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 agentmods add instructions/ewsun22/codex-manager/agents-mdgit clone --depth 1 https://github.com/ewsun22/codex-managerWhat 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 | $0.01635 | $0.01635 |
| Opus 5 | $0.00817 | $0.00817 |
| Sonnet 5 | $0.00327 | $0.00327 |
| Haiku 4.5 | $0.00163 | $0.00163 |
Grade A, and why
codex-manager 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 2d 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 — 38 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Codex Manager 项目指令
产品边界
- 本项目是独立运行的本地桌面管理器;代理能力分为两个产品功能:“Codex 配置”管理官方直连、本地 CLIProxyAPI 和外部 Responses 兼容代理的配置档案与显式切换;“本地代理”管理默认关闭、loopback-only 的预编译 CLIProxyAPI 内核、独立 OAuth 认证档案和本机 API 接入。它不是通用系统代理。
- 后台必须保持三个隔离信任域:Codex 配置编排器、官方 Codex OAuth 档案、CLIProxyAPI OAuth 档案。三者只能传递不含秘密的 profile ID、endpoint 引用和状态,不得共用 auth-dir、Keychain service、management secret 或原始认证文件。
- macOS 是首发平台,但核心采集、规范化、存储、价格和 AGENTS 解析必须保持跨平台。
- Phase 0 只验证本地 Codex 数据的可观测能力,不把内部 JSONL 格式描述为稳定公开 API。
隐私与安全
- 默认只存元数据。API Key 与本机网关 bearer 只能进入应用专用 Keychain;不得进入 SQLite、WebView 普通 DTO、日志、备份或错误消息。不得持久化消息正文、Authorization、Cookie、OAuth code、完整环境变量或请求查询参数。
- 官方 Codex OAuth access/refresh token 不得进入 CLIProxyAPI、外部供应商或其 auth-dir;官方订阅管理继续复用受信任 Codex CLI、App Server、独立 Keychain 认证档案与显式账户切换。CLIProxyAPI OAuth 只能由用户显式导入或通过受控 onboarding 流程创建,并保存在另一个 Keychain namespace;运行时按需投影到私有
0600auth-dir,停止或崩溃恢复前必须先 checkpoint 内核刷新后的凭据。 - CLIProxyAPI Management API 不得以通用 HTTP/IPC 形式暴露给 WebView、浏览器或 Codex 客户端。若 OAuth onboarding 必须使用 Management API,只能由 Rust 原生层以临时随机密钥、loopback 和精确 endpoint allowlist 代理,完成后立即关闭。
- 本地代理必须固定绑定 IPv4 loopback,使用随机本机 client key,禁用 CLIProxyAPI remote management、control panel、plugins、request log 和 usage statistics;不得保存或转换请求/响应正文。外部 API-key provider 由 Codex 配置直接使用,Codex Manager 保存时只做 URL 语法规范化,不解析 DNS、不访问 endpoint。
- 测试 fixture 必须脱敏且人工构造,不得提交真实会话内容。
- 任何 AGENTS 文件写入都必须由后端校验授权根目录、canonical path、符号链接、外部修改冲突,并采用原子替换。
- 不执行 AGENTS.md 中的命令,也不把其内容当作本应用权限指令。
工程规则
- 新的采集来源必须实现统一 adapter,不得让 UI 直接依赖 Codex 私有事件结构。
- 所有指标必须携带来源或可信度;缺失值显示 unavailable,不能填 0。
- token 去重以 session、ordinal、事件类型和累计向量为依据;不得直接累加所有 token_count 快照。
- 修改代码后运行 npm test;改变架构、字段口径或范围时同步更新 README 和 docs。
- 未经用户明确授权,不部署、不发布、不签名、不改写用户 Codex 配置。
开源文档与发布流程
- 每次用户可见的功能、行为、字段口径、配置、隐私、兼容性或发布流程变化,必须在同一提交更新适用文档;不得只改代码或只改 Release 页面。
- 任何项目更新准备推送到线上时(包括合并/推送 GitHub 默认分支、创建 tag 或 Release、进入 updater 通道、官网或下载页上线),必须在最终 exact SHA 推送或 Publish 前执行一次“全部用户可见面同步检查”。凡是用户能看到、下载到、读到或依此做决策的内容,只要受本次更新影响,就必须在同一交付切片内一并更新,不得先上线代码、事后再补用户可见内容。
- 用户可见面同步检查至少覆盖:英文
