agentperf 0.1.0

package/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/README.md

@@ -0,0 +1,37 @@
+# agentperf
+
+**Measure what AI agents pay to use a website.** Same page, same model, same
+task, same verification — two lanes:
+
+- **dom** — the agent reads the accessibility tree and clicks/types, the way
+  browser-driving agents work today. (Text-based driving is *cheaper* than
+  screenshots, so this baseline is conservative in the DOM lane's favor.)
+- **tools** — the agent calls the page's own WebMCP tools. The harness injects
+  a `document.modelContext` host before page load, so it benchmarks any
+  WebMCP page in stock Chromium — no browser flag, no origin trial.
+
+Success is verified by the harness against the rendered page, never claimed by
+the model. Failed runs are reported, not discarded.
+
+```bash
+npm i -g agentperf
+npx playwright install chromium
+export OPENAI_API_KEY=sk-…            # any OpenAI-compatible endpoint works
+agentperf run --url https://your-app.example --runs 3
+```
+
+Output: per-run and median wall-clock, tokens (prompt+completion), model
+round-trips, actions, success rate — as a markdown table and JSON under
+`results/`.
+
+Options: `--lane both|tools|dom` · `--runs N` · `--model <id>` (default
+`gpt-5.6-luna`) · `--base-url <url>` for compatible providers · `--task
+booking|path/to/task.json` · `--max-turns N` · `--out dir`. A task file is
+`{ id, prompt, successPattern, maxTurns }` — the prompt is handed to both
+lanes verbatim and `successPattern` is a regex the page's visible text must
+match when the task is truly done.
+
+Instrument your app with [`@agentperf/react`](https://www.npmjs.com/package/@agentperf/react)
+to give it the tools lane.
+
+Apache-2.0

