cindy AGENTS.md

Repository instructions for the Cindy desktop and mobile client, including how its rules are organized and which documents govern development, product behavior, design, setup, and security.

In plain words
What is it for?
Use them before working on the Cindy client, especially when setting up the environment or modifying desktop windows, rendering, IPC, navigation, credentials, or stored user data.
Why use it?
They help coding agents respect the repository boundary, locate the right guidance, and avoid changing the separate server repository without permission.

Instructions file for CodexOpenCode

Install

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.

agentmods
npx agentmods add instructions/makecindy/cindy/agents-md
Clone the repo
git clone --depth 1 https://github.com/makecindy/cindy

Made for: Codex, OpenCode.

Per session 4,178 This file is loaded in full into every session.
When invoked 4,178 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.04178 $0.04178
Opus 5 $0.02089 $0.02089
Sonnet 5 $0.00836 $0.00836
Haiku 4.5 $0.00418 $0.00418

Measured yesterday against content hash a83bd2a45e25, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

cindy 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 yesterday.

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.

AGENTS.md · 184 lines

How it starts

The opening of the file, as written. The whole thing — 184 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Cindy 客户端仓:Agent 工作入口

本文件是 Codex 与 Claude Code 共用的项目指令正本。CLAUDE.md 只保留 @AGENTS.md,不要在两处重复维护规则。

仓库边界

  • 本仓库只负责 desktop、mobile 及其共享 packages。
  • 服务端位于独立仓库;除非用户明确要求,不要跨仓修改服务端。
  • 开始工作前先检查工作区状态和相关源码,不覆盖、不回退用户已有改动。

规则组织

  • 开发与工程规则统一放在 docs/dev-rules/
  • 产品行为与体验规则统一放在 docs/product-rules/
  • UI 视觉、交互与内容设计规则统一放在 docs/design-rules/,权威视觉规范正文为 docs/design-rules/DESIGN.md(根目录 DESIGN.md 仅为跳转入口),目录索引为 docs/design-rules/cindy-design-system.md
  • AGENTS.md 只保留所有任务都适用的规则、风险入口和文档索引。
  • 目录或模块专属规则优先放到对应目录的嵌套 AGENTS.md;需要跨目录复用的 专题说明放在 docs/,并由本文件写明触发条件。

