Datasets:

LucioLiu
/

user-md

+# CREATION LOG — user-md
+## Packaging fix — 2026-07-18
+- Drafted a Claude Code packaging manifest so `claude plugin validate` can discover the package. *(That manifest lives in the not-yet-published packaging workspace and is **not** part of this release — do not look for a `.claude-plugin` directory in what you installed.)*
+- Retained root `plugin.json` as a compatibility mirror for existing skill-package consumers. Both manifests carry identical metadata, and the operative skill entry remains root `SKILL.md` for both packaging paths.
+- This is packaging-only; no behavior fixture, privacy sweep, or cross-harness test was run or newly marked PASS.
+## Phase 2 contract layer — 2026-07-18
+- Added an evidence-bearing agent-modpack component manifest, pinned-schema structural runner, privacy scan, and a real root-instance gitignore probe.
+- Added `SK-UM-P-001` and `SK-UM-N-001` as independently executable case specifications. Both remain `not-run`; cross-harness loading and real profile behavior are not claimed.
+## 1.0.0 — 2026-07-17 · Initial draft
+### What this is
+First full draft of the user-md skill: a discipline for keeping an AI's understanding of its user in a user-owned Markdown file family (`USER.md` + `USER.local.md` + `user/*.md` overflow), with a hand-written lane and an agent-curated lane.
+### Design decisions
+1. **Five-section `USER.md`.** Identity facts and preferences are kept in separate sections — separating the two is an established pattern in shipping agent-memory products (distinct namespaces for facts vs. preferences). The dynamic layer (`What I'm up to`) follows the long user-modeling research tradition of treating the user model as explicit, revisable hypotheses rather than fixed truth.
+2. **Two lanes, no silent rewriting.** Cross-product consensus supports both: hand-written and auto-collected user data are kept apart, and everything stays user-visible, editable, and deletable. user-md takes a deliberate anti-position against background mechanisms that quietly rewrite a user's profile: the curated lane is append-only, dated, sourced, confidence-tagged, and consolidation (the only legal rewrite of that lane) must leave a changelog line.
+3. **Machine-local split.** `USER.local.md` (gitignored) mirrors the established pattern of machine-specific, gitignored companion files in agent instruction systems. One-sentence test: if it stops being true when you switch machines, it does not belong in the portable file.
+4. **"persona" is banned throughout.** The word is doubly occupied — UX archetype users on one side, an agent's own character on the other — so any use is ambiguous. Section names follow the facts/preferences lineage instead.
+5. **Single-source thresholds.** Every size/frequency threshold (main-file budget, overflow line, promotion bar, consolidation cadence, short-message filter) is defined exactly once, in the threshold table of `references/curation.md`; every other file references the table instead of repeating numbers. Prevents drift when thresholds are tuned.
+6. **No phantom runtime.** Growth is a discipline plus optional wiring: a session-start hook is documented as *optional* with an example configuration, and the skill never claims an automated pipeline it does not ship.
+7. **Convention lineage stated plainly.** The `USER.md` name and every-session loading model come from the OpenClaw agent workspace; this is credited in the README, along with the clarification that the AGENTS.md standard has no user-profile slot. user-md occupies the user axis only and leaves agent-side files untouched.
+8. **Templates are 100% fictional.** All example content uses invented placeholders (Alex Rivera, Lisbon, a roguelike side project); nothing in the distribution derives from any real user's data.
+### Naming
+- Skill/repo name: `user-md` (adopts the OpenClaw `USER.md` convention).
+- README title candidates (slash-phrase describes the mechanism; settled in editorial review):
+  - A. `user-md / user memory as a file` — used in the first draft; dropped (misses the growth mechanism, which is the differentiator)
+  - B. `user-md / a USER.md that grows` — **final pick** (file shape + growth mechanism, matches the "This is a file that grows" positioning line)
+  - C. `user-md / the file that knows you` — dropped (effect, not mechanism)
+- Tagline: "Your AI's understanding of you — in a file you own. Moves when you move."
+- Spare: "Every repo has a README. Your agent deserves a USER.md."
+- Anchor sentence: "Every agent has an About Me. Yours should have an About You."
+### Known gaps / pre-release checklist
+- [x] Verify current OpenClaw docs for any official `USER.md` *inner* structure — verified in ops review: the official workspace page defines the file's role and loading model only, no inner structure; this skill's five sections stand as a de-facto structural proposal.
+- [x] Name-collision sweep (GitHub / npm / marketplace) for `user-md` — done: npm unclaimed; GitHub exact-name hits limited to one unrelated, inactive zero-star repo; no substantive collision.
+- [x] Editorial pass on README title candidate — settled on B (see Naming above).
+- [x] Manual PII sweep of every tracked public file and automated private-path/secret-pattern scan; templates re-confirmed fictional. — 2026-07-18 Phase 2 structural pass. This does not mark the independent privacy behavior case PASS.

