../config.js";
|
||||||
import { getShellEnv, killProcessTree, sanitizeBinaryOutput } from "../../utils/shell.js";
|
import {
|
||||||
|
getShellEnv,
|
||||||
|
killProcessTree,
|
||||||
|
sanitizeBinaryOutput,
|
||||||
|
} from "../../utils/shell.js";
|
||||||
|
|
||||||
const browserActions = [
|
const browserActions = [
|
||||||
"open",
|
"open",
|
||||||
"snapshot",
|
"snapshot",
|
||||||
"click",
|
"click",
|
||||||
"fill",
|
"fill",
|
||||||
"wait",
|
"wait",
|
||||||
"screenshot",
|
"screenshot",
|
||||||
"state_save",
|
"state_save",
|
||||||
"state_load",
|
"state_load",
|
||||||
"close",
|
"close",
|
||||||
] as const;
|
] as const;
|
||||||
|
|
||||||
const browserSnapshotModes = ["interactive", "full"] as const;
|
const browserSnapshotModes = ["interactive", "full"] as const;
|
||||||
const browserLoadStates = ["load", "domcontentloaded", "networkidle"] as const;
|
const browserLoadStates = ["load", "domcontentloaded", "networkidle"] as const;
|
||||||
|
|
||||||
const DEFAULT_BROWSER_COMMAND = process.env.PI_AGENT_BROWSER_COMMAND || "agent-browser";
|
const DEFAULT_BROWSER_COMMAND =
|
||||||
|
process.env.PI_AGENT_BROWSER_COMMAND || "agent-browser";
|
||||||
const DEFAULT_BROWSER_TIMEOUT_SECONDS = 90;
|
const DEFAULT_BROWSER_TIMEOUT_SECONDS = 90;
|
||||||
|
|
||||||
const browserSchema = Type.Object({
|
const browserSchema = Type.Object({
|
||||||
action: Type.Union(
|
action: Type.Union(
|
||||||
browserActions.map((action) => Type.Literal(action)),
|
browserActions.map((action) => Type.Literal(action)),
|
||||||
{ description: "Browser action to execute" },
|
{ description: "Browser action to execute" },
|
||||||
),
|
),
|
||||||
url: Type.Optional(Type.String({ description: "URL to open, or URL glob to wait for" })),
|
url: Type.Optional(
|
||||||
mode: Type.Optional(
|
Type.String({ description: "URL to open, or URL glob to wait for" }),
|
||||||
Type.Union(
|
),
|
||||||
browserSnapshotModes.map((mode) => Type.Literal(mode)),
|
mode: Type.Optional(
|
||||||
{ description: "Snapshot mode. Defaults to interactive." },
|
Type.Union(
|
||||||
),
|
browserSnapshotModes.map((mode) => Type.Literal(mode)),
|
||||||
),
|
{ description: "Snapshot mode. Defaults to interactive." },
|
||||||
ref: Type.Optional(
|
),
|
||||||
Type.String({
|
),
|
||||||
description: "Element ref from snapshot output, such as @e2",
|
ref: Type.Optional(
|
||||||
}),
|
Type.String({
|
||||||
),
|
description: "Element ref from snapshot output, such as @e2",
|
||||||
value: Type.Optional(Type.String({ description: "Text value to fill into a field" })),
|
}),
|
||||||
text: Type.Optional(Type.String({ description: "Visible text to wait for" })),
|
),
|
||||||
ms: Type.Optional(
|
value: Type.Optional(
|
||||||
Type.Number({
|
Type.String({ description: "Text value to fill into a field" }),
|
||||||
description: "Milliseconds to wait",
|
),
|
||||||
minimum: 0,
|
text: Type.Optional(Type.String({ description: "Visible text to wait for" })),
|
||||||
}),
|
ms: Type.Optional(
|
||||||
),
|
Type.Number({
|
||||||
loadState: Type.Optional(
|
description: "Milliseconds to wait",
|
||||||
Type.Union(
|
minimum: 0,
|
||||||
browserLoadStates.map((state) => Type.Literal(state)),
|
}),
|
||||||
{ description: "Page load state to wait for" },
|
),
|
||||||
),
|
loadState: Type.Optional(
|
||||||
),
|
Type.Union(
|
||||||
path: Type.Optional(
|
browserLoadStates.map((state) => Type.Literal(state)),
|
||||||
Type.String({
|
{ description: "Page load state to wait for" },
|
||||||
description: "Output path for screenshots, relative to the current working directory if not absolute",
|
),
|
||||||
}),
|
),
|
||||||
),
|
path: Type.Optional(
|
||||||
fullPage: Type.Optional(Type.Boolean({ description: "Capture a full-page screenshot" })),
|
Type.String({
|
||||||
stateName: Type.Optional(
|
description:
|
||||||
Type.String({
|
"Output path for screenshots, relative to the current working directory if not absolute",
|
||||||
description: "Named browser state checkpoint stored under ~/.pi/agent/browser/states/",
|
}),
|
||||||
}),
|
),
|
||||||
),
|
fullPage: Type.Optional(
|
||||||
|
Type.Boolean({ description: "Capture a full-page screenshot" }),
|
||||||
|
),
|
||||||
|
stateName: Type.Optional(
|
||||||
|
Type.String({
|
||||||
|
description:
|
||||||
|
"Named browser state checkpoint stored under ~/.pi/agent/browser/states/",
|
||||||
|
}),
|
||||||
|
),
|
||||||
});
|
});
|
||||||
|
|
||||||
export type BrowserToolAction = (typeof browserActions)[number];
|
export type BrowserToolAction = (typeof browserActions)[number];
|
||||||
|
|
@ -76,433 +89,483 @@ export type BrowserLoadState = (typeof browserLoadStates)[number];
|
||||||
export type BrowserToolInput = Static<typeof browserSchema>;
|
export type BrowserToolInput = Static<typeof browserSchema>;
|
||||||
|
|
||||||
export interface BrowserToolDetails {
|
export interface BrowserToolDetails {
|
||||||
action: BrowserToolAction;
|
action: BrowserToolAction;
|
||||||
command: string;
|
command: string;
|
||||||
args: string[];
|
args: string[];
|
||||||
profilePath: string;
|
profilePath: string;
|
||||||
screenshotPath?: string;
|
screenshotPath?: string;
|
||||||
statePath?: string;
|
statePath?: string;
|
||||||
}
|
}
|
||||||
|
|
||||||
export interface BrowserOperations {
|
export interface BrowserOperations {
|
||||||
exec: (
|
exec: (
|
||||||
command: string,
|
command: string,
|
||||||
args: string[],
|
args: string[],
|
||||||
options: {
|
options: {
|
||||||
cwd: string;
|
cwd: string;
|
||||||
env: NodeJS.ProcessEnv;
|
env: NodeJS.ProcessEnv;
|
||||||
onData: (data: Buffer) => void;
|
onData: (data: Buffer) => void;
|
||||||
signal?: AbortSignal;
|
signal?: AbortSignal;
|
||||||
timeout?: number;
|
timeout?: number;
|
||||||
},
|
},
|
||||||
) => Promise<{ exitCode: number | null }>;
|
) => Promise<{ exitCode: number | null }>;
|
||||||
}
|
}
|
||||||
|
|
||||||
const defaultBrowserOperations: BrowserOperations = {
|
const defaultBrowserOperations: BrowserOperations = {
|
||||||
exec: (command, args, { cwd, env, onData, signal, timeout }) => {
|
exec: (command, args, { cwd, env, onData, signal, timeout }) => {
|
||||||
return new Promise((resolvePromise, rejectPromise) => {
|
return new Promise((resolvePromise, rejectPromise) => {
|
||||||
const child = spawn(command, args, {
|
const child = spawn(command, args, {
|
||||||
cwd,
|
cwd,
|
||||||