当前规则索引

  • 首次接触本仓、需要定位功能代码位置或判断新代码归属模块时,先读仓库地图 docs/dev-rules/repo-map.md
  • 首次安装、修复依赖或准备新 worktree 时,必须先读 docs/dev-rules/environment-setup.md
  • 启动、调试或验证 Desktop 时,必须先读 docs/dev-rules/desktop-development.md
  • 修改 Desktop Renderer、preload、BrowserWindow、WebView、IPC、CSP、导航或 Electron 特权能力前,必须先读 docs/dev-rules/electron-security-and-process-boundaries.md
  • 新增或修改 Desktop 独立窗口、辅助窗口或弹出型 BrowserWindow 前,必须遵守 docs/dev-rules/electron-security-and-process-boundaries.md 的「独立辅助窗口统一生命周期 基线」:复用既有控制器/基础设施,不得另造平行的预热、就绪握手、隐藏复用与崩溃恢复 状态机。
  • 修改凭证或授权信息处理、文件落盘位置、用户持久数据、临时文件或测试目录前,必须 先读 docs/dev-rules/credentials-and-local-storage.md
  • 新增或修改媒体生成、导入、缓存、附件、持久化、协议解析或回收逻辑前,必须先读 docs/dev-rules/media-storage-and-protocols.md
  • 修改 Desktop 数据库 schema、migration、companion script 或运行期数据库访问前,必须 先读 docs/dev-rules/database-and-migrations.md
  • 开发、调试或验证 Mobile 时,必须先读 docs/dev-rules/mobile-development.md
  • 修改 apps/mobile 的原生配置、原生依赖、config plugin 或原生模块(app.jsonapp.config.jseas.jsonapps/mobile/package.jsonplugins/modules/ 等会 进入 runtime fingerprint 的输入)前,必须先读 docs/dev-rules/mobile-development.md 的「冷更边界」:除非必要,不得提交会改变指纹的改动;会触发冷更的 PR 与技术框架 变动同级,必须由仓库指定的把关人针对冷更明确确认后才能合并——不看改动大小,也不看谁 提的,提交者身份不构成例外。
  • 新增或调整产品功能、判断能力应进入 Core / Skill / 插件、设计人机交互或多端体验 前,必须先读 docs/product-rules/core-product-principles.md
  • 新增或修改 /review、Reviewer 任务、成果快照、Finding 协议、复核入口、结果呈现或 复核生命周期前,必须先读 docs/product-rules/review-product-direction.md
  • 新增或修改按区域(cn / global)分支的逻辑、构建身份与命名、端点选择、区域相关 UI 标注,或涉及两个版本关系的对外文案前,必须先读 docs/product-rules/region-and-editions.md无限定词身份归 Global,未显式指定 区域一律落在 global,只标注中国大陆版
  • 新增或修改任何界面、组件、布局、样式、动效或 UI 文案前,必须先读权威设计规范 docs/design-rules/DESIGN.md;设计文档索引见 docs/design-rules/cindy-design-system.md
  • 新增或修改设计 Token、主题系统、标准 UI 组件(primitive / pattern)、视觉类门禁脚本, 或参与设计系统迁移 PR 前,必须先读治理合同 docs/design-rules/design-governance.md:真相源边界、兼容红线、两级证据合同与 PR 风险分类均以它为准;视觉规则本身仍以 DESIGN.md 为准。
  • 新增或修改任何 UI 文案里的产品术语前,必须先查术语表 i18n/GLOSSARY.md:已裁决 的术语照用,不自造译法;表里没有或拿不准的,在 i18n/glossary.jsonstatus: "proposed" 条目再讨论。门禁为 pnpm check:i18n-glossary,规则见 docs/dev-rules/engineering-conventions.md §5.1。
  • 新增或修改任一 Telegram bot 的用户可见行为(命令、消息呈现、收口策略、群行为、 权限口径、附件与表情)前,必须先读能力台账 docs/product-rules/telegram-bot-parity.md: 两个 bot 是两套架构,差异可以有但必须登记在表里;表里标「有意不同」的行不要去 "统一",动它要先推翻对应裁决。改完记得把对应行写回去。
  • 文案里出现任务 / 对话 / 消息这几个词时,必须先读 docs/product-rules/task-and-conversation-naming.mdsession 面向用户叫「任务」, 「对话」只用于任务内的交流过程与内容,单条往来叫「消息」;「任务」与 task 同句出现 时必须消解歧义。这三个词的边界拿不准会直接做出用户能看见的不一致。
  • 所有新增或修改的 UI 必须同时实现 Light 与 Dark 两种模式(颜色一律走语义 token, 禁止只适配一种模式的硬编码或条件补丁);只实现一种模式视为未完成。两种模式的实机 目检不是硬性门槛——能目检更好,做不到时如实写明哪种模式未验证,不得把「复用了 themed 样式」当成「双模式已验证」。具体要求以 docs/design-rules/DESIGN.md 的双模式交付门槛 为准。
  • 修改 Orca 多 Agent 协同时,必须先读 docs/dev-rules/orca-team-architecture.md