package/dist/chunk-BGU3NILX.js

@@ -0,0 +1,395 @@
+// src/llm.ts
+async function chat(config, messages, tools) {
+  const response = await fetch(`${config.baseUrl}/chat/completions`, {
+    method: "POST",
+    headers: {
+      "content-type": "application/json",
+      authorization: `Bearer ${config.apiKey}`
+    },
+    body: JSON.stringify({
+      model: config.model,
+      messages,
+      // gpt-5.6 models 400 on function tools unless reasoning is off.
+      ...config.model.startsWith("gpt-5.6") && { reasoning_effort: "none" },
+      tools: tools.map((tool) => ({
+        type: "function",
+        function: {
+          name: tool.name,
+          description: tool.description,
+          parameters: tool.parameters
+        }
+      })),
+      tool_choice: "auto"
+    })
+  });
+  if (!response.ok) {
+    const body = await response.text();
+    throw new Error(`LLM request failed (${response.status}): ${body.slice(0, 400)}`);
+  }
+  const data = await response.json();
+  const message = data.choices[0]?.message;
+  if (!message) throw new Error("LLM response had no choices");
+  const toolCalls = (message.tool_calls ?? []).map((call) => {
+    let args = {};
+    try {
+      args = JSON.parse(call.function.arguments || "{}");
+    } catch {
+    }
+    return { id: call.id, name: call.function.name, args, raw: call };
+  });
+  return {
+    content: message.content,
+    toolCalls,
+    rawMessage: message,
+    promptTokens: data.usage?.prompt_tokens ?? 0,
+    completionTokens: data.usage?.completion_tokens ?? 0
+  };
+}
+
+// src/agent-loop.ts
+var TASK_COMPLETE = {
+  name: "task_complete",
+  description: "Call this exactly once, when the task is fully done and its outcome is visible on the page. Summarize the outcome (include any code or value the task asked for).",
+  parameters: {
+    type: "object",
+    properties: { summary: { type: "string" } },
+    required: ["summary"]
+  }
+};
+async function runAgentLoop(options) {
+  const messages = [
+    { role: "system", content: options.systemPrompt },
+    { role: "user", content: options.taskPrompt }
+  ];
+  const tools = [...options.tools, TASK_COMPLETE];
+  const outcome = {
+    turns: 0,
+    actions: 0,
+    promptTokens: 0,
+    completionTokens: 0,
+    claimedDone: false,
+    claimSummary: ""
+  };
+  let nudged = false;
+  while (outcome.turns < options.maxTurns) {
+    outcome.turns += 1;
+    const turn = await chat(options.llm, messages, tools);
+    outcome.promptTokens += turn.promptTokens;
+    outcome.completionTokens += turn.completionTokens;
+    messages.push(turn.rawMessage);
+    if (turn.toolCalls.length === 0) {
+      if (nudged) {
+        outcome.failure = "model stopped calling tools before claiming completion";
+        return outcome;
+      }
+      nudged = true;
+      messages.push({
+        role: "user",
+        content: "Continue using the tools. When the task is done, call task_complete."
+      });
+      continue;
+    }
+    for (const call of turn.toolCalls) {
+      if (call.name === TASK_COMPLETE.name) {
+        outcome.claimedDone = true;
+        outcome.claimSummary = String(call.args.summary ?? "");
+        messages.push({ role: "tool", tool_call_id: call.id, content: "Recorded." });
+        return outcome;
+      }
+      outcome.actions += 1;
+      let result;
+      try {
+        result = await options.invoke(call.name, call.args);
+      } catch (error) {
+        result = `ACTION FAILED: ${error instanceof Error ? error.message : String(error)}`;
+      }
+      messages.push({ role: "tool", tool_call_id: call.id, content: result });
+    }
+  }
+  outcome.failure = `hit the ${options.maxTurns}-turn cap without claiming completion`;
+  return outcome;
+}
+
+// src/shim.ts
+var HOST_SHIM = `
+(() => {
+  const tools = new Map();
+  const api = {
+    listTools: () =>
+      Array.from(tools.values()).map((t) => ({
+        name: t.name,
+        description: t.description,
+        inputSchema: t.inputSchema || { type: "object", properties: {} },
+        annotations: t.annotations || {}
+      })),
+    call: async (name, args) => {
+      const tool = tools.get(name);
+      if (!tool) {
+        return {
+          content: [{ type: "text", text: "No tool named \\"" + name + "\\" is registered on this page." }],
+          isError: true
+        };
+      }
+      const result = await tool.execute(args || {});
+      if (typeof result === "string") return { content: [{ type: "text", text: result }] };
+      return result;
+    }
+  };
+  Object.defineProperty(window, "__agentperf", { value: api });
+
+  const modelContext = {
+    registerTool: (tool, options) => {
+      tools.set(tool.name, tool);
+      if (options && options.signal) {