LICENSE ADDED Viewed

	@@ -0,0 +1,77 @@

+PolyForm Noncommercial License 1.0.0
+Copyright (c) 2026 Lucio Liu
+Required Notice: Copyright (c) 2026 Lucio Liu
+# PolyForm Noncommercial License 1.0.0
+<https://polyformproject.org/licenses/noncommercial/1.0.0>
+## Acceptance
+In order to get any license under these terms, you must agree to them as both strict obligations and conditions to all your licenses.
+## Copyright License
+The licensor grants you a copyright license for the software to do everything you might do with the software that would otherwise infringe the licensor's copyright in it for any permitted purpose.  However, you may only distribute the software according to [Distribution License](#distribution-license) and make changes or new works based on the software according to [Changes and New Works License](#changes-and-new-works-license).
+## Distribution License
+The licensor grants you an additional copyright license to distribute copies of the software.  Your license to distribute covers distributing the software with changes and new works permitted by [Changes and New Works License](#changes-and-new-works-license).
+## Notices
+You must ensure that anyone who gets a copy of any part of the software from you also gets a copy of these terms or the URL for them above, as well as copies of any plain-text lines beginning with `Required Notice:` that the licensor provided with the software.  For example:
+> Required Notice: Copyright Yoyodyne, Inc. (http://example.com)
+## Changes and New Works License
+The licensor grants you an additional copyright license to make changes and new works based on the software for any permitted purpose.
+## Patent License
+The licensor grants you a patent license for the software that covers patent claims the licensor can license, or becomes able to license, that you would infringe by using the software.
+## Noncommercial Purposes
+Any noncommercial purpose is a permitted purpose.
+## Personal Uses
+Personal use for research, experiment, and testing for the benefit of public knowledge, personal study, private entertainment, hobby projects, amateur pursuits, or religious observance, without any anticipated commercial application, is use for a permitted purpose.
+## Noncommercial Organizations
+Use by any charitable organization, educational institution, public research organization, public safety or health organization, environmental protection organization, or government institution is use for a permitted purpose regardless of the source of funding or obligations resulting from the funding.
+## Fair Use
+You may have "fair use" rights for the software under the law. These terms do not limit them.
+## No Other Rights
+These terms do not allow you to sublicense or transfer any of your licenses to anyone else, or prevent the licensor from granting licenses to anyone else.  These terms do not imply any other licenses.
+## Patent Defense
+If you make any written claim that the software infringes or contributes to infringement of any patent, your patent license for the software granted under these terms ends immediately. If your company makes such a claim, your patent license ends immediately for work on behalf of your company.
+## Violations
+The first time you are notified in writing that you have violated any of these terms, or done anything with the software not covered by your licenses, your licenses can nonetheless continue if you come into full compliance with these terms, and take practical steps to correct past violations, within 32 days of receiving notice.  Otherwise, all your licenses end immediately.
+## No Liability
+***As far as the law allows, the software comes as is, without any warranty or condition, and the licensor will not be liable to you for any damages arising out of these terms or the use or nature of the software, under any kind of legal claim.***
+## Definitions
+The **licensor** is the individual or entity offering these terms, and the **software** is the software the licensor makes available under these terms.
+**You** refers to the individual or entity agreeing to these terms.
+**Your company** is any legal entity, sole proprietorship, or other kind of organization that you work for, plus all organizations that have control over, are under the control of, or are under common control with that organization.  **Control** means ownership of substantially all the assets of an entity, or the power to direct its management and policies by vote, contract, or otherwise.  Control can be direct or indirect.
+**Your licenses** are all the licenses granted to you for the software under these terms.
+**Use** means anything you do with the software requiring one of your licenses.