detached: true,
|
detached: true,
|
||||||
env,
|
env,
|
||||||
stdio: ["ignore", "pipe", "pipe"],
|
stdio: ["ignore", "pipe", "pipe"],
|
||||||
});
|
});
|
||||||
|
|
||||||
let timedOut = false;
|
let timedOut = false;
|
||||||
let timeoutHandle: NodeJS.Timeout | undefined;
|
let timeoutHandle: NodeJS.Timeout | undefined;
|
||||||
|
|
||||||
if (timeout !== undefined && timeout > 0) {
|
if (timeout !== undefined && timeout > 0) {
|
||||||
timeoutHandle = setTimeout(() => {
|
timeoutHandle = setTimeout(() => {
|
||||||
timedOut = true;
|
timedOut = true;
|
||||||
if (child.pid) {
|
if (child.pid) {
|
||||||
killProcessTree(child.pid);
|
killProcessTree(child.pid);
|
||||||
}
|
}
|
||||||
}, timeout * 1000);
|
}, timeout * 1000);
|
||||||
}
|
}
|
||||||
|
|
||||||
if (child.stdout) {
|
if (child.stdout) {
|
||||||
child.stdout.on("data", onData);
|
child.stdout.on("data", onData);
|
||||||
}
|
}
|
||||||
if (child.stderr) {
|
if (child.stderr) {
|
||||||
child.stderr.on("data", onData);
|
child.stderr.on("data", onData);
|
||||||
}
|
}
|
||||||
|
|
||||||
const onAbort = () => {
|
const onAbort = () => {
|
||||||
if (child.pid) {
|
if (child.pid) {
|
||||||
killProcessTree(child.pid);
|
killProcessTree(child.pid);
|
||||||
}
|
}
|
||||||
};
|
};
|
||||||
|
|
||||||
if (signal) {
|
if (signal) {
|
||||||
if (signal.aborted) {
|
if (signal.aborted) {
|
||||||
onAbort();
|
onAbort();
|
||||||
} else {
|
} else {
|
||||||
signal.addEventListener("abort", onAbort, { once: true });
|
signal.addEventListener("abort", onAbort, { once: true });
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
child.on("error", (error) => {
|
child.on("error", (error) => {
|
||||||
if (timeoutHandle) clearTimeout(timeoutHandle);
|
if (timeoutHandle) clearTimeout(timeoutHandle);
|
||||||
if (signal) signal.removeEventListener("abort", onAbort);
|
if (signal) signal.removeEventListener("abort", onAbort);
|
||||||
rejectPromise(error);
|
rejectPromise(error);
|
||||||
});
|
});
|
||||||
|
|
||||||
child.on("close", (code) => {
|
child.on("close", (code) => {
|
||||||
if (timeoutHandle) clearTimeout(timeoutHandle);
|
if (timeoutHandle) clearTimeout(timeoutHandle);
|
||||||
if (signal) signal.removeEventListener("abort", onAbort);
|
if (signal) signal.removeEventListener("abort", onAbort);
|
||||||
|
|
||||||
if (signal?.aborted) {
|
if (signal?.aborted) {
|
||||||
rejectPromise(new Error("aborted"));
|
rejectPromise(new Error("aborted"));
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
|
|
||||||
if (timedOut) {
|
if (timedOut) {
|
||||||
rejectPromise(new Error(`timeout:${timeout}`));
|
rejectPromise(new Error(`timeout:${timeout}`));
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
|
|
||||||
resolvePromise({ exitCode: code });
|
resolvePromise({ exitCode: code });
|
||||||
});
|
});
|
||||||
});
|
});
|
||||||
},
|
},
|
||||||
};
|
};
|
||||||
|
|
||||||
export interface BrowserToolOptions {
|
export interface BrowserToolOptions {
|
||||||
operations?: BrowserOperations;
|
operations?: BrowserOperations;
|
||||||
command?: string;
|
command?: string;
|
||||||
defaultTimeoutSeconds?: number;
|
defaultTimeoutSeconds?: number;
|
||||||
profileDir?: string;
|
profileDir?: string;
|
||||||
stateDir?: string;
|
stateDir?: string;
|
||||||
agentDir?: string;
|
agentDir?: string;
|
||||||
}
|
}
|
||||||
|
|
||||||
interface BrowserCommandContext {
|
interface BrowserCommandContext {
|
||||||
action: BrowserToolAction;
|
action: BrowserToolAction;
|
||||||
args: string[];
|
args: string[];
|
||||||
statusMessage: string;
|
statusMessage: string;
|
||||||
successMessage: string;
|
successMessage: string;
|
||||||
profilePath: string;
|
profilePath: string;
|
||||||
screenshotPath?: string;
|
screenshotPath?: string;
|
||||||
statePath?: string;
|
statePath?: string;
|
||||||
}
|
}
|
||||||
|
|
||||||
type BrowserCommandContextWithoutProfile = Omit<BrowserCommandContext, "profilePath">;
|
type BrowserCommandContextWithoutProfile = Omit<
|
||||||
|
BrowserCommandContext,
|
||||||
|
"profilePath"
|
||||||
|
>;
|
||||||
|
|
||||||
function resolveCommandPath(cwd: string, inputPath: string): string {
|
function resolveCommandPath(cwd: string, inputPath: string): string {
|
||||||
return resolve(cwd, inputPath);
|
return resolve(cwd, inputPath);
|
||||||
}
|
}
|
||||||
|
|
||||||
function getBrowserRootDir(options?: BrowserToolOptions): string {
|
function getBrowserRootDir(options?: BrowserToolOptions): string {
|
||||||
const baseAgentDir = options?.agentDir ?? getAgentDir();
|
const baseAgentDir = options?.agentDir ?? getAgentDir();
|
||||||
return join(baseAgentDir, "browser");
|
return join(baseAgentDir, "browser");
|
||||||
}
|
}
|
||||||
|
|
||||||
function getBrowserProfilePath(cwd: string, options?: BrowserToolOptions): string {
|
function getBrowserProfilePath(
|
||||||
const profilePath = options?.profileDir ?? join(getBrowserRootDir(options), "profile");
|
cwd: string,
|
||||||
return resolveCommandPath(cwd, profilePath);
|
options?: BrowserToolOptions,
|
||||||
|
): string {
|
||||||
|
const profilePath =
|
||||||
|
options?.profileDir ?? join(getBrowserRootDir(options), "profile");
|
||||||
|
return resolveCommandPath(cwd, profilePath);
|
||||||
}
|
}
|
||||||
|
|
||||||
function getBrowserStateDir(cwd: string, options?: BrowserToolOptions): string {
|
function getBrowserStateDir(cwd: string, options?: BrowserToolOptions): string {
|
||||||
const stateDir = options?.stateDir ?? join(getBrowserRootDir(options), "states");
|
const stateDir =
|
||||||
return resolveCommandPath(cwd, stateDir);
|
options?.stateDir ?? join(getBrowserRootDir(options), "states");
|
||||||
|
return resolveCommandPath(cwd, stateDir);
|
||||||
}
|
}
|
||||||
|
|
||||||
function createTempScreenshotPath(): string {
|
function createTempScreenshotPath(): string {
|
||||||
const id = randomBytes(8).toString("hex");
|
const id = randomBytes(8).toString("hex");
|
||||||
return join(tmpdir(), `pi-browser-screenshot-${id}.png`);
|
return join(tmpdir(), `pi-browser-screenshot-${id}.png`);
|
||||||
}
|
}
|
||||||
|
|
||||||
function normalizeOutput(chunks: Buffer[]): string {
|
function normalizeOutput(chunks: Buffer[]): string {
|
||||||
return sanitizeBinaryOutput(Buffer.concat(chunks).toString("utf-8")).trim();
|
return sanitizeBinaryOutput(Buffer.concat(chunks).toString("utf-8")).trim();
|
||||||
}
|
}
|
||||||
|
|
||||||