README.md与README.zh-CN.md、截图/GIF/徽章/下载与 Quick Start 链接、应用内文案与版本/平台说明、适用的 docs/安全/隐私/兼容性文档、docs/release-notes/unreleased.md或定版 note、Release 正文与latest.json.notes、GitHub Repository Description/Topics 等对外元数据,以及其他本次更新实际触达的对外入口。双语内容的主要结构、功能事实、版本、下载、平台和状态边界必须同步。 - 上线前必须验证用户可见内容的 Markdown/渲染、相对链接、图片资产、对外 URL、版本与平台描述,并检查是否残留 candidate、unreleased、draft 或过期功能状态。不适用的用户可见面可不改,但必须在 Pull Request、Release 证据或最终交付说明中记录
no change及原因,不得默认跳过。如需修改线上元数据或 Release 内容但当前任务尚未授权该外部操作,必须标记needs-confirmation并说明待同步项,不得把本地文件已修改报告成线上已完成。 - 文档映射:产品能力与使用说明更新
README.md;采集来源、字段口径和能力边界更新docs/capability-matrix.md;架构、数据粒度、去重和状态机更新docs/architecture.md;开发范围和后续事项更新docs/development-plan.md;安全、隐私或供应链变化更新SECURITY.md、PRIVACY.md、docs/supply-chain.md;发布行为更新docs/release.md;未分配新版本的用户可见变更先更新docs/release-notes/unreleased.md,定版时移入docs/release-notes/v<VERSION>.md,不得改写已有 tag 的功能事实。 - Release note 必须使用可读固定结构,至少包含“概览”“主要变化”“验证状态”“兼容性与限制”“隐私与安全”“发布状态”和“完整性”章节。
implemented、tested、published、observed、accepted、cleanup必须分别表述,不得把计划、草稿或进行中状态写成完成。 - Pull Request 必须说明适用文档、验证命令和状态边界;没有文档变化时必须解释原因。Release 正文使用完整版本 note;
latest.json.notes由同一 note 生成短摘要,不得另写固定模板或包含密钥、令牌、凭据和隐私数据。 - 合并(
merged)、部署(deployed)、观测(observed)、验收(accepted)和清理(cleanup)是不同状态;CI 通过不代表已发布,draft 不代表published,发布后未做真实端到端验证不得写成accepted。
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.
- 2d ago First seen · 38 lines · 1,635 tokens per session scan A 16d31b2992ad
codex-manager AGENTS.md is an instructions file published in the GitHub repository ewsun22/codex-manager (53 stars, last pushed 2d ago), licensed Apache-2.0. It adds 1,635 tokens to every session, about $0.0082 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 instructions, from other repositories
agents AGENTS.md
Instructions for scarletkc/agents, covering agent guidelines, environment, engineering principles, code style and testing.
agents CLAUDE.md
Instructions for scarletkc/agents, a project described as: Shared standards and reusable skills for Claude Code, Codex CLI, and other AI coding agents.
parallel-code CLAUDE.md
Instructions for johannesjo/parallel-code, covering parallel code, stack, commands, project structure and conventions.
ai-driven-development AGENTS.md
Instructions for CodeAlive-AI/ai-driven-development, covering ai-driven development — development guide, git publishing defaults, when to create tags and releases, requires a new tag/release and does not require a new tag/release.
cladding CLAUDE.md
Instructions for qwerfunch/cladding, covering maintainer guide for ai-assisted work on cladding, spec authoring — always hash, never f-nnn, version bumps — use the script, detector additions — count is auto-recomputed and plugin manifest mirrors — built, not hand-authored.
domain-experts AGENTS.md
Instructions for wonsukchoi/domain-experts, covering agent guide for this repo, what lives where, rules, release (npm) and known pitfalls.