README.md ADDED Viewed

	@@ -0,0 +1,172 @@

+---
+license: other
+license_name: polyform-noncommercial-1.0.0
+license_link: https://polyformproject.org/licenses/noncommercial/1.0.0/
+pretty_name: "user-md - keep your AI's understanding of you in a file you own"
+language:
+  - en
+  - zh
+tags:
+  - agent-skill
+  - agent-skills
+  - claude-skill
+  - claude-code
+  - skill
+  - agentskills
+  - prompt-engineering
+task_categories:
+  - text-generation
+size_categories:
+  - n<1K
+viewer: false
+---
+> **user-md** - Keep your AI's understanding of you in a Markdown file you own - a five-section USER.md with a hand-written lane and an append-only agent-curated lane.
+>
+> **Get it.** This Hugging Face repo mirrors the source; pick whichever works for you:
+>
+> ```bash
+> # Hugging Face (this page)
+> hf download LucioLiu/user-md --repo-type dataset --local-dir ./user-md
+> git clone https://huggingface.co/datasets/LucioLiu/user-md
+>
+> # GitLab mirror - plain git, always current
+> git clone https://gitlab.com/LucioLiu/user-md.git
+> ```
+>
+> Then move the folder into your agent's skills directory - `~/.claude/skills/user-md/` for Claude Code, `~/.agents/skills/user-md/` for Codex, or your client's equivalent.
+>
+> **Licence: PolyForm Noncommercial 1.0.0** - see the `LICENSE` file in this repo, which is authoritative.
+---
+**English** | [简体中文](README.zh-CN.md)
+# user-md / a USER.md that grows
+> Your AI's understanding of you — in a file you own. Moves when you move.
+## Know your user
+user-md is a discipline for keeping your AI's understanding of you in a single Markdown file that you own: `USER.md`. It ends the re-explaining — your role, your standards, how you like answers delivered — and unlike vendor cloud memory, it moves when your folder moves. Hand-written sections capture who you are; an agent-curated section grows with dated, confidence-tagged observations.
+**Every agent has an About Me. Yours should have an About You.**
+Leave `USER.md` empty and your agent still runs — but every session starts with zero knowledge of you. And unlike memory features locked inside a vendor's cloud account, this file sits in your folder — readable and editable by you, usable by any agent you point at it. As one reviewer put it about a vendor's GDPR-only memory export: "The export feature exists to satisfy regulators, not users." (Nick Sawinyh)
+## Three layers, two lanes
+```mermaid
+flowchart TB
+    subgraph U["USER.md — portable, loaded every session"]
+        direction TB
+        subgraph H["Hand-written lane — you author it, the agent may only suggest"]
+            W1["Who I am"]
+            W2["How I work"]
+            W3["Our relationship"]
+            W4["What I'm up to"]
+        end
+        subgraph A["Agent-curated lane — append-only, dated, confidence-tagged"]
+            W5["Learned by your agent"]
+        end
+    end
+    LOCAL["USER.local.md — machine-local, gitignored<br/>env · toolchain · paths · credential pointers"]
+    PROJ["Project-local files<br/>facts only true inside one project"]
+    OVER["user/*.md overflow files<br/>details sunk when the main file outgrows its budget"]
+    W5 -- "repeated observation + your explicit OK" --> H
+    W5 -- "overflow" --> OVER
+    U -. "breaks on machine switch? → goes here instead" .-> LOCAL
+    U -. "only true in one project? → goes here instead" .-> PROJ
+```
+**Three layers** — where a fact about you belongs:
+| Layer | File | Test |
+|---|---|---|
+| Portable profile | `USER.md` | still true on any machine, in any project |
+| Machine-local | `USER.local.md` (gitignored) | stops being true when you switch machines |
+| Project-local | that project's own context file | stops being true outside one project |
+**Two lanes** inside `USER.md`:
+- **Hand-written lane** — `Who I am`, `How I work`, `Our relationship`, `What I'm up to`. You author these. The agent may suggest edits and may apply one only after your explicit confirmation; it never rewrites this lane silently.
+- **Agent-curated lane** — `Learned by your agent`. Append-only. Every entry carries a date, a source event, and a confidence tag; single observations are explicitly marked unverified. Repeated observations can be *promoted* into the hand-written lane — only with your explicit OK. Nothing is ever silently rewritten; everything stays visible, editable, and deletable.
+The main file is kept deliberately small so it can be loaded in full every session; details overflow into `user/*.md` helper files. Exact size, promotion, and consolidation thresholds live in one place: [`references/curation.md`](references/curation.md).
+## How it compares