function sanitizeStateName(stateName: string): string {
|
function sanitizeStateName(stateName: string): string {
|
||||||
const trimmed = stateName.trim();
|
const trimmed = stateName.trim();
|
||||||
if (trimmed.length === 0) {
|
if (trimmed.length === 0) {
|
||||||
throw new Error("stateName is required for browser state actions");
|
throw new Error("stateName is required for browser state actions");
|
||||||
}
|
}
|
||||||
|
|
||||||
const withoutJsonSuffix = trimmed.endsWith(".json") ? trimmed.slice(0, -".json".length) : trimmed;
|
const withoutJsonSuffix = trimmed.endsWith(".json")
|
||||||
const sanitized = withoutJsonSuffix.replace(/[^a-zA-Z0-9._-]+/g, "-").replace(/^-+|-+$/g, "");
|
? trimmed.slice(0, -".json".length)
|
||||||
|
: trimmed;
|
||||||
|
const sanitized = withoutJsonSuffix
|
||||||
|
.replace(/[^a-zA-Z0-9._-]+/g, "-")
|
||||||
|
.replace(/^-+|-+$/g, "");
|
||||||
|
|
||||||
if (sanitized.length === 0) {
|
if (sanitized.length === 0) {
|
||||||
throw new Error(`Invalid browser state name: "${stateName}"`);
|
throw new Error(`Invalid browser state name: "${stateName}"`);
|
||||||
}
|
}
|
||||||
|
|
||||||
return sanitized;
|
return sanitized;
|
||||||
}
|
}
|
||||||
|
|
||||||
function ensureBrowserDirs(profilePath: string, stateDir: string): void {
|
function ensureBrowserDirs(profilePath: string, stateDir: string): void {
|
||||||
mkdirSync(profilePath, { recursive: true });
|
mkdirSync(profilePath, { recursive: true });
|
||||||
mkdirSync(stateDir, { recursive: true });
|
mkdirSync(stateDir, { recursive: true });
|
||||||
}
|
}
|
||||||
|
|
||||||
function createBrowserCommandContext(
|
function createBrowserCommandContext(
|
||||||
profilePath: string,
|
profilePath: string,
|
||||||
stateDir: string,
|
stateDir: string,
|
||||||
context: BrowserCommandContextWithoutProfile,
|
context: BrowserCommandContextWithoutProfile,
|
||||||
): BrowserCommandContext {
|
): BrowserCommandContext {
|
||||||
ensureBrowserDirs(profilePath, stateDir);
|
ensureBrowserDirs(profilePath, stateDir);
|
||||||
return {
|
return {
|
||||||
...context,
|
...context,
|
||||||
profilePath,
|
profilePath,
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
|
|
||||||
function buildWaitArgs(input: BrowserToolInput): { args: string[]; status: string } {
|
function buildWaitArgs(input: BrowserToolInput): {
|
||||||
const targets = [
|
args: string[];
|
||||||
input.ref !== undefined ? "ref" : undefined,
|
status: string;
|
||||||
input.url !== undefined ? "url" : undefined,
|
} {
|
||||||
input.text !== undefined ? "text" : undefined,
|
const targets = [
|
||||||
input.ms !== undefined ? "ms" : undefined,
|
input.ref !== undefined ? "ref" : undefined,
|
||||||
input.loadState !== undefined ? "loadState" : undefined,
|
input.url !== undefined ? "url" : undefined,
|
||||||
].filter((target): target is string => target !== undefined);
|
input.text !== undefined ? "text" : undefined,
|
||||||
|
input.ms !== undefined ? "ms" : undefined,
|
||||||
|
input.loadState !== undefined ? "loadState" : undefined,
|
||||||
|
].filter((target): target is string => target !== undefined);
|
||||||
|
|
||||||
if (targets.length !== 1) {
|
if (targets.length !== 1) {
|
||||||
throw new Error("browser wait requires exactly one of ref, url, text, ms, or loadState");
|
throw new Error(
|
||||||
}
|
"browser wait requires exactly one of ref, url, text, ms, or loadState",
|
||||||
|
);
|
||||||
|
}
|
||||||
|
|
||||||
if (input.ref !== undefined) {
|
if (input.ref !== undefined) {
|
||||||
return { args: ["wait", input.ref], status: `Waiting for ${input.ref}...` };
|
return { args: ["wait", input.ref], status: `Waiting for ${input.ref}...` };
|
||||||
}
|
}
|
||||||
if (input.url !== undefined) {
|
if (input.url !== undefined) {
|
||||||
return {
|
return {
|
||||||
args: ["wait", "--url", input.url],
|
args: ["wait", "--url", input.url],
|
||||||
status: `Waiting for URL ${input.url}...`,
|
status: `Waiting for URL ${input.url}...`,
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
if (input.text !== undefined) {
|
if (input.text !== undefined) {
|
||||||
return {
|
return {
|
||||||
args: ["wait", "--text", input.text],
|
args: ["wait", "--text", input.text],
|
||||||
status: `Waiting for text "${input.text}"...`,
|
status: `Waiting for text "${input.text}"...`,
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
if (input.ms !== undefined) {
|
if (input.ms !== undefined) {
|
||||||
return {
|
return {
|
||||||
args: ["wait", String(input.ms)],
|
args: ["wait", String(input.ms)],
|
||||||
status: `Waiting ${input.ms}ms...`,
|
status: `Waiting ${input.ms}ms...`,
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
|
|
||||||
return {
|
return {
|
||||||
args: ["wait", "--load", input.loadState!],
|
args: ["wait", "--load", input.loadState!],
|
||||||
status: `Waiting for load state ${input.loadState}...`,
|
status: `Waiting for load state ${input.loadState}...`,
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
|
|
||||||
function buildBrowserCommand(
|
function buildBrowserCommand(
|
||||||
cwd: string,
|
cwd: string,
|
||||||
input: BrowserToolInput,
|
input: BrowserToolInput,
|
||||||
options?: BrowserToolOptions,
|
options?: BrowserToolOptions,
|
||||||
): BrowserCommandContext {
|
): BrowserCommandContext {
|
||||||
const profilePath = getBrowserProfilePath(cwd, options);
|
const profilePath = getBrowserProfilePath(cwd, options);
|
||||||
const stateDir = getBrowserStateDir(cwd, options);
|
const stateDir = getBrowserStateDir(cwd, options);
|
||||||
const baseArgs = ["--profile", profilePath];
|
const baseArgs = ["--profile", profilePath];
|
||||||
|
|
||||||
switch (input.action) {
|
switch (input.action) {
|
||||||
case "open": {
|
case "open": {
|
||||||
if (!input.url) {
|
if (!input.url) {
|
||||||
throw new Error("browser open requires url");
|
throw new Error("browser open requires url");
|
||||||
}
|
}
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
action: input.action,
|
action: input.action,
|
||||||
args: [...baseArgs, "open", input.url],
|
args: [...baseArgs, "open", input.url],
|
||||||
statusMessage: `Opening ${input.url}...`,
|
statusMessage: `Opening ${input.url}...`,
|
||||||
successMessage: `Opened ${input.url}`,
|
successMessage: `Opened ${input.url}`,
|
||||||
});
|
});
|
||||||
}
|
}
|
||||||
case "snapshot": {
|
case "snapshot": {
|
||||||
const mode = input.mode ?? "interactive";
|
