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 open-octo/octo-agent --skill web-accessgit clone --depth 1 https://github.com/open-octo/octo-agentWrote 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/open-octo/octo-agent/web-access)<a href="https://agentmods.dev/skills/open-octo/octo-agent/web-access"><img src="https://agentmods.dev/badge/skills/open-octo/octo-agent/web-access/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/open-octo/octo-agent/web-access"><img src="https://agentmods.dev/badge/skills/open-octo/octo-agent/web-access.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00139 | $0.02562 |
| Opus 5 | $0.00069 | $0.01281 |
| Sonnet 5 | $0.00028 | $0.00512 |
| Haiku 4.5 | $0.00014 | $0.00256 |
Grade A, and why
web-access 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 9d 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 — 150 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Skill: web-access
复杂 web 任务的方法论：把内置的 web_search / web_fetch 与 browser 工具（驱动你日常的、已登录的 Chrome/Edge）按场景编排起来，并跨 session 积累站点经验。
前置：浏览器自动化
web_search / web_fetch 开箱即用，不需要任何配置。
只有当任务需要操作浏览器界面 / 登录态 / 动态渲染页面时，才需要 browser 工具能连上你日常的浏览器。一次性配置：
- 在要用的浏览器打开 inspect 页面并勾选 Allow remote debugging for this browser instance（可能需重启浏览器）：
- Chrome：
chrome://inspect/#remote-debugging - Edge：
edge://inspect，然后点左侧 Remote debugging
- Chrome：
- 或直接运行
octo browser setup，它会给出上面的步骤、验证连接、并把端口写进配置。
连上后 browser 工具天然携带登录态——大多数常用网站都已登录。无独立浏览器、无需命令行参数。
部分站点对浏览器自动化检测严格，存在账号限流/封禁风险。操作社交平台（小红书等）强烈建议用小号。Agent 继续操作即视为接受。
浏览哲学
像人一样思考，兼顾高效与适应性地完成任务。
带着目标进入，边看边判断，遇到阻碍就解决，发现内容不够就深入——全程围绕「我要达成什么」做决策。
① 拿到请求 — 先明确成功标准：什么算完成？需要获取什么信息、执行什么操作、达到什么结果？这是后续所有判断的锚点。
② 选择起点 — 根据任务性质、平台特征、达成条件，选一个最可能直达的方式作为第一步去验证。需要操作页面、需要登录态、已知静态方式不可达的平台（小红书、微信公众号等）→ 直接用 browser。
③ 过程校验 — 每一步的结果都是证据。用结果对照①的成功标准：路径在推进吗？结果的质量、相关度、量级是否指向目标可达？发现方向错了立即调整，不在同一方式上反复重试——搜索没命中不等于"还没找对方法"，也可能是"目标不存在"。遇到弹窗、登录墙，先判断它是否真的挡住了目标：内容可能已在 DOM 中，交互只是展示手段。
④ 完成判断 — 对照成功标准确认完成才停止；但也不为了"完整"过度操作、浪费代价。
联网工具选择
确保信息真实性，一手信息优于二手。搜索引擎和聚合平台是发现入口，不是真伪的证明。
| 场景 | 工具 |
|---|---|
| 搜索摘要、关键词结果、发现信息来源 | web_search |
| URL 已知，按 prompt 从页面提取信息 | web_fetch（直接传原始 URL；HTML 自动转成干净 Markdown，正文优先、链接绝对化） |
| URL 已知，需要页面原始 HTML（meta、JSON-LD 等结构化字段） | web_fetch 加 clean=false |
| 非公开内容，或已知静态层无效的平台（小红书、公众号等） | browser（直接，跳过静态层） |
| 需要登录态、交互操作，或要像人一样在浏览器内自由导航探索 | browser |
web_search / web_fetch 都不处理登录态。browser 不要求 URL 已知——可从任意入口出发，靠页面内搜索、点击、跳转找到目标。
browser 工具要点
action 级用法（observe / click / eval / record / replay 等的参数与语义）以 browser 工具自身的 schema 描述为准，此处不重复。schema 之外的判断准则：
- 重复性流程（批量操作、定期取数）优先录制回放（record → replay），而非每次盲驱动。
- 收尾用
close关闭自己开的标签页，保留用户原有标签页。
程序化 vs GUI 交互
- 程序化（navigate 构造 URL、eval 操作 DOM）：快、精确，但对网站不是正常用户行为，可能触发反爬。
- GUI 交互（observe→click→type→scroll）：网站不限制正常 UI 操作，确定性最高，但步骤多、慢。
根据对平台的了解灵活选择。GUI 交互也是有效探测：一次真实交互能观察站点实际行为（URL 模式、必需参数、跳转逻辑），为后续程序化操作提供依据；程序化受阻时它是可靠兜底。
What ships with it
2 files 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.
- 9d ago First seen · 150 lines · 139 tokens per session scan A 99b6cc5aadf5
web-access is a skill published in the GitHub repository open-octo/octo-agent (97 stars, last pushed today), licensed MIT. It adds 139 tokens to every session and 2,562 once invoked, about $0.0007 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 2026-08-30.
Other skills, from other repositories
ha-browser
Hope Agent browser automation — the standard status → tabs → snapshot → act loop, stale-ref recovery rules, and what to do when login / 2FA / captcha / camera-prompt / dialog blocks progress. Load this skill whenever you reach for the browser tool. Trigger on: user asks the agent to open / control / click / scrape /…
claude-in-chrome
Automates your Chrome browser to interact with web pages - clicking elements, filling forms, capturing screenshots, reading console logs, and navigating sites. Opens pages in new tabs within your existing Chrome session. Requires site-level permissions before executing (configured in the extension).
hive.browser-automation
Required before any hive-browser CLI command. The browser is driven from the terminal by running hive-browser ... --json via terminalexec — not via MCP tools. Teaches the browser lifecycle rules (the bridge attaches to the USER'S running Chrome — never kill or launch browser processes; timeouts are transport issues…
hive.linkedin-automation
Read before automating LinkedIn with browser tools. LinkedIn combines shadow DOM (#interop-outlet), strict Trusted Types CSP that silently drops innerHTML, Lexical composer, native beforeunload dialogs that hang the bridge, and aggressive spam filters — each has bitten us at least once. Verified flows for profile…
hive.x-automation
Read before automating X / Twitter with browser tools. Verified flows for post, reply, delete, search-and-engage, plus the Draft.js compose quirks that silently disable the send button. Includes the daily-reply and job-market-reply playbooks. Requires hive.browser-automation for the underlying screenshot + coordinate…
hive.slack-notifications-setup
Set up a Slack notification channel (Sentinel) for a colony by driving the browser — reuse or create the "Hive Sentinel" Slack app from a JSON manifest, install it, capture the bot + app tokens, create/select the channel via the Slack API, and turn Sentinel on so the colony can ping the user on Slack and accept…