+| | **user-md** | Cloud memory features | Memory-layer products (service + DB) | Vanilla OpenClaw `USER.md` |
+|---|---|---|---|---|
+| Where it lives | a file in your folder | your vendor account | a service + database | a file in your folder |
+| Read/edit directly | plain Markdown, any editor | product UI only | API/SDK | plain Markdown |
+| Portable across vendors | yes — copy the folder | no | depends on self-hosting | yes |
+| Growth mechanism | two-lane curation discipline | automatic, sometimes silent | automatic pipeline | none — static, hand-filled |
+| Audit trail | every entry dated + sourced | limited | varies | none |
+The gap user-md fills: tools that are file-shaped have no growth mechanism, and tools that grow aren't file-shaped. This is a file that grows — under explicit, auditable rules.
+## Compatibility
+The `USER.md` convention comes from the **OpenClaw** agent workspace, where the file is defined as "Who the user is and how to address them. Loaded every session." (docs.openclaw.ai). It is *not* part of the AGENTS.md standard — that standard has no user-profile slot. user-md adopts the OpenClaw name and every-session loading model, and adds the layering and curation discipline on top.
+### OpenClaw
+Drop-in. `USER.md` occupies the user axis only; `SOUL.md`, `IDENTITY.md`, and `MEMORY.md` stay untouched. Copy the templates into your workspace and you are done.
+### Claude Code
+Add one reference line to your `CLAUDE.md` (project-level, or `~/.claude/CLAUDE.md` for global use):
+```
+@USER.md
+```
+Optionally, a SessionStart hook can be wired instead — an example configuration ships in [`SKILL.md`](SKILL.md). The hook is optional wiring, not a built-in runtime; the discipline works without it.
+### Any other harness
+Any tool that reads instructions at session start can use this. Add a line like *"Read the relevant sections of USER.md before starting work"* to its system prompt or rules file. Cross-vendor transfer and session injection require the user's authorization; load only the sections needed for the task instead of treating "portable" as blanket disclosure consent.
+## Install (2 minutes)
+1. Copy `templates/USER.md` and `templates/USER.local.md` into your workspace root.
+2. Add `USER.local.md` to `.gitignore` — before your first commit.
+3. Wire session loading for your harness (see Compatibility above).
+Then spend five minutes filling in the first four sections of `USER.md`. Honest bullet points beat polished prose.
+To teach your agent the curation rules, also install this folder as a skill/plugin:
+```
+gh skill install LucioLiu/user-md user-md --agent claude-code --scope user
+```
+or copy the folder into your harness's skills directory (Chinese-speaking users should copy [`SKILL.zh-CN.md`](SKILL.zh-CN.md) instead and rename it to `SKILL.md` locally). Or just tell your agent: *"Read `SKILL.md` in this folder and set up user-md for me."*
+This release ships the root `plugin.json` as its package manifest. (A `.claude-plugin` packaging variant exists in an unpublished workspace and is deliberately not part of this release — don't look for that directory in what you installed.)
+## Uninstall
+Remove the plugin/skill folder. Your `USER.md` file family stays where it is — those files are yours, which is rather the point.
+## Version & self-check
+- Current version: **1.1.0** (see `plugin.json`).
+- Repository: `https://github.com/LucioLiu/user-md`
+- Self-check: diff your installed `SKILL.md` and `plugin.json` against the repo's main branch.
+## Verification boundary
+The unified structural entrypoint, privacy/gitignore checks, and independent behavior bundle are **not published in this repository yet** — don't expect a `tests` directory in what you installed. A structural PASS would not prove curation behavior or cross-harness loading anyway; both frozen behavior cases remain `not-run` until independently executed and judged.
+## Privacy
+This file family is PII by nature. `USER.local.md` must be gitignored. Review `USER.md` manually before pushing it to any public repository. Never store plaintext credentials in either file — pointers only. "Moves when you move" means the user can transfer the file; it does not grant automatic cross-project, cross-vendor, or cross-model sharing. Each transfer/load must follow explicit user authorization and minimum necessary disclosure. All examples shipped in the templates are fictional.
+## License
+[PolyForm Noncommercial 1.0.0](LICENSE) — free for personal and non-commercial use.
+---
+> Both language editions are maintained in sync. If they ever diverge, the **English edition** is authoritative — please open an [issue](https://github.com/LucioLiu/user-md/issues) if you spot one.