const mode = input.mode ?? "interactive";
|
||||||
const args = mode === "interactive" ? [...baseArgs, "snapshot", "-i"] : [...baseArgs, "snapshot"];
|
const args =
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
mode === "interactive"
|
||||||
action: input.action,
|
? [...baseArgs, "snapshot", "-i"]
|
||||||
args,
|
: [...baseArgs, "snapshot"];
|
||||||
statusMessage: "Capturing browser snapshot...",
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
successMessage: "Captured browser snapshot",
|
action: input.action,
|
||||||
});
|
args,
|
||||||
}
|
statusMessage: "Capturing browser snapshot...",
|
||||||
case "click": {
|
successMessage: "Captured browser snapshot",
|
||||||
if (!input.ref) {
|
});
|
||||||
throw new Error("browser click requires ref");
|
}
|
||||||
}
|
case "click": {
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
if (!input.ref) {
|
||||||
action: input.action,
|
throw new Error("browser click requires ref");
|
||||||
args: [...baseArgs, "click", input.ref],
|
}
|
||||||
statusMessage: `Clicking ${input.ref}...`,
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
successMessage: `Clicked ${input.ref}`,
|
action: input.action,
|
||||||
});
|
args: [...baseArgs, "click", input.ref],
|
||||||
}
|
statusMessage: `Clicking ${input.ref}...`,
|
||||||
case "fill": {
|
successMessage: `Clicked ${input.ref}`,
|
||||||
if (!input.ref || input.value === undefined) {
|
});
|
||||||
throw new Error("browser fill requires ref and value");
|
}
|
||||||
}
|
case "fill": {
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
if (!input.ref || input.value === undefined) {
|
||||||
action: input.action,
|
throw new Error("browser fill requires ref and value");
|
||||||
args: [...baseArgs, "fill", input.ref, input.value],
|
}
|
||||||
statusMessage: `Filling ${input.ref}...`,
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
successMessage: `Filled ${input.ref}`,
|
action: input.action,
|
||||||
});
|
args: [...baseArgs, "fill", input.ref, input.value],
|
||||||
}
|
statusMessage: `Filling ${input.ref}...`,
|
||||||
case "wait": {
|
successMessage: `Filled ${input.ref}`,
|
||||||
const wait = buildWaitArgs(input);
|
});
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
}
|
||||||
action: input.action,
|
case "wait": {
|
||||||
args: [...baseArgs, ...wait.args],
|
const wait = buildWaitArgs(input);
|
||||||
statusMessage: wait.status,
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
successMessage: "Browser wait condition satisfied",
|
action: input.action,
|
||||||
});
|
args: [...baseArgs, ...wait.args],
|
||||||
}
|
statusMessage: wait.status,
|
||||||
case "screenshot": {
|
successMessage: "Browser wait condition satisfied",
|
||||||
const screenshotPath = input.path ? resolveCommandPath(cwd, input.path) : createTempScreenshotPath();
|
});
|
||||||
const args = [...baseArgs, "screenshot"];
|
}
|
||||||
if (input.fullPage) {
|
case "screenshot": {
|
||||||
args.push("--full");
|
const screenshotPath = input.path
|
||||||
}
|
? resolveCommandPath(cwd, input.path)
|
||||||
args.push(screenshotPath);
|
: createTempScreenshotPath();
|
||||||
|
const args = [...baseArgs, "screenshot"];
|
||||||
|
if (input.fullPage) {
|
||||||
|
args.push("--full");
|
||||||
|
}
|
||||||
|
args.push(screenshotPath);
|
||||||
|
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
action: input.action,
|
action: input.action,
|
||||||
args,
|
args,
|
||||||
statusMessage: "Taking browser screenshot...",
|
statusMessage: "Taking browser screenshot...",
|
||||||
successMessage: `Saved browser screenshot to ${screenshotPath}`,
|
successMessage: `Saved browser screenshot to ${screenshotPath}`,
|
||||||
screenshotPath,
|
screenshotPath,
|
||||||
});
|
});
|
||||||
}
|
}
|
||||||
case "state_save": {
|
case "state_save": {
|
||||||
if (!input.stateName) {
|
if (!input.stateName) {
|
||||||
throw new Error("browser state_save requires stateName");
|
throw new Error("browser state_save requires stateName");
|
||||||
}
|
}
|
||||||
const statePath = join(stateDir, `${sanitizeStateName(input.stateName)}.json`);
|
const statePath = join(
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
stateDir,
|
||||||
action: input.action,
|
`${sanitizeStateName(input.stateName)}.json`,
|
||||||
args: [...baseArgs, "state", "save", statePath],
|
);
|
||||||
statusMessage: `Saving browser state "${input.stateName}"...`,
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
successMessage: `Saved browser state "${input.stateName}" to ${statePath}`,
|
action: input.action,
|
||||||
statePath,
|
args: [...baseArgs, "state", "save", statePath],
|
||||||
});
|
statusMessage: `Saving browser state "${input.stateName}"...`,
|
||||||
}
|
successMessage: `Saved browser state "${input.stateName}" to ${statePath}`,
|
||||||
case "state_load": {
|
statePath,
|
||||||
if (!input.stateName) {
|
});
|
||||||
throw new Error("browser state_load requires stateName");
|
}
|
||||||
}
|
case "state_load": {
|
||||||
const statePath = join(stateDir, `${sanitizeStateName(input.stateName)}.json`);
|
if (!input.stateName) {
|
||||||
if (!existsSync(statePath)) {
|
throw new Error("browser state_load requires stateName");
|
||||||
throw new Error(`Saved browser state "${input.stateName}" not found at ${statePath}`);
|
}
|
||||||
}
|
const statePath = join(
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
stateDir,
|
||||||
action: input.action,
|
`${sanitizeStateName(input.stateName)}.json`,
|
||||||
args: [...baseArgs, "state", "load", statePath],
|
);
|
||||||
statusMessage: `Loading browser state "${input.stateName}"...`,
|
if (!existsSync(statePath)) {
|
||||||
successMessage: `Loaded browser state "${input.stateName}" from ${statePath}`,
|
throw new Error(
|
||||||
statePath,
|
`Saved browser state "${input.stateName}" not found at ${statePath}`,
|
||||||
});
|
);
|
||||||
}
|
}
|
||||||
case "close":
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
return createBrowserCommandContext(profilePath, stateDir, {
|
action: input.action,
|
||||||
action: input.action,
|
args: [...baseArgs, "state", "load", statePath],
|
||||||
args: [...baseArgs, "close"],
|
statusMessage: `Loading browser state "${input.stateName}"...`,
|
||||||
statusMessage: "Closing browser...",
|
successMessage: `Loaded browser state "${input.stateName}" from ${statePath}`,
|
||||||
successMessage: "Closed browser",
|
statePath,
|
||||||
});