  • 修改 packages/maker-core 的 Agent 编排、prompt 组装、tool/MCP 暴露、translator、 model 映射、usage 计量,或任何进入模型 system 段的提示词前,必须先读 docs/dev-rules/maker-core-and-agent-behavior.md
  • 修改 PI harness 集成(packages/maker-core/src/agents/pi/**pi-host.tspiEnvironment.ts)、PI 会话权限/配置/system prompt/桥接,或 PI 相关的上线判断前, 必须先读 docs/dev-rules/pi-harness.md(含设计原则、维护不变量与上线清单)。
  • 修改插件(.cindy)运行时、沙箱、权限、能力 slot、面板供片、网络/凭证/文件交接, 或身份卡、管子协议、打包与编写手册前,必须先读 docs/dev-rules/plugin-security-and-authoring.md。其中存量插件兼容是红线:任何 插件系统改动(含批准状态 schema、指纹格式、manifest 校验、安装布局、包格式)都必须 向下兼容——用户升级后什么都不做,已装、已批准、已启用的插件必须照旧可用,绝不允许 要求用户重新安装、重新确认权限或重新配置。做不到就必须自带从旧版数据的自动迁移; 自动迁移也做不到时,必须有明确提示 + 一次性批量恢复入口,且不丢用户已存的凭证与偏好。 漏迁移 = P0,规则正文见该文件第 5 节。改到插件基座(运行时/沙箱、批准状态记录、 能力 slot、打包与内容判据、manifest 契约、装入与权限确认 UI、已装列表投影)的 PR 一律 走白名单确认门,需放行人明确 Approve 才能合并,不看 diff 大小、不因「是 bugfix/纯技术 改动」豁免。
  • 修改插件发现链(花名册注入、ghost_list / ghost_info / ghost_call)、插件运行期 可见性门禁或 FORGE_GUIDE 作者契约前,必须先读 docs/ghost-progressive-discovery.md
  • 新增或修改插件持久 Library(library 槽、binding / 目录选择、随时迁移、 回收站删除、SQLite 语句门或 /library/ 面板投影)前,必须先读 docs/dev-rules/plugin-library-storage.md
  • 修改客户端自动更新链路(cindy-updater 或 Electron 侧更新服务)前,必须先读 docs/dev-rules/cindy-updater.md
  • 新增或修改 Desktop 日志、IPC 错误处理、main 侧业务逻辑与测试、跨平台(macOS/ Windows)行为,或任何 UI 文案的 i18n 落地前,必须先读 docs/dev-rules/engineering-conventions.md
  • 修改客户端日志采集/脱敏/上报链路(apps/desktop/src/main/log-upload/**)、logger.ts 的 main 日志行格式、崩溃判定或待补传标记前,必须先读 docs/dev-rules/log-upload-and-redaction.md。其中三条是不变量记录边界(写侧 续行转义与读侧切分是同一条不变量的两半,破坏任一侧即隐私逃逸)、白名单方向 (deny-by-default,不得改成黑名单——调试级别的功能日志是用户内容的主要泄漏源)、 标记代次 + 原子清除(并发实例下仅靠时间戳会误删另一实例刚写的新崩溃标记)。 脱敏规则只增不减,放宽任一条视为隐私变更、需重新评审。
  • 修改本地协议 package、插件分发来源边界或 device-link 协议/relay/隧道 payload/IPC allowlist,或任何改动跨端 wire protocol 前,必须先读 docs/dev-rules/protocol-compatibility.md
  • 修改 package 依赖方向、main 进程模块加载方式,或主界面布局树结构前,必须先读 docs/dev-rules/architecture-invariants.md
  • 新增或修改 Settings UI、配置文件、本地偏好、运行时 profile,或 agent/MCP/provider 开关前,必须先读 docs/dev-rules/configuration-and-overrides.md
  • 新增或修改涉及 workdir 文件、agent 进程、会话数据的功能,新增 IPC channel/推送事件, 或修改 device-link 的重试/超时/断链恢复逻辑前,必须先读 docs/dev-rules/remote-and-mobile-adaptation.md;其中恢复路径改动必须回答该文件的 「故障半径三问」。
  • 在 Cindy 内嵌 worktree 会话里工作、准备提交或直推、或做 code review 前,必须先读 docs/dev-rules/development-workflow.md

Read the full file on GitHub · 184 lines

Changes

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.

  1. yesterday First seen · 184 lines · 4,178 tokens per session scan A a83bd2a45e25

Subscribe to this mod's changes

cindy AGENTS.md is an instructions file published in the GitHub repository makecindy/cindy (2,360 stars, last pushed yesterday), licensed Apache-2.0. It adds 4,178 tokens to every session, about $0.0209 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.