README.zh-CN.md ADDED Viewed

	@@ -0,0 +1,132 @@

+[English](README.md) | **简体中文**
+# user-md / 一份会成长的 USER.md
+> 你的 AI 对你的理解——存在一个你自己拥有的文件里。你的文件夹搬到哪，它跟到哪。
+## 让你的 AI 认识你
+user-md 是一套纪律，把你的 AI 对你的理解统一保存在一个你自己拥有的 Markdown 文件里：`USER.md`。它终结了"每次都要重新解释一遍"——你的角色、你的标准、你喜欢答案怎么给你——而且和厂商的云端记忆不同，它随你的文件夹一起搬家。手写区记录你是谁；agent 养护区随着带日期、带置信标签的观察持续生长。
+**每个 agent 都有一份"关于我"。你也该有一份"关于你"。**
+`USER.md` 留空，agent 照样能跑——只是每次会话都对你一无所知。而且和被锁在厂商云端账号里的记忆功能不同，这份文件就放在你的文件夹里——你可读可改，你指给哪个 agent 用它就能用。正如一位评论者谈到某厂商"仅为满足 GDPR 而做"的记忆导出功能时所说："这个导出功能是为了应付监管，不是为了用户。"（Nick Sawinyh）
+## 三层结构，两条轨道
+```mermaid
+flowchart TB
+    subgraph U["USER.md —— 可迁移，每会话加载"]
+        direction TB
+        subgraph H["手写轨道 —— 你亲笔写，agent 只能建议"]
+            W1["我是谁"]
+            W2["我怎么工作"]
+            W3["我们的关系"]
+            W4["我在忙什么"]
+        end
+        subgraph A["agent 养护轨道 —— 只追加、带日期、带置信标签"]
+            W5["agent 学到的"]
+        end
+    end
+    LOCAL["USER.local.md —— 机器本地，已 gitignore<br/>环境 · 工具链 · 路径 · 凭据指针"]
+    PROJ["项目本地文件<br/>只在单个项目里成立的事实"]
+    OVER["user/*.md 溢出文件<br/>主文件体量超预算时下沉的细节"]
+    W5 -- "重复观察 + 你明确同意" --> H
+    W5 -- "溢出" --> OVER
+    U -. "换机器就失效？→ 改存这里" .-> LOCAL
+    U -. "只在单个项目里成立？→ 改存这里" .-> PROJ
+```
+**三层**——关于你的一条事实该归哪：
+| 层 | 文件 | 判据 |
+|---|---|---|
+| 可迁移画像 | `USER.md` | 换任何机器、任何项目依然成立 |
+| 机器本地 | `USER.local.md`（已 gitignore） | 换机器就不再成立 |
+| 项目本地 | 该项目自己的上下文文件 | 出了这个项目就不成立 |
+`USER.md` 内部的**两条轨道**：
+- **手写轨道**——`我是谁`、`我怎么工作`、`我们的关系`、`我在忙什么`。由你亲笔撰写。agent 可以建议修改，但只有在你明确确认后才能应用——绝不静默改写这条轨道。
+- **agent 养护轨道**——`agent 学到的`。只追加。每条记录都带日期、来源事件与置信标签；单次观察必须显式标注"待验证"。重复出现的观察可以被*升级*进手写轨道——但必须经你明确同意。任何内容都不会被静默改写；一切内容对你可见、可编辑、可删除。
+主文件刻意保持小体量，好让它能每次会话都全量加载；细节溢出到 `user/*.md` 辅助文件里。确切的体量、升级与整合阈值统一定义在一处：[`references/curation.md`](references/curation.md)。
+## 它和同类方案的区别
+| | **user-md** | 云端记忆功能 | 记忆层产品（服务+数据库） | 原生 OpenClaw `USER.md` |
+|---|---|---|---|---|
+| 存在哪 | 你文件夹里的一个文件 | 你的厂商账号里 | 一个服务+数据库 | 你文件夹里的一个文件 |
+| 能否直接读写 | 纯 Markdown，任意编辑器 | 仅限产品自带界面 | API/SDK | 纯 Markdown |
+| 能否跨厂商迁移 | 能——拷贝文件夹即可 | 不能 | 取决于是否自建 | 能 |
+| 成长机制 | 双轨养护纪律 | 自动，有时静默 | 自动化管道 | 无——静态、纯手填 |