+        options.signal.addEventListener("abort", () => { tools.delete(tool.name); });
+      }
+    },
+    getTools: async () => api.listTools()
+  };
+  Object.defineProperty(document, "modelContext", { value: modelContext, configurable: true });
+  Object.defineProperty(navigator, "modelContext", { value: modelContext, configurable: true });
+})();
+`;
+
+// src/lanes.ts
+import { chromium } from "playwright";
+async function verify(page, task) {
+  const text = await page.evaluate(() => document.body.innerText);
+  return new RegExp(task.successPattern).test(text);
+}
+function metrics(lane, task) {
+  return {
+    lane,
+    taskId: task.id,
+    success: false,
+    wallClockMs: 0,
+    turns: 0,
+    actions: 0,
+    promptTokens: 0,
+    completionTokens: 0,
+    totalTokens: 0
+  };
+}
+function finish(run, startedMs, loop, verified, claimedDone) {
+  run.wallClockMs = Math.round(performance.now() - startedMs);
+  run.turns = loop.turns;
+  run.actions = loop.actions;
+  run.promptTokens = loop.promptTokens;
+  run.completionTokens = loop.completionTokens;
+  run.totalTokens = loop.promptTokens + loop.completionTokens;
+  run.success = verified;
+  if (!verified) {
+    run.failure = loop.failure ?? (claimedDone ? "model claimed completion but the page does not show the expected outcome" : "run ended without completion");
+  }
+  return run;
+}
+async function runToolsLane(url, task, llm) {
+  const browser = await chromium.launch();
+  const run = metrics("tools", task);
+  const startedMs = performance.now();
+  try {
+    const context = await browser.newContext();
+    await context.addInitScript(HOST_SHIM);
+    const page = await context.newPage();
+    await page.goto(url, { waitUntil: "load" });
+    await page.waitForFunction(() => window.__agentperf?.listTools().length > 0, void 0, {
+      timeout: 15e3
+    });
+    const pageTools = await page.evaluate(() => window.__agentperf.listTools());
+    const tools = pageTools.map((tool) => ({
+      name: tool.name,
+      description: tool.description,
+      parameters: tool.inputSchema
+    }));
+    const loop = await runAgentLoop({
+      llm,
+      systemPrompt: "You complete tasks on a web page through the page's own tools (WebMCP). Call get_page_state first to orient when it exists. Use only the tools. When the task is fully done, call task_complete.",
+      taskPrompt: task.prompt,
+      tools,
+      maxTurns: task.maxTurns,
+      invoke: async (name, args) => {
+        const result = await page.evaluate(
+          ([toolName, toolArgs]) => window.__agentperf.call(
+            toolName,
+            toolArgs
+          ),
+          [name, args]
+        );
+        const text = result.content.map((c) => c.text).join("\n");
+        return result.isError ? `TOOL ERROR:
+${text}` : text;
+      }
+    });
+    const verified = loop.claimedDone && await verify(page, task);
+    return finish(run, startedMs, loop, verified, loop.claimedDone);
+  } catch (error) {
+    run.wallClockMs = Math.round(performance.now() - startedMs);
+    run.failure = error instanceof Error ? error.message : String(error);
+    return run;
+  } finally {
+    await browser.close();
+  }
+}
+var DOM_TOOLS = [
+  {
+    name: "read_page",
+    description: "Read the page's current accessibility tree (roles and names). Call after every action that changes the page.",
+    parameters: { type: "object", properties: {} }
+  },
+  {
+    name: "click",
+    description: "Click the element with this ARIA role and accessible name.",
+    parameters: {
+      type: "object",
+      properties: {
+        role: { type: "string", description: "ARIA role, e.g. button, link, textbox" },
+        name: { type: "string", description: "Accessible name, may be partial" }
+      },
+      required: ["role", "name"]
+    }
+  },
+  {
+    name: "fill",
+    description: "Type a value into the input with this ARIA role and accessible name (replaces content).",
+    parameters: {
+      type: "object",
+      properties: {
+        role: { type: "string" },
+        name: { type: "string" },
+        value: { type: "string" }
+      },
+      required: ["role", "name", "value"]
+    }
+  }
+];
+var MAX_SNAPSHOT_CHARS = 24e3;
+async function runDomLane(url, task, llm) {
+  const browser = await chromium.launch();
+  const run = metrics("dom", task);
+  const startedMs = performance.now();
+  try {
+    const context = await browser.newContext();
+    const page = await context.newPage();
+    await page.goto(url, { waitUntil: "load" });
+    const target = (role, name) => page.getByRole(String(role), {
+      name: String(name),
+      exact: false
+    }).first();
+    const loop = await runAgentLoop({
+      llm,