|
});
|
||||||
default: {
|
}
|
||||||
const unsupportedAction: never = input.action;
|
case "close":
|
||||||
throw new Error(`Unsupported browser action: ${unsupportedAction}`);
|
return createBrowserCommandContext(profilePath, stateDir, {
|
||||||
}
|
action: input.action,
|
||||||
}
|
args: [...baseArgs, "close"],
|
||||||
|
statusMessage: "Closing browser...",
|
||||||
|
successMessage: "Closed browser",
|
||||||
|
});
|
||||||
|
default: {
|
||||||
|
const unsupportedAction: never = input.action;
|
||||||
|
throw new Error(`Unsupported browser action: ${unsupportedAction}`);
|
||||||
|
}
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
function buildBrowserErrorMessage(action: BrowserToolAction, output: string, exitCode: number | null): string {
|
function buildBrowserErrorMessage(
|
||||||
const base =
|
action: BrowserToolAction,
|
||||||
exitCode === null
|
output: string,
|
||||||
? `Browser action "${action}" failed`
|
exitCode: number | null,
|
||||||
: `Browser action "${action}" exited with code ${exitCode}`;
|
): string {
|
||||||
return output.length > 0 ? `${output}\n\n${base}` : base;
|
const base =
|
||||||
|
exitCode === null
|
||||||
|
? `Browser action "${action}" failed`
|
||||||
|
: `Browser action "${action}" exited with code ${exitCode}`;
|
||||||
|
return output.length > 0 ? `${output}\n\n${base}` : base;
|
||||||
}
|
}
|
||||||
|
|
||||||
function getMissingBrowserCommandMessage(command: string): string {
|
function getMissingBrowserCommandMessage(command: string): string {
|
||||||
return [
|
return [
|
||||||
`Browser tool could not find "${command}".`,
|
`Browser tool could not find "${command}".`,
|
||||||
"Install agent-browser so the first-class browser tool can run.",
|
"Install agent-browser so the first-class browser tool can run.",
|
||||||
"Recommended setup:",
|
"Recommended setup:",
|
||||||
" npm install -g agent-browser",
|
" npm install -g agent-browser",
|
||||||
" agent-browser install",
|
" agent-browser install",
|
||||||
"If Chromium lives at a custom path, set AGENT_BROWSER_EXECUTABLE_PATH.",
|
"If Chromium lives at a custom path, set AGENT_BROWSER_EXECUTABLE_PATH.",
|
||||||
].join("\n");
|
].join("\n");
|
||||||
}
|
}
|
||||||
|
|
||||||
export function createBrowserTool(cwd: string, options?: BrowserToolOptions): AgentTool<typeof browserSchema> {
|
export function createBrowserTool(
|
||||||
const operations = options?.operations ?? defaultBrowserOperations;
|
cwd: string,
|
||||||
const command = options?.command ?? DEFAULT_BROWSER_COMMAND;
|
options?: BrowserToolOptions,
|
||||||
const defaultTimeoutSeconds = options?.defaultTimeoutSeconds ?? DEFAULT_BROWSER_TIMEOUT_SECONDS;
|
): AgentTool<typeof browserSchema> {
|
||||||
|
const operations = options?.operations ?? defaultBrowserOperations;
|
||||||
|
const command = options?.command ?? DEFAULT_BROWSER_COMMAND;
|
||||||
|
const defaultTimeoutSeconds =
|
||||||
|
options?.defaultTimeoutSeconds ?? DEFAULT_BROWSER_TIMEOUT_SECONDS;
|
||||||
|
|
||||||
return {
|
return {
|
||||||
name: "browser",
|
name: "browser",
|
||||||
label: "browser",
|
label: "browser",
|
||||||
description:
|
description:
|
||||||
"Use a persistent browser for websites: open pages, inspect them with snapshot, click or fill elements, wait for changes, take screenshots, and save or load named browser state.",
|
"Use a persistent browser for websites: open pages, inspect them with snapshot, click or fill elements, wait for changes, take screenshots, and save or load named browser state.",
|
||||||
parameters: browserSchema,
|
parameters: browserSchema,
|
||||||
execute: async (_toolCallId, input, signal, onUpdate) => {
|
execute: async (_toolCallId, input, signal, onUpdate) => {
|
||||||
const commandContext = buildBrowserCommand(cwd, input, options);
|
const commandContext = buildBrowserCommand(cwd, input, options);
|
||||||
const details: BrowserToolDetails = {
|
const details: BrowserToolDetails = {
|
||||||
action: commandContext.action,
|
action: commandContext.action,
|
||||||
command,
|
command,
|
||||||
args: commandContext.args,
|
args: commandContext.args,
|
||||||
profilePath: commandContext.profilePath,
|
profilePath: commandContext.profilePath,
|
||||||
screenshotPath: commandContext.screenshotPath,
|
screenshotPath: commandContext.screenshotPath,
|
||||||
statePath: commandContext.statePath,
|
statePath: commandContext.statePath,
|
||||||
};
|
};
|
||||||
|
|
||||||
onUpdate?.({
|
onUpdate?.({
|
||||||
content: [{ type: "text", text: commandContext.statusMessage }],
|
content: [{ type: "text", text: commandContext.statusMessage }],
|
||||||
details,
|
details,
|
||||||
});
|
});
|
||||||
|
|
||||||
const chunks: Buffer[] = [];
|
const chunks: Buffer[] = [];
|
||||||
|
|
||||||
try {
|
try {
|
||||||
const { exitCode } = await operations.exec(command, commandContext.args, {
|
const { exitCode } = await operations.exec(
|
||||||
cwd,
|
command,
|
||||||
env: getShellEnv(),
|
commandContext.args,
|
||||||
onData: (data) => chunks.push(data),
|
{
|
||||||
signal,
|
cwd,
|
||||||
timeout: defaultTimeoutSeconds,
|
env: getShellEnv(),
|
||||||
});
|
onData: (data) => chunks.push(data),
|
||||||
|
signal,
|
||||||
|
timeout: defaultTimeoutSeconds,
|
||||||
|
},
|
||||||
|
);
|
||||||
|
|
||||||
const output = normalizeOutput(chunks);
|
const output = normalizeOutput(chunks);
|
||||||
if (exitCode !== 0) {
|
if (exitCode !== 0) {
|
||||||
throw new Error(buildBrowserErrorMessage(commandContext.action, output, exitCode));
|
throw new Error(
|
||||||
}
|
buildBrowserErrorMessage(commandContext.action, output, exitCode),
|
||||||
|
);
|
||||||
|
}
|
||||||
|
|
||||||
if (commandContext.action === "snapshot") {
|
if (commandContext.action === "snapshot") {
|
||||||
if (output.length === 0) {
|
if (output.length === 0) {
|
||||||
throw new Error("Browser snapshot returned no output");
|
throw new Error("Browser snapshot returned no output");
|
||||||
}
|
}
|
||||||
return {
|
return {
|
||||||
content: [{ type: "text", text: output }],
|
content: [{ type: "text", text: output }],
|
||||||
details,
|
details,
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
|
|
||||||
const text = output.length > 0 ? output : commandContext.successMessage;
|
const text = output.length > 0 ? output : commandContext.successMessage;
|
||||||
return {
|
return {
|
||||||
content: [{ type: "text", text }],
|
content: [{ type: "text", text }],
