Borrowing it
Nothing to install: this file belongs to yangpeng366/openeyes. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/yangpeng366/openeyes/main/AGENTS.mdgit clone --depth 1 https://github.com/yangpeng366/openeyesWrote 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/instructions/yangpeng366/openeyes/agents-md)<a href="https://agentmods.dev/instructions/yangpeng366/openeyes/agents-md"><img src="https://agentmods.dev/badge/instructions/yangpeng366/openeyes/agents-md/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/instructions/yangpeng366/openeyes/agents-md"><img src="https://agentmods.dev/badge/instructions/yangpeng366/openeyes/agents-md.svg" alt="Reviewed on agentmods" width="80" 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 | $0.02675 | $0.02675 |
| Opus 5.5 | $0.01070 | $0.01070 |
| Sonnet 5.5 | $0.00535 | $0.00535 |
| Haiku 4.5 | $0.00267 | $0.00267 |
Grade A, and why
openeyes AGENTS.md 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 today.
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 — 126 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — OpenEyes 交接要点
面向接手的 agent / 人类。目标:读完这页就能动手改,不必修完再回头翻长文。
这项目是干什么的
AI 友好的 Windows 桌面操作原语:see → resolve → act。
任何大模型 agent 通过 CLI(eyes)或 MCP server 驱动桌面 GUI。
核心资产是结构化 element schema 与降级链,不是模型。
六条主要思路
- 结构化优先:
uia → cdp → vision。能走无障碍树 / DOM 就不看像素; 截图是最后手段,且走解析式 vision(OmniParser v2:先解析成结构化元素再决策)。 每次结果都带source,说明这答案从哪来。 - 易点击:UIA 命中点中心 → 未命中点 vision bbox 中心 → 都不命中走字母 hint overlay。 任何一步都要有可解释的降级原因。
- 默认 dry-run:所有改变状态的操作先 dry-run,显式
--go才真执行。 验收门禁据此判定「真跑了」还是「只是演练」。 - MCP 优先:23 个工具的契约是一份稳定接口,改内部不许改契约;
examples/mcp-stdio-probe.py就是为了守住它。 - Record & Replay 无需 vision:录的是逻辑步骤(UIAEvent)而非原始事件流,
回放翻译成
openeyes.actuate()调用;backend 可注入(auto/jev真模型 /mock), 真模型缺凭据时回落 mock,不静默卡住。 - 判断交给模型,阈值留在代码:
openeyes/decision/的decide()把打分分四带 (ACT / FALLBACK / UNCERTAIN / FLAG_FOR_REVIEW),只有 FLAG_FOR_REVIEW 这一带 会置needs_attention=True——人的注意力只花在真需要人看的地方。 - 验收要能判红:浏览器四路探针(type / shot / 双跑 / 失败面)统一一套 verdict 模型, 双跑比对要真哈希;判据必须「有可能失败」,否则门禁形同虚设。
改动规矩
- 跑
pwsh -NoProfile -File examples\quality-gate.ps1:ruff、UTF-8 无 BOM、MCP 契约 23 工具、pytest 全套。 CI 跑的是同一串步骤,别只跑局部。 - 全量测试目前 1336 passed / 8 skipped(2026-11-18 实测);改到公共路径就把全量跑一遍。
- 真机验收用可见窗口(真实 Edge profile),不用无头。
- 源文件 UTF-8 无 BOM(
examples/bom-sweep.py会扫)。 - 提交直接推
origin main(单人维护,不走 PR、不 force push)。
接口速查(agent 接入只看这一层)
- CLI:
eyes <command> --json,字段固定;子命令先 dry-run 再--go。 - MCP:
openeyes-mcp,工具名与 CLI 一一对应。 - 退出码:0 成功 / 1 失败 / 2 参数错 / 3 没匹配到目标 / 4 后端不可达 / 5 动作没生效。 JSON 走 stdout,人读错误走 stderr。
- 元素字段:
bbox / center / control_type / name / automation_id / class / state。 - 要更深:README 的 Architecture 与 Record & Replay 两节。
已知边界
- 只有 Windows 的 UIA 后端真正可用;macOS / Linux 在路线图上。
- 远程浏览器走 SSH 端口转发 + Host 头改写,隧道断了就是断了,不自动修。
- vision backend 依赖外部模型权重,离线环境只到 uia / cdp 这层。
实践与调试
每条都是实测过的坑:症状 → 原因 → 对策。命令照仓库实情写,先说「跑什么、看什么」。
1. 中断与回执(四路验收探针):Ctrl+Break → exit 130 + verdict=INTERRUPTED,且 --out
目录必须为空(不写回执);taskkill 走 Windows STATUS_CONTROL_C_EXIT(0xC000013A);
TerminateProcess 硬杀 → exit 1 且无回执——只存在 PNG 不算通过,回执是唯一凭证。
对策:探针一律用空的 --out 起步,跑完先看回执再看产物。
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.
- today Changed · +4 lines · +139 tokens per session fc8deaed55ba
- 3d ago Changed · +15 lines · +473 tokens per session 3996186b9b05
- 4d ago First seen · 107 lines · 2,063 tokens per session scan A b9e4a798a11f
openeyes AGENTS.md is an instructions file published in the GitHub repository yangpeng366/openeyes (2 stars, last pushed today), licensed MIT. It adds 2,675 tokens to every session, about $0.0107 per session on Opus 5.5. 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 2026-11-13.
openeyes AGENTS.md compared with instructions from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| openeyes AGENTS.mdyangpeng366/openeyes | Instructions file | 2 | today | MIT | A | 2,675 | Codex, OpenCode |
| vscode buildNext.instructions.mdmicrosoft/vscode | Instructions file | 196k | today | MIT | A | 9,719 | GitHub Copilot |
| next.js AGENTS.mdvercel/next.js | Instructions file | 144k | today | MIT | A | 8,130 | Codex, OpenCode |
| vscode oss-third-party-notices.instructions.mdmicrosoft/vscode | Instructions file | 196k | today | MIT | A | 5,685 | GitHub Copilot |
| langchain AGENTS.mdlangchain-ai/langchain | Instructions file | 150k | today | MIT | A | 4,680 | Codex, OpenCode |
| vscode copilot-instructions.mdmicrosoft/vscode | Instructions file | 196k | today | MIT | A | 3,153 | GitHub Copilot |
| generative-ai-for-beginners AGENTS.mdmicrosoft/generative-ai-for-beginners | Instructions file | 124k | today | MIT | A | 2,981 | Codex, OpenCode |