+      systemPrompt: "You complete tasks on a web page the way a browser-driving agent does: read the accessibility tree with read_page, then click and fill elements by role and name. Re-read the page after actions that change it. When the task is fully done, call task_complete.",
+      taskPrompt: task.prompt,
+      tools: DOM_TOOLS,
+      maxTurns: task.maxTurns,
+      invoke: async (name, args) => {
+        switch (name) {
+          case "read_page": {
+            const snapshot = await page.locator("body").ariaSnapshot();
+            return snapshot.length > MAX_SNAPSHOT_CHARS ? snapshot.slice(0, MAX_SNAPSHOT_CHARS) + "\n\u2026 (truncated)" : snapshot;
+          }
+          case "click": {
+            await target(args.role, args.name).click({ timeout: 5e3 });
+            await page.waitForTimeout(250);
+            return `Clicked ${String(args.role)} "${String(args.name)}".`;
+          }
+          case "fill": {
+            await target(args.role, args.name).fill(String(args.value ?? ""), { timeout: 5e3 });
+            return `Filled ${String(args.role)} "${String(args.name)}".`;
+          }
+          default:
+            return `ACTION FAILED: unknown action "${name}"`;
+        }
+      }
+    });
+    const verified = loop.claimedDone && await verify(page, task);
+    return finish(run, startedMs, loop, verified, loop.claimedDone);
+  } catch (error) {
+    run.wallClockMs = Math.round(performance.now() - startedMs);
+    run.failure = error instanceof Error ? error.message : String(error);
+    return run;
+  } finally {
+    await browser.close();
+  }
+}
+
+// src/report.ts
+import { mkdirSync, writeFileSync } from "fs";
+import { join } from "path";
+function median(values) {
+  if (values.length === 0) return 0;
+  const sorted = [...values].sort((a, b) => a - b);
+  const mid = Math.floor(sorted.length / 2);
+  return sorted.length % 2 === 1 ? sorted[mid] : Math.round((sorted[mid - 1] + sorted[mid]) / 2);
+}
+function aggregate(lane, runs) {
+  return {
+    lane,
+    runs,
+    successRate: runs.length === 0 ? 0 : runs.filter((r) => r.success).length / runs.length,
+    medianWallClockMs: median(runs.map((r) => r.wallClockMs)),
+    medianTotalTokens: median(runs.map((r) => r.totalTokens)),
+    medianTurns: median(runs.map((r) => r.turns))
+  };
+}
+function seconds(ms) {
+  return (ms / 1e3).toFixed(1) + "s";
+}
+function toMarkdown(report) {
+  const lines = [
+    `# AgentPerf report \u2014 ${report.task.id}`,
+    "",
+    `- **URL:** ${report.url}`,
+    `- **Model:** ${report.model}`,
+    `- **Task:** ${report.task.prompt}`,
+    `- **Started:** ${report.startedAt}`,
+    "",
+    "| lane | success | median wall-clock | median tokens | median round-trips |",
+    "|------|---------|-------------------|---------------|--------------------|"
+  ];
+  for (const lane of report.lanes) {
+    const success = `${Math.round(lane.successRate * 100)}% (${lane.runs.filter((r) => r.success).length}/${lane.runs.length})`;
+    lines.push(
+      `| ${lane.lane} | ${success} | ${seconds(lane.medianWallClockMs)} | ${lane.medianTotalTokens.toLocaleString()} | ${lane.medianTurns} |`
+    );
+  }
+  const tools = report.lanes.find((l) => l.lane === "tools");
+  const dom = report.lanes.find((l) => l.lane === "dom");
+  if (tools && dom && tools.medianTotalTokens > 0 && tools.medianWallClockMs > 0) {
+    lines.push(
+      "",
+      `**DOM lane pays ${(dom.medianTotalTokens / tools.medianTotalTokens).toFixed(1)}x the tokens and ${(dom.medianWallClockMs / tools.medianWallClockMs).toFixed(1)}x the wall-clock of the tools lane** (medians; failed runs included in success rate, excluded from nothing).`
+    );
+  }
+  lines.push("", "## Runs", "");
+  lines.push("| lane | # | success | wall-clock | tokens | round-trips | actions | failure |");
+  lines.push("|------|---|---------|------------|--------|-------------|---------|---------|");
+  for (const lane of report.lanes) {
+    lane.runs.forEach((r, i) => {
+      lines.push(
+        `| ${r.lane} | ${i + 1} | ${r.success ? "\u2713" : "\u2717"} | ${seconds(r.wallClockMs)} | ${r.totalTokens.toLocaleString()} | ${r.turns} | ${r.actions} | ${r.failure ?? ""} |`
+      );
+    });
+  }
+  lines.push("");
+  return lines.join("\n");
+}
+function writeReport(report, outDir) {
+  const stamp = report.startedAt.replace(/[:.]/g, "-");
+  const dir = join(outDir, stamp);
+  mkdirSync(dir, { recursive: true });
+  const json = join(dir, "report.json");
+  const md = join(dir, "report.md");
+  writeFileSync(json, JSON.stringify(report, null, 2));
+  writeFileSync(md, toMarkdown(report));
+  return { json, md };
+}
+
+export {
+  runAgentLoop,
+  HOST_SHIM,
+  runToolsLane,
+  runDomLane,
+  aggregate,
+  toMarkdown,
+  writeReport
+};