|
||||||
details,
|
details,
|
||||||
};
|
};
|
||||||
} catch (error) {
|
} catch (error) {
|
||||||
if (error instanceof Error && "code" in error && error.code === "ENOENT") {
|
if (
|
||||||
throw new Error(getMissingBrowserCommandMessage(command));
|
error instanceof Error &&
|
||||||
}
|
"code" in error &&
|
||||||
if (error instanceof Error && error.message === "aborted") {
|
error.code === "ENOENT"
|
||||||
throw new Error(`Browser action "${commandContext.action}" aborted`);
|
) {
|
||||||
}
|
throw new Error(getMissingBrowserCommandMessage(command));
|
||||||
if (error instanceof Error && error.message.startsWith("timeout:")) {
|
}
|
||||||
const seconds = error.message.split(":")[1];
|
if (error instanceof Error && error.message === "aborted") {
|
||||||
throw new Error(`Browser action "${commandContext.action}" timed out after ${seconds} seconds`);
|
throw new Error(`Browser action "${commandContext.action}" aborted`);
|
||||||
}
|
}
|
||||||
throw error;
|
if (error instanceof Error && error.message.startsWith("timeout:")) {
|
||||||
}
|
const seconds = error.message.split(":")[1];
|
||||||
},
|
throw new Error(
|
||||||
};
|
`Browser action "${commandContext.action}" timed out after ${seconds} seconds`,
|
||||||
|
);
|
||||||
|
}
|
||||||
|
throw error;
|
||||||
|
}
|
||||||
|
},
|
||||||
|
};
|
||||||
}
|
}
|
||||||
|
|
||||||
export const browserTool = createBrowserTool(process.cwd());
|
export const browserTool = createBrowserTool(process.cwd());
|
||||||
|
|
|
||||||
|
|
@ -5,270 +5,288 @@ import { afterEach, describe, expect, it } from "vitest";
|
||||||
import { parseArgs } from "../src/cli/args.js";
|
import { parseArgs } from "../src/cli/args.js";
|
||||||
import { buildSystemPrompt } from "../src/core/system-prompt.js";
|
import { buildSystemPrompt } from "../src/core/system-prompt.js";
|
||||||
import {
|
import {
|
||||||
type BrowserOperations,
|
type BrowserOperations,
|
||||||
type BrowserToolDetails,
|
type BrowserToolDetails,
|
||||||
createAllTools,
|
createAllTools,
|
||||||
createBrowserTool,
|
createBrowserTool,
|
||||||
defaultCodingToolNames,
|
defaultCodingToolNames,
|
||||||
} from "../src/core/tools/index.js";
|
} from "../src/core/tools/index.js";
|
||||||
|
|
||||||
interface TextBlock {
|
interface TextBlock {
|
||||||
type: "text";
|
type: "text";
|
||||||
text: string;
|
text: string;
|
||||||
}
|
}
|
||||||
|
|
||||||
type ToolContentBlock = TextBlock | { type: string };
|
type ToolContentBlock = TextBlock | { type: string };
|
||||||
|
|
||||||
interface ToolResultLike {
|
interface ToolResultLike {
|
||||||
content: ToolContentBlock[];
|
content: ToolContentBlock[];
|
||||||
details?: unknown;
|
details?: unknown;
|
||||||
}
|
}
|
||||||
|
|
||||||
interface BrowserExecCall {
|
interface BrowserExecCall {
|
||||||
command: string;
|
command: string;
|
||||||
args: string[];
|
args: string[];
|
||||||
cwd: string;
|
cwd: string;
|
||||||
env: NodeJS.ProcessEnv;
|
env: NodeJS.ProcessEnv;
|
||||||
timeout?: number;
|
timeout?: number;
|
||||||
}
|
}
|
||||||
|
|
||||||
function getTextOutput(result: ToolResultLike): string {
|
function getTextOutput(result: ToolResultLike): string {
|
||||||
return result.content
|
return result.content
|
||||||
.filter((block): block is TextBlock => block.type === "text")
|
.filter((block): block is TextBlock => block.type === "text")
|
||||||
.map((block) => block.text)
|
.map((block) => block.text)
|
||||||
.join("\n");
|
.join("\n");
|
||||||
}
|
}
|
||||||
|
|
||||||
function createMockBrowserOperations(
|
function createMockBrowserOperations(
|
||||||
output = "",
|
output = "",
|
||||||
exitCode: number | null = 0,
|
exitCode: number | null = 0,
|
||||||
): {
|
): {
|
||||||
calls: BrowserExecCall[];
|
calls: BrowserExecCall[];
|
||||||
operations: BrowserOperations;
|
operations: BrowserOperations;
|
||||||
} {
|
} {
|
||||||
const calls: BrowserExecCall[] = [];
|
const calls: BrowserExecCall[] = [];
|
||||||
|
|
||||||
return {
|
return {
|
||||||
calls,
|
calls,
|
||||||
operations: {
|
operations: {
|
||||||
exec: async (command, args, options) => {
|
exec: async (command, args, options) => {
|
||||||
calls.push({
|
calls.push({
|
||||||
command,
|
command,
|
||||||
args,
|
args,
|
||||||
cwd: options.cwd,
|
cwd: options.cwd,
|
||||||
env: options.env,
|
env: options.env,
|
||||||
timeout: options.timeout,
|
timeout: options.timeout,
|
||||||
});
|
});
|
||||||
if (output.length > 0) {
|
if (output.length > 0) {
|
||||||
options.onData(Buffer.from(output, "utf-8"));
|
options.onData(Buffer.from(output, "utf-8"));
|
||||||
}
|
}
|
||||||
return { exitCode };
|
return { exitCode };
|
||||||
},
|
},
|
||||||
},
|
},
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
|
|
||||||
describe("browser tool", () => {
|
describe("browser tool", () => {
|
||||||
const tempDirs: string[] = [];
|
const tempDirs: string[] = [];
|
||||||
|
|
||||||
afterEach(() => {
|
afterEach(() => {
|
||||||
while (tempDirs.length > 0) {
|
while (tempDirs.length > 0) {
|
||||||
const tempDir = tempDirs.pop();
|
const tempDir = tempDirs.pop();
|
||||||
if (tempDir) {
|
if (tempDir) {
|
||||||
rmSync(tempDir, { recursive: true, force: true });
|
rmSync(tempDir, { recursive: true, force: true });
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
});
|
});
|
||||||
|
|
||||||
function createTempDir(prefix: string): string {
|
function createTempDir(prefix: string): string {
|
||||||
const tempDir = mkdtempSync(join(tmpdir(), prefix));
|
const tempDir = mkdtempSync(join(tmpdir(), prefix));
|
||||||
tempDirs.push(tempDir);
|
tempDirs.push(tempDir);
|
||||||
return tempDir;
|
return tempDir;
|
||||||
}
|
}
|
||||||
|
|
||||||
it("opens pages through agent-browser with a shared profile", async () => {
|
it("opens pages through agent-browser with a shared profile", async () => {
|
||||||
const cwd = createTempDir("coding-agent-browser-open-");
|
const cwd = createTempDir("coding-agent-browser-open-");
|
||||||
const profileDir = join(cwd, "profile");
|
const profileDir = join(cwd, "profile");
|
||||||
const stateDir = join(cwd, "states");
|
const stateDir = join(cwd, "states");
|
||||||
const { calls, operations } = createMockBrowserOperations();
|
const { calls, operations } = createMockBrowserOperations();
|
||||||
|
|
||||||
const browserTool = createBrowserTool(cwd, {
|
const browserTool = createBrowserTool(cwd, {
