Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add cc10143/annotation-overlay-mcp --skill annotation-browser-launchgit clone --depth 1 https://github.com/cc10143/annotation-overlay-mcpWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/cc10143/annotation-overlay-mcp/annotation-browser-launch)<a href="https://agentmods.dev/skills/cc10143/annotation-overlay-mcp/annotation-browser-launch"><img src="https://agentmods.dev/badge/skills/cc10143/annotation-overlay-mcp/annotation-browser-launch.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00086 | $0.01181 |
| Opus 5 | $0.00043 | $0.00590 |
| Sonnet 5 | $0.00017 | $0.00236 |
| Haiku 4.5 | $0.00009 | $0.00118 |
Grade A, and why
annotation-browser-launch scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 81 lines — stays where its author put it; the contents beside it link to each section on GitHub.
标注浏览器启动（Annotation Browser Launch）
何时用
agent 需要用户在页面上做视觉标注反馈，标注发生在 agent 自己拉起的浏览器窗口 里（方案 D，主流程）。配合 annotation-feedback-wait 使用：本 skill 拉窗口 + 激活标注 → wait skill 收数据 → 本 skill 重连复验。
为什么是 agent 的浏览器
标注界面放在 agent 控制的浏览器（chromium-956323）而不是用户真实 Chrome：agent 原生控制导航/刷新/截图，闭环最顺。已验证：该 chromium 构建能加载我们的扩展、https 页能 fetch localhost server（旧 chromium-1228 不行 —— 它 https→localhost 挂起，submit 会失败，必须用 956323）。
配置表（本机路径）
| 项 | 值 |
|---|---|
| chromium | C:/Users/26462/AppData/Local/ms-playwright/chromium-956323/chrome-win/chrome.exe |
| 扩展目录 | D:/KaiFa/annotation-overlay/extension |
| 持久 profile | C:/Users/26462/.annotation-overlay/browser-profile（登录态跨轮次保留） |
| CDP 端口 | 9223（避开 chrome-debug 的 9222） |
| playwright NODE_PATH | C:/Users/26462/AppData/Local/OpenAI/Codex/runtimes/cua_node/1b23c930bdf84ed6/bin/node_modules |
| 启动脚本 | ~/.claude/skills/annotation-browser-launch/scripts/launch.cjs |
脚本支持 env 覆盖：ANNO_BROWSER_CHROME / ANNO_BROWSER_EXT / ANNO_BROWSER_PROFILE / ANNO_BROWSER_PORT。
流程
1. 启动标注浏览器（后台运行）
先查 9223 端口是否已在监听 —— 如果上次的标注浏览器还开着，直接跳到「重连」复用，不要重复启动（同 profile + 端口互斥）。
NODE_PATH="<配置表 playwright NODE_PATH>" node ~/.claude/skills/annotation-browser-launch/scripts/launch.cjs "<目标URL>"
用 run_in_background 运行。脚本会：
- 拉起 chromium-956323 + 扩展 + 持久 profile + 9223 端口
- 导航到目标 URL，等 overlay 注入，
activate()显示工具栏 - 输出
[annotation-browser] READY url=... | overlay=true | toolbar=true | port=9223
首次使用：profile 是空的，用户需要在 agent 浏览器里登录目标 app 一次，之后跨轮次保留。
2. 请用户标注
窗口已在用户屏幕、工具栏已显示。告诉用户用标注工具圈画问题区域并点 Submit。然后转 annotation-feedback-wait 收数据。
3. 重连 + 复验
agent 处理反馈后，用 node_repl 连回浏览器操作（刷新/截图/重激活）：
const pw = await import("playwright");
const c = await pw.chromium.connectOverCDP("http://127.0.0.1:9223");
const page = c.contexts().flatMap(cx => cx.pages())[0];
await page.reload(); // 加载修复
await page.waitForLoadState("domcontentloaded");
await page.evaluate(() => window.__annotationOverlay.activate()); // 重激活标注（提交后会自动停用）
// 复验截图：await page.screenshot({ path: "C:/Users/26462/.annotation-overlay/screenshots/verify.png" })
await c.close();
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago First seen · 81 lines · 86 tokens per session scan A b5d73f1e9679
annotation-browser-launch is a skill published in the GitHub repository cc10143/annotation-overlay-mcp (0 stars, last pushed 5d ago), licensed MIT. It adds 86 tokens to every session and 1,181 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
peek
Use when the user mentions a recent browser session, an error they just reproduced, "what was the user doing before X", DOM state at some past moment, or wants to turn a manual repro into a Playwright test. Peek exposes 18 MCP tools backed by a local SQLite store of rrweb-captured browser sessions.
verify-bb
Verify BB user journeys in an isolated source dev app using dev-browser@next and the matching source CLI. Use after BB feature changes or when asked to verify or smoke-test BB. The feature map covers core UI and agent interfaces, every repository plugin, desktop, mobile, and hosted services; select the affected…
browser-automation
Use the Browser Automation BB plugin to inspect and automate persistent browser pages in an explicit desktop or local headless session. Use for browser navigation, snapshots, clicking, forms, and verification screenshots.
d3k
Use when the user asks to use d3k, run/dev/test/debug a web project with d3k, or reproduce a browser issue. Own the runtime: reuse or background-start d3k non-interactively, wait for readiness, use its project-stable managed Chrome profile, and inspect unified browser/server evidence.
create-verification-skill
Create a repo-local verification skill and exhaustive feature map for driving a real app through its UI, CLI, or API. Use for "create a verification skill", "make a verify skill for this repo", or "document how agents can verify this app".
maintain-verification-skill
Audit an existing verification skill and feature map against source and the running app, then repair proven documentation or harness drift. Use for "maintain the verification skill", "audit the verify skill", or "refresh the verification feature map".