package/dist/cli.d.ts

@@ -0,0 +1 @@
+#!/usr/bin/env node

package/dist/cli.js

@@ -0,0 +1,75 @@
+#!/usr/bin/env node
+import {
+  aggregate,
+  runDomLane,
+  runToolsLane,
+  toMarkdown,
+  writeReport
+} from "./chunk-BGU3NILX.js";
+
+// src/cli.ts
+import { readFileSync } from "fs";
+var BUILTIN_TASKS = {
+  booking: {
+    id: "booking",
+    prompt: "You are on the booking page of the Fringe & Co. salon. Book the 'Beard trim' service for Dana Smith (dana@example.com): pick any open slot on the earliest available day, complete the booking, and report the confirmation code the page shows.",
+    successPattern: "FR-[A-Z0-9]{5}",
+    maxTurns: 20
+  }
+};
+function arg(name, fallback) {
+  const index = process.argv.indexOf(`--${name}`);
+  if (index !== -1 && process.argv[index + 1]) return process.argv[index + 1];
+  return fallback;
+}
+async function main() {
+  if (process.argv[2] !== "run") {
+    console.log("Usage: agentperf run --url <url> [--lane both|tools|dom] [--runs 3] [--model <model>]");
+    process.exit(process.argv[2] ? 1 : 0);
+  }
+  const url = arg("url");
+  if (!url) throw new Error("--url is required");
+  const apiKey = process.env.AGENTPERF_API_KEY ?? process.env.OPENAI_API_KEY;
+  if (!apiKey) throw new Error("Set OPENAI_API_KEY (or AGENTPERF_API_KEY)");
+  const llm = {
+    apiKey,
+    baseUrl: arg("base-url", process.env.OPENAI_BASE_URL ?? "https://api.openai.com/v1"),
+    model: arg("model", process.env.AGENTPERF_MODEL ?? "gpt-5.6-luna")
+  };
+  const taskArg = arg("task", "booking");
+  const task = BUILTIN_TASKS[taskArg] ?? JSON.parse(readFileSync(taskArg, "utf8"));
+  task.maxTurns = Number(arg("max-turns", String(task.maxTurns)));
+  const laneArg = arg("lane", "both");
+  const lanes = laneArg === "both" ? ["dom", "tools"] : [laneArg];
+  const runs = Number(arg("runs", "3"));
+  const outDir = arg("out", "results");
+  console.log(`AgentPerf \u2014 ${task.id} @ ${url}`);
+  console.log(`model ${llm.model}, ${runs} run(s) per lane: ${lanes.join(", ")}
+`);
+  const report = {
+    url,
+    model: llm.model,
+    task,
+    startedAt: (/* @__PURE__ */ new Date()).toISOString(),
+    lanes: []
+  };
+  for (const lane of lanes) {
+    const results = [];
+    for (let i = 1; i <= runs; i++) {
+      process.stdout.write(`[${lane}] run ${i}/${runs}\u2026 `);
+      const result = lane === "tools" ? await runToolsLane(url, task, llm) : await runDomLane(url, task, llm);
+      results.push(result);
+      console.log(
+        result.success ? `\u2713 ${(result.wallClockMs / 1e3).toFixed(1)}s, ${result.totalTokens} tokens, ${result.turns} turns` : `\u2717 ${result.failure}`
+      );
+    }
+    report.lanes.push(aggregate(lane, results));
+  }
+  const files = writeReport(report, outDir);
+  console.log("\n" + toMarkdown(report));
+  console.log(`Report: ${files.md}`);
+}
+main().catch((error) => {
+  console.error(error instanceof Error ? error.message : error);
+  process.exit(1);
+});