+| 审计轨迹 | 每条都带日期+来源 | 有限 | 因产品而异 | 无 |
+user-md 填补的空档：文件形态的工具没有成长机制，会成长的工具又不是文件形态。这是一份**会成长的文件**——在显式、可审计的规则之下成长。
+## 兼容性
+`USER.md` 这个约定源自 **OpenClaw** agent 工作台，该文件在那里被定义为"用户是谁、该怎么称呼他们。每次会话都会加载。"（docs.openclaw.ai）。它*不属于* AGENTS.md 标准的一部分——那个标准里没有用户画像这个槽位。user-md 沿用了 OpenClaw 的命名与每会话加载模型，并在其上加了分层与养护纪律。
+### OpenClaw
+即插即用。`USER.md` 只占据"用户"这一根轴；`SOUL.md`、`IDENTITY.md`、`MEMORY.md` 保持不动。把模板拷进你的工作台即可。
+### Claude Code
+在你的 `CLAUDE.md`（项目级，或全局用 `~/.claude/CLAUDE.md`）里加一行引用：
+```
+@USER.md
+```
+也可以选择改挂 SessionStart hook——示例配置见 [`SKILL.md`](SKILL.md)。这个 hook 是可选接法，不是内置运行时；不挂它纪律照样成立。
+### 其他任意 harness
+任何在会话开始时读取指令的工具都能用这套。在它的系统提示或规则文件里加一句类似"开工前先读 USER.md 的相关小节"。跨厂商传输与会话内注入都需要用户授权；只加载任务所需的小节，别把"可迁移"当成无条件的全量披露许可。
+## 安装（2 分钟）
+1. 把 `templates/USER.md` 和 `templates/USER.local.md` 拷进你的工作台根目录。
+2. 在第一次 commit 之前，把 `USER.local.md` 加进 `.gitignore`。
+3. 按你的 harness 接上会话加载（见上面「兼容性」）。
+然后花五分钟填好 `USER.md` 的前四节。诚实的要点胜过打磨过的长句。
+想让你的 agent 也学会养护规则，把这个文件夹也装成 skill/plugin：
+```
+gh skill install LucioLiu/user-md user-md --agent claude-code --scope user
+```
+或者直接把文件夹拷进你 harness 的 skills 目录（中文用户请改拷贝 [`SKILL.zh-CN.md`](SKILL.zh-CN.md) 并在本地重命名为 `SKILL.md`）。或者干脆跟你的 agent 说："读一下这个文件夹里的 `SKILL.md`，帮我把 user-md 配起来。"
+本次发行以根目录的 `plugin.json` 作为包清单。（另有一份 `.claude-plugin` 打包变体留在未发布的工作区里，**刻意不随本次发行分发**——别在你装到的副本里找那个目录。）
+## 卸载
+删掉 plugin/skill 文件夹即可。你的 `USER.md` 文件族原地不动——那些文件是你的，这正是重点所在。
+## 版本与自检
+- 当前版本：**1.1.0**（见 `plugin.json`）。
+- 仓库地址：`https://github.com/LucioLiu/user-md`
+- 自检方式：把你装的 `SKILL.md` 与 `plugin.json` 对比仓库 main 分支的 diff。
+## 验证边界
+统一结构测试入口、隐私/gitignore 检查、独立行为 case bundle **尚未在本仓发布**——别指望你装到的副本里有 `tests` 目录。何况结构测试 PASS 本来也不能证明养护行为或跨 harness 加载已验证：两条冻结行为用例在被独立执行与裁决之前保持 `not-run`。
+## 隐私
+这个文件族天生就是 PII——按你手上最敏感的内容来对待它。`USER.local.md` 必须 gitignore。推到任何公开仓库之前必须人工审阅一遍 `USER.md`。两个文件都绝不能存明文凭据——只放指针。"随你搬家"指的是用户自己可以携带这份文件，不代表自动获得跨项目、跨厂商或跨模型的共享授权。每一次转移/加载都必须遵循用户的明确授权与最小必要披露原则。模板里附带的全部示例均为虚构。
+## License
+[PolyForm Noncommercial 1.0.0](LICENSE) — 个人与非商业使用免费。
+---
+> 中英文双版同步维护；如有出入，以**英文版**为准——发现不一致请开 [issue](https://github.com/LucioLiu/user-md/issues)。