|
||||||
operations,
|
operations,
|
||||||
command: "agent-browser-test",
|
command: "agent-browser-test",
|
||||||
profileDir,
|
profileDir,
|
||||||
stateDir,
|
stateDir,
|
||||||
});
|
});
|
||||||
|
|
||||||
const result = (await browserTool.execute("browser-open", {
|
const result = (await browserTool.execute("browser-open", {
|
||||||
action: "open",
|
action: "open",
|
||||||
url: "https://example.com",
|
url: "https://example.com",
|
||||||
})) as ToolResultLike;
|
})) as ToolResultLike;
|
||||||
|
|
||||||
expect(calls).toHaveLength(1);
|
expect(calls).toHaveLength(1);
|
||||||
expect(calls[0]).toMatchObject({
|
expect(calls[0]).toMatchObject({
|
||||||
command: "agent-browser-test",
|
command: "agent-browser-test",
|
||||||
args: ["--profile", profileDir, "open", "https://example.com"],
|
args: ["--profile", profileDir, "open", "https://example.com"],
|
||||||
cwd,
|
cwd,
|
||||||
timeout: 90,
|
timeout: 90,
|
||||||
});
|
});
|
||||||
expect(getTextOutput(result)).toBe("Opened https://example.com");
|
expect(getTextOutput(result)).toBe("Opened https://example.com");
|
||||||
|
|
||||||
const details = result.details as BrowserToolDetails | undefined;
|
const details = result.details as BrowserToolDetails | undefined;
|
||||||
expect(details?.profilePath).toBe(profileDir);
|
expect(details?.profilePath).toBe(profileDir);
|
||||||
});
|
});
|
||||||
|
|
||||||
it("uses interactive snapshots by default and returns snapshot text", async () => {
|
it("uses interactive snapshots by default and returns snapshot text", async () => {
|
||||||
const cwd = createTempDir("coding-agent-browser-snapshot-");
|
const cwd = createTempDir("coding-agent-browser-snapshot-");
|
||||||
const profileDir = join(cwd, "profile");
|
const profileDir = join(cwd, "profile");
|
||||||
const stateDir = join(cwd, "states");
|
const stateDir = join(cwd, "states");
|
||||||
const { calls, operations } = createMockBrowserOperations("main [ref=@e1]\nbutton [ref=@e2] Sign in");
|
const { calls, operations } = createMockBrowserOperations(
|
||||||
|
"main [ref=@e1]\nbutton [ref=@e2] Sign in",
|
||||||
|
);
|
||||||
|
|
||||||
const browserTool = createBrowserTool(cwd, {
|
const browserTool = createBrowserTool(cwd, {
|
||||||
operations,
|
operations,
|
||||||
profileDir,
|
profileDir,
|
||||||
stateDir,
|
stateDir,
|
||||||
});
|
});
|
||||||
|
|
||||||
const result = (await browserTool.execute("browser-snapshot", {
|
const result = (await browserTool.execute("browser-snapshot", {
|
||||||
action: "snapshot",
|
action: "snapshot",
|
||||||
})) as ToolResultLike;
|
})) as ToolResultLike;
|
||||||
|
|
||||||
expect(calls[0]?.args).toEqual(["--profile", profileDir, "snapshot", "-i"]);
|
expect(calls[0]?.args).toEqual(["--profile", profileDir, "snapshot", "-i"]);
|
||||||
expect(getTextOutput(result)).toContain("button [ref=@e2] Sign in");
|
expect(getTextOutput(result)).toContain("button [ref=@e2] Sign in");
|
||||||
});
|
});
|
||||||
|
|
||||||
it("validates wait targets before spawning agent-browser", async () => {
|
it("validates wait targets before spawning agent-browser", async () => {
|
||||||
const cwd = createTempDir("coding-agent-browser-wait-");
|
const cwd = createTempDir("coding-agent-browser-wait-");
|
||||||
const profileDir = join(cwd, "profile");
|
const profileDir = join(cwd, "profile");
|
||||||
const stateDir = join(cwd, "states");
|
const stateDir = join(cwd, "states");
|
||||||
const { calls, operations } = createMockBrowserOperations();
|
const { calls, operations } = createMockBrowserOperations();
|
||||||
|
|
||||||
const browserTool = createBrowserTool(cwd, {
|
const browserTool = createBrowserTool(cwd, {
|
||||||
operations,
|
operations,
|
||||||
profileDir,
|
profileDir,
|
||||||
stateDir,
|
stateDir,
|
||||||
});
|
});
|
||||||
|
|
||||||
await expect(
|
await expect(
|
||||||
browserTool.execute("browser-wait-missing", {
|
browserTool.execute("browser-wait-missing", {
|
||||||
action: "wait",
|
action: "wait",
|
||||||
}),
|
}),
|
||||||
).rejects.toThrow("browser wait requires exactly one of ref, url, text, ms, or loadState");
|
).rejects.toThrow(
|
||||||
|
"browser wait requires exactly one of ref, url, text, ms, or loadState",
|
||||||
|
);
|
||||||
|
|
||||||
await expect(
|
await expect(
|
||||||
browserTool.execute("browser-wait-ambiguous", {
|
browserTool.execute("browser-wait-ambiguous", {
|
||||||
action: "wait",
|
action: "wait",
|
||||||
ref: "@e2",
|
ref: "@e2",
|
||||||
text: "Done",
|
text: "Done",
|
||||||
}),
|
}),
|
||||||
).rejects.toThrow("browser wait requires exactly one of ref, url, text, ms, or loadState");
|
).rejects.toThrow(
|
||||||
|
"browser wait requires exactly one of ref, url, text, ms, or loadState",
|
||||||
|
);
|
||||||
|
|
||||||
expect(calls).toHaveLength(0);
|
expect(calls).toHaveLength(0);
|
||||||
});
|
});
|
||||||
|
|
||||||
it("preserves empty string wait targets instead of falling through to loadState", async () => {
|
it("preserves empty string wait targets instead of falling through to loadState", async () => {
|
||||||
const cwd = createTempDir("coding-agent-browser-wait-empty-");
|
const cwd = createTempDir("coding-agent-browser-wait-empty-");
|
||||||
const profileDir = join(cwd, "profile");
|
const profileDir = join(cwd, "profile");
|
||||||
const stateDir = join(cwd, "states");
|
const stateDir = join(cwd, "states");
|
||||||
const { calls, operations } = createMockBrowserOperations();
|
const { calls, operations } = createMockBrowserOperations();
|
||||||
|
|
||||||
const browserTool = createBrowserTool(cwd, {
|
const browserTool = createBrowserTool(cwd, {
|
||||||
operations,
|
operations,
|
||||||
profileDir,
|
profileDir,
|
||||||
stateDir,
|
stateDir,
|
||||||
});
|
});
|
||||||
|
|
||||||
await browserTool.execute("browser-wait-empty-text", {
|
await browserTool.execute("browser-wait-empty-text", {
|
||||||
action: "wait",
|
action: "wait",
|
||||||
text: "",
|
text: "",
|
||||||
});
|
});
|
||||||
|
|
||||||
expect(calls[0]?.args).toEqual(["--profile", profileDir, "wait", "--text", ""]);
|
expect(calls[0]?.args).toEqual([
|
||||||
});
|
"--profile",
|
||||||
|
profileDir,
|
||||||
|
"wait",
|
||||||
|
"--text",
|
||||||
|
"",
|
||||||
|
]);
|
||||||
|
});
|
||||||
|
|
||||||
it("does not create browser directories when validation fails before command construction", async () => {
|