package/dist/index.d.ts

@@ -0,0 +1,121 @@
+/**
+ * Minimal OpenAI-compatible chat client with tool calling. No SDK — the
+ * harness talks to any /chat/completions endpoint (OpenAI, or a compatible
+ * provider via --base-url) and accounts every token it spends.
+ */
+interface LlmConfig {
+    apiKey: string;
+    baseUrl: string;
+    model: string;
+}
+interface LlmToolDef {
+    name: string;
+    description: string;
+    parameters: Record<string, unknown>;
+}
+
+type Lane = "tools" | "dom";
+interface TaskSpec {
+    id: string;
+    /** The instruction handed to the agent, identical across lanes. */
+    prompt: string;
+    /**
+     * Regex (source string) that must match the page's visible text when the
+     * task is genuinely done — checked by the harness, never by the model.
+     */
+    successPattern: string;
+    /** Hard cap on model round-trips before the run is failed. */
+    maxTurns: number;
+}
+interface RunMetrics {
+    lane: Lane;
+    taskId: string;
+    success: boolean;
+    /** Why a run failed, when it did. */
+    failure?: string;
+    wallClockMs: number;
+    /** Model round-trips (chat completions). */
+    turns: number;
+    /** Tool/browser actions executed. */
+    actions: number;
+    promptTokens: number;
+    completionTokens: number;
+    totalTokens: number;
+}
+interface LaneAggregate {
+    lane: Lane;
+    runs: RunMetrics[];
+    successRate: number;
+    medianWallClockMs: number;
+    medianTotalTokens: number;
+    medianTurns: number;
+}
+interface BenchmarkReport {
+    url: string;
+    model: string;
+    task: TaskSpec;
+    startedAt: string;
+    lanes: LaneAggregate[];
+}
+
+declare global {
+    interface Window {
+        __agentperf: {
+            listTools: () => Array<{
+                name: string;
+                description: string;
+                inputSchema: Record<string, unknown>;
+            }>;
+            call: (name: string, args: Record<string, unknown>) => Promise<{
+                content: Array<{
+                    type: string;
+                    text: string;
+                }>;
+                isError?: boolean;
+            }>;
+        };
+    }
+}
+declare function runToolsLane(url: string, task: TaskSpec, llm: LlmConfig): Promise<RunMetrics>;
+declare function runDomLane(url: string, task: TaskSpec, llm: LlmConfig): Promise<RunMetrics>;
+
+/**
+ * The agent loop both lanes share, so nothing about the driver differs
+ * between them except the tools it holds. task_complete is a harness tool —
+ * the model claims completion, the harness verifies it against the page.
+ */
+
+interface LoopOutcome {
+    turns: number;
+    actions: number;
+    promptTokens: number;
+    completionTokens: number;
+    claimedDone: boolean;
+    claimSummary: string;
+    failure?: string;
+}
+declare function runAgentLoop(options: {
+    llm: LlmConfig;
+    systemPrompt: string;
+    taskPrompt: string;
+    tools: LlmToolDef[];
+    invoke: (name: string, args: Record<string, unknown>) => Promise<string>;
+    maxTurns: number;
+}): Promise<LoopOutcome>;
+
+declare function aggregate(lane: Lane, runs: RunMetrics[]): LaneAggregate;
+declare function toMarkdown(report: BenchmarkReport): string;
+declare function writeReport(report: BenchmarkReport, outDir: string): {
+    json: string;
+    md: string;
+};
+
+/**
+ * Injected before any page script runs. Implements enough of the WebMCP
+ * standard (`document.modelContext.registerTool`) that the harness can act
+ * as the agent host in stock Chromium — no browser flag, no origin trial.
+ * Registered tools are captured and called through `window.__agentperf`.
+ */
+declare const HOST_SHIM = "\n(() => {\n  const tools = new Map();\n  const api = {\n    listTools: () =>\n      Array.from(tools.values()).map((t) => ({\n        name: t.name,\n        description: t.description,\n        inputSchema: t.inputSchema || { type: \"object\", properties: {} },\n        annotations: t.annotations || {}\n      })),\n    call: async (name, args) => {\n      const tool = tools.get(name);\n      if (!tool) {\n        return {\n          content: [{ type: \"text\", text: \"No tool named \\\"\" + name + \"\\\" is registered on this page.\" }],\n          isError: true\n        };\n      }\n      const result = await tool.execute(args || {});\n      if (typeof result === \"string\") return { content: [{ type: \"text\", text: result }] };\n      return result;\n    }\n  };\n  Object.defineProperty(window, \"__agentperf\", { value: api });\n\n  const modelContext = {\n    registerTool: (tool, options) => {\n      tools.set(tool.name, tool);\n      if (options && options.signal) {\n        options.signal.addEventListener(\"abort\", () => { tools.delete(tool.name); });\n      }\n    },\n    getTools: async () => api.listTools()\n  };\n  Object.defineProperty(document, \"modelContext\", { value: modelContext, configurable: true });\n  Object.defineProperty(navigator, \"modelContext\", { value: modelContext, configurable: true });\n})();\n";
+
+export { type BenchmarkReport, HOST_SHIM, type Lane, type LaneAggregate, type RunMetrics, type TaskSpec, aggregate, runAgentLoop, runDomLane, runToolsLane, toMarkdown, writeReport };