SKILL.md ADDED Viewed

	@@ -0,0 +1,134 @@

+---
+name: user-md
+description: Keep your AI's understanding of you in a Markdown file you own - a five-section USER.md with a hand-written lane and an append-only agent-curated lane (dated, sourced, confidence-tagged), machine-local facts split into a gitignored USER.local.md. No silent rewrites; moves when your folder moves. Use when setting up a USER.md or user profile for an agent, recording user preferences or working style, deciding which file a fact about the user belongs in, curating agent-observed habits with confidence tags, onboarding a new agent to an existing user, or carrying personalization across machines and harnesses. 中文触发：用户画像、用户偏好沉淀、记住我的习惯、USER.md 建档养护。
+---
+**English** | [简体中文](SKILL.zh-CN.md)
+# user-md: user-profile file discipline
+> Your AI's understanding of you should live in a file **you own** — not locked inside some vendor's cloud account. This document is the profile file's **discipline** (it teaches an agent how to build, maintain, and use this file), not a memory system (it bundles no runtime and no automatic pipeline). Read the details in `references/` on demand — **those two files are written in Chinese**, and this English SKILL is self-contained without them: curation details in `curation.md`, layering criteria and host integration in `layering.md`. **The size and frequency thresholds for the whole file family are defined exactly once, in the threshold table in `references/curation.md`** — this document and every other file only ever reference it, never re-state the numbers.
+## Core stance (four sentences)
+1. **You own the file**: plain Markdown, readable/editable/deletable by any editor, moves when your folder moves.
+2. **Two lanes, clearly separated**: the hand-written lane belongs to the user — the agent may only suggest, never rewrite it; the curated lane belongs to the agent — append-only, every entry leaves a trace.
+3. **No silent rewrites**: any rewrite leaves a trace (one line in the change log); every piece of content stays visible, editable, and deletable by the user.
+4. **No fabricated facts**: the curated lane only records genuine observations; a single observation must carry an unverified tag; anything not actually observed doesn't get written.
+## 1. Build: the file family and the five sections
+### The three-file family
+| File | What it holds | Traits |
+|---|---|---|
+| `USER.md` | The user profile that still holds across machines and projects | Portable; injected in full every session, must stay small (threshold in `references/curation.md`) |
+| `USER.local.md` | Content that stops being true on a different machine: environment / paths / toolchain / credential pointers | Machine-local; **must be gitignored** |
+| `user/*.md` overflow helper files | Details moved out once the curated lane crosses its sink-down line | Main file keeps a one-line index, loaded on demand |
+### USER.md's five sections (settled structure — template in `templates/USER.md`)
+1. `## Who I am` — identity facts: role / background / how to address them. **Keep facts and preferences stored separately** (splitting these two kinds of information is a common practice across several product-grade memory systems).