it("does not create browser directories when validation fails before command construction", async () => {
|
||||||
const cwd = createTempDir("coding-agent-browser-invalid-open-");
|
const cwd = createTempDir("coding-agent-browser-invalid-open-");
|
||||||
const profileDir = join(cwd, "profile");
|
const profileDir = join(cwd, "profile");
|
||||||
const stateDir = join(cwd, "states");
|
const stateDir = join(cwd, "states");
|
||||||
const { operations } = createMockBrowserOperations();
|
const { operations } = createMockBrowserOperations();
|
||||||
|
|
||||||
const browserTool = createBrowserTool(cwd, {
|
const browserTool = createBrowserTool(cwd, {
|
||||||
operations,
|
operations,
|
||||||
profileDir,
|
profileDir,
|
||||||
stateDir,
|
stateDir,
|
||||||
});
|
});
|
||||||
|
|
||||||
await expect(
|
await expect(
|
||||||
browserTool.execute("browser-open-missing-url", {
|
browserTool.execute("browser-open-missing-url", {
|
||||||
action: "open",
|
action: "open",
|
||||||
}),
|
}),
|
||||||
).rejects.toThrow("browser open requires url");
|
).rejects.toThrow("browser open requires url");
|
||||||
|
|
||||||
expect(existsSync(profileDir)).toBe(false);
|
expect(existsSync(profileDir)).toBe(false);
|
||||||
expect(existsSync(stateDir)).toBe(false);
|
expect(existsSync(stateDir)).toBe(false);
|
||||||
});
|
});
|
||||||
|
|
||||||
it("stores named state under the managed browser state directory", async () => {
|
it("stores named state under the managed browser state directory", async () => {
|
||||||
const cwd = createTempDir("coding-agent-browser-state-");
|
const cwd = createTempDir("coding-agent-browser-state-");
|
||||||
const profileDir = join(cwd, "profile");
|
const profileDir = join(cwd, "profile");
|
||||||
const stateDir = join(cwd, "states");
|
const stateDir = join(cwd, "states");
|
||||||
const { calls, operations } = createMockBrowserOperations();
|
const { calls, operations } = createMockBrowserOperations();
|
||||||
|
|
||||||
const browserTool = createBrowserTool(cwd, {
|
const browserTool = createBrowserTool(cwd, {
|
||||||
operations,
|
operations,
|
||||||
profileDir,
|
profileDir,
|
||||||
stateDir,
|
stateDir,
|
||||||
});
|
});
|
||||||
|
|
||||||
const result = (await browserTool.execute("browser-state-save", {
|
const result = (await browserTool.execute("browser-state-save", {
|
||||||
action: "state_save",
|
action: "state_save",
|
||||||
stateName: "my session/prod",
|
stateName: "my session/prod",
|
||||||
})) as ToolResultLike;
|
})) as ToolResultLike;
|
||||||
|
|
||||||
const expectedStatePath = join(stateDir, "my-session-prod.json");
|
const expectedStatePath = join(stateDir, "my-session-prod.json");
|
||||||
expect(calls[0]?.args).toEqual(["--profile", profileDir, "state", "save", expectedStatePath]);
|
expect(calls[0]?.args).toEqual([
|
||||||
|
"--profile",
|
||||||
|
profileDir,
|
||||||
|
"state",
|
||||||
|
"save",
|
||||||
|
expectedStatePath,
|
||||||
|
]);
|
||||||
|
|
||||||
const details = result.details as BrowserToolDetails | undefined;
|
const details = result.details as BrowserToolDetails | undefined;
|
||||||
expect(details?.statePath).toBe(expectedStatePath);
|
expect(details?.statePath).toBe(expectedStatePath);
|
||||||
expect(getTextOutput(result)).toContain(expectedStatePath);
|
expect(getTextOutput(result)).toContain(expectedStatePath);
|
||||||
});
|
});
|
||||||
|
|
||||||
it("treats null exit codes as browser failures", async () => {
|
it("treats null exit codes as browser failures", async () => {
|
||||||
const cwd = createTempDir("coding-agent-browser-null-exit-");
|
const cwd = createTempDir("coding-agent-browser-null-exit-");
|
||||||
const profileDir = join(cwd, "profile");
|
const profileDir = join(cwd, "profile");
|
||||||
const stateDir = join(cwd, "states");
|
const stateDir = join(cwd, "states");
|
||||||
const { operations } = createMockBrowserOperations("browser crashed", null);
|
const { operations } = createMockBrowserOperations("browser crashed", null);
|
||||||
|
|
||||||
const browserTool = createBrowserTool(cwd, {
|
const browserTool = createBrowserTool(cwd, {
|
||||||
operations,
|
operations,
|
||||||
profileDir,
|
profileDir,
|
||||||
stateDir,
|
stateDir,
|
||||||
});
|
});
|
||||||
|
|
||||||
await expect(
|
await expect(
|
||||||
browserTool.execute("browser-open-null-exit", {
|
browserTool.execute("browser-open-null-exit", {
|
||||||
action: "open",
|
action: "open",
|
||||||
url: "https://example.com",
|
url: "https://example.com",
|
||||||
}),
|
}),
|
||||||
).rejects.toThrow('browser crashed\n\nBrowser action "open" failed');
|
).rejects.toThrow('browser crashed\n\nBrowser action "open" failed');
|
||||||
});
|
});
|
||||||
|
|
||||||
it("accepts browser in --tools and exposes it in default tool wiring", () => {
|
it("accepts browser in --tools and exposes it in default tool wiring", () => {
|
||||||
const parsed = parseArgs(["--tools", "browser,read"]);
|
const parsed = parseArgs(["--tools", "browser,read"]);
|
||||||
expect(parsed.tools).toEqual(["browser", "read"]);
|
expect(parsed.tools).toEqual(["browser", "read"]);
|
||||||
|
|
||||||
expect(defaultCodingToolNames).toContain("browser");
|
expect(defaultCodingToolNames).toContain("browser");
|
||||||
expect(createAllTools(process.cwd()).browser.name).toBe("browser");
|
expect(createAllTools(process.cwd()).browser.name).toBe("browser");
|
||||||
});
|
});
|
||||||
|
|
||||||
it("mentions browser in the default system prompt", () => {
|
it("mentions browser in the default system prompt", () => {
|
||||||
const prompt = buildSystemPrompt();
|
const prompt = buildSystemPrompt();
|
||||||
|
|
||||||
expect(prompt).toContain(
|
expect(prompt).toContain(
|
||||||
"- browser: Open websites, inspect pages with snapshot, click/fill/wait, take screenshots, and save/load browser state",
|
"- browser: Open websites, inspect pages with snapshot, click/fill/wait, take screenshots, and save/load browser state",
|
||||||
);
|
);
|
||||||
expect(prompt).toContain(
|
expect(prompt).toContain(
|
||||||
"Use browser for website tasks. Open the page, use snapshot to inspect interactive elements, then click, fill, wait, or screenshot as needed",
|
"Use browser for website tasks. Open the page, use snapshot to inspect interactive elements, then click, fill, wait, or screenshot as needed",
|
||||||
);
|
);
|
||||||
});
|
});
|
||||||
});
|
});
|
||||||
|
|
|
||||||
Loading…
Add table
Add a link
Reference in a new issue