package/dist/index.js

@@ -0,0 +1,18 @@
+import {
+  HOST_SHIM,
+  aggregate,
+  runAgentLoop,
+  runDomLane,
+  runToolsLane,
+  toMarkdown,
+  writeReport
+} from "./chunk-BGU3NILX.js";
+export {
+  HOST_SHIM,
+  aggregate,
+  runAgentLoop,
+  runDomLane,
+  runToolsLane,
+  toMarkdown,
+  writeReport
+};

package/package.json

@@ -0,0 +1,49 @@
+{
+  "name": "agentperf",
+  "version": "0.1.0",
+  "description": "Measure what AI agents pay to use a website: the same task via DOM driving vs. WebMCP tool calls — wall-clock, tokens, round-trips, success rate.",
+  "license": "Apache-2.0",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/N-45div/AgentPerf.git",
+    "directory": "packages/harness"
+  },
+  "keywords": [
+    "webmcp",
+    "benchmark",
+    "agents",
+    "mcp",
+    "playwright",
+    "agentperf"
+  ],
+  "type": "module",
+  "bin": {
+    "agentperf": "./dist/cli.js"
+  },
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js"
+    }
+  },
+  "files": [
+    "dist",
+    "README.md"
+  ],
+  "dependencies": {
+    "playwright": "^1.49.0"
+  },
+  "devDependencies": {
+    "@types/node": "^22.0.0",
+    "tsup": "^8.3.0",
+    "typescript": "^5.7.0",
+    "vitest": "^3.0.0"
+  },
+  "scripts": {
+    "build": "tsup src/cli.ts src/index.ts --format esm --dts --clean",
+    "test": "vitest run",
+    "typecheck": "tsc -p tsconfig.json"
+  }
+}