+2. `## How I work` — preferences and style: communication style / deliverable standards / acceptance habits / default authorization boundary.
+3. `## Our relationship` — relationship and collaboration: what to ask about, what to just do, how feedback should be given.
+4. `## What I'm up to` — the dynamic-cognition layer: current goals and what's in progress — **explicitly revisable assumptions**, updated as evidence comes in, cleared once stale.
+5. `## Learned by your agent` — the agent-curated lane (rules in the two-lane section below).
+**Naming discipline**: section names follow the facts/preferences lineage; **the word "persona" is banned throughout** — in a UX context it means a fictional archetype user, in an agent context it means the agent's own personality; the double meaning is guaranteed to cause ambiguity if used.
+**Getting started**: copy the two templates from `templates/` into your workspace root, add `USER.local.md` to `.gitignore`, and spend five minutes filling in the first four sections. It still runs even left empty — but the agent starts every session with zero knowledge of you.
+## 2. Maintain: the two-lane system and curation discipline
+### The two-lane system
+- **The hand-written lane (the first four sections)**: written by the user themselves. When the agent notices content that's stale or contradictory, it **may only suggest** (giving suggested text plus a reason) — the user makes the change themselves, or it's applied only after explicit consent, and one line goes into the change log at the end of the file once it's done.
+- **The curated lane (the fifth section)**: appended by the agent. Each entry's format:
+  `- [YYYY-MM-DD · source event] the rule text itself. [single observation, unverified]`
+  Date + source is every entry's birth certificate; a single observation must carry the unverified tag. Once the same type of observation recurs enough to cross the promotion threshold (in `references/curation.md`), a promotion proposal may be raised — **it only moves into the hand-written lane after the user confirms**. Full format and process in `references/curation.md`.
+### Signals and false positives
+- Three categories of signal worth curating: **correction** (the user corrected your approach or understanding), **repeated preference** (the same kind of request recurs across different occasions), **explicit statement** (the user directly says "remember this" / "always do X from now on").
+- **Hard rule: append on the spot**: once a signal hits one of the three categories and clears the false-positive filter → **append it to the curated lane on the spot** — don't batch it up, and don't leave it to be reconstructed from memory at the end of the session. Recognizing it and not writing it down is the same as not having this discipline at all.
+- False-positive filtering: don't curate one-off context (a requirement specific to this task ≠ a standing preference); ignore extremely short messages by default (filter threshold in `references/curation.md`); down-weight confidence for anything said in an emotional moment; instructions tightly bound to a specific task belong to the task, not the profile.
+### Consolidation and size
+