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/xerrors/yuxi/agents-mdgit clone --depth 1 https://github.com/xerrors/YuxiWhat 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 | $0.01873 | $0.01873 |
| Opus 5 | $0.00937 | $0.00937 |
| Sonnet 5 | $0.00375 | $0.00375 |
| Haiku 4.5 | $0.00187 | $0.00187 |
Grade A, and why
Yuxi 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.
How it starts
The opening of the file, as written. The whole thing — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Yuxi Agent 开发约定
Yuxi 是基于 LangGraph、FastAPI、Vue 和多种持久化服务构建的知识库与多智能体平台。Docker Compose 是开发拓扑的事实来源;修改不熟悉的模块前先阅读 ARCHITECTURE.md,再用符号搜索确认真实实现。
每次任务先加载什么
- ARCHITECTURE.md:稳定边界、主链路和架构不变量。
- Yuxi Spec Loop:非平凡变更从提案、证据到收敛的流程。
- 工程信任系统:语义 Owner、证据、决策记录、派生审计和 gate 规则。
- 测试规范:unit、integration、E2E 的职责与命令。
- 贡献指南:分支、独立 Review、commit 和 PR 流程。
- 用户在当前任务中的明确要求优先于本文件;修改
backend/、web/或docs/时同时遵循该子树的AGENTS.md。子树规则只补充本目录,不复制回根文件。
任务与决策
- 开始实现前,把请求压缩为可验证的目标、非目标和验收主张;多步任务先给出每步带验证方式的简短计划。
- 假设显式记录后继续;只有当不同解释会改变验收结果、数据、安全或外部状态时才阻塞询问,并把候选解释一并列出而不是默默选择其一。存在明显更简单的方案时,先说明取舍再动手。
- 把"可以""也可以""类似这样""例如"当作简单方向,不是设计更大机制、配置项或兼容层的许可。
- 在真实语义 Owner 处闭合受影响的工程主张:源码、数据约束或当前契约拥有事实;独立 oracle 与负向案例证明它;实际 workflow 或可问责 Reviewer 产生拒绝后果。不要建立可独立编辑的中央主张清单或 claim ID 体系。
- 非平凡变更必须新增或更新一份 tracked 决策记录。记录问题、当前决定、真实替代项、后果和验证,不保存推理流水账或实现过程叙事。
docs/vibe/只用于本地临时计划,被 Git 忽略,不是组织记忆,也不能作为已完成事实的唯一来源。- 只修改验收标准需要的范围;不顺手重构、格式化或添加想象中的配置、兼容层和扩展点。
不能破坏的系统事实
- HTTP 路由保持薄;用例流程属于
yuxi.services,持久化查询属于yuxi.repositories。 - 普通请求先在 PostgreSQL 中持久化 Message 和 AgentRunRequest;只有 ready FIFO 队头创建 AgentRun,且每次投递 ARQ 前 owning transaction 都已提交。Redis 负责投递、短期事件、取消和缓存,不拥有最终业务状态。
- 同一用户、Agent、线程的普通请求按 FIFO 串行派发;Request 和 Run 是不同状态模型。
- AgentRun 的输出、事件、artifact 和错误必须绑定同一 request/run;禁止从相邻 Run 猜测结果。
- 非终态 Run 必须有明确执行 Owner、lease/heartbeat 或等价机制,以及崩溃后的可观察结局。
/api/system/health只表达进程 liveness;接流量前置条件由/api/system/ready证明,业务正确性仍由真实链路测试证明。- LangGraph checkpoint 只使用 PostgreSQL;API、worker 与 Agent 不提供本地后端选择或静默降级。
- 权限在后端依赖与 repository 可见性查询处最终执行;前端守卫、prompt、schema omission 和 UI 隐藏不是授权边界。
- LITE 的 shipping 启动、路由注册和能力发现不得导入或初始化知识库、图谱和评估的重运行时;仍保留的附件解析入口只在真实解析动作发生时惰性加载 parser。
- 沙盒虚拟路径、对象 URL 和宿主机路径不可混用;所有用户路径必须在 owning filesystem boundary 校验。
证据规则
- Agent 的完成报告、HTTP 200、日志关键词或 mock 调用次数都不是最终事实;重新读取数据库、文件、对象、DOM 或协议结果。
- 每个新 guard 都要有一个能恢复目标缺陷并使其在正确原因上失败的负向测试。
- 测试从最小相关集合开始,再按风险扩大;不要用 unit 结果替代真实 PostgreSQL、真实 HTTP、worker 或浏览器语义。
- Expected output、snapshot 和 fixture 只能显式更新并审阅;CI 不得一边生成 oracle 一边验证 oracle。
- 实际执行命令、结果和未执行原因写入 PR;未验证不能写成通过。
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.
- yesterday First seen · 78 lines · 1,873 tokens per session scan A 70fc05bf1261
Yuxi AGENTS.md is an instructions file published in the GitHub repository xerrors/Yuxi (6,591 stars, last pushed yesterday), licensed MIT. It adds 1,873 tokens to every session, about $0.0094 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
full-stack-ai-agent-template CLAUDE.md
Instructions for vstorm-co/full-stack-ai-agent-template, covering claude.md, project overview, commands, install dependencies and run tests.
full-stack-ai-agent-template AGENTS.md
Instructions for vstorm-co/full-stack-ai-agent-template, covering agents.md, project overview, commands, cli and architecture.
txtify CLAUDE.md
Instructions for lkmeta/txtify, covering txtify — agent guide, architecture (read this before touching job flow), commands, invariants & known traps and conventions.
openharness AGENTS.md
Instructions for mifunedev/openharness, covering open harness — orchestrator, what open harness is, 1. agent work stays inside the sandbox, 2. coding-harness choice does not change the workspace and 3. remote and unattended operation are normal.
pulse8-ai-cortex-knowledge-vault AGENTS.md
Instructions for synpulse8-opensource/pulse8-ai-cortex-knowledge-vault, covering agents.md — pulse8.ai cortex, repository map, environment variables, essential commands and dependencies.
GooFish-AIMonitor AGENTS.md
Instructions for tristanwqy/GooFish-AIMonitor, covering agents.md — 给 ai 协作者的工程指南, 黄金规则(先读这条), 验证命令(完成前必跑), 代码地图 and 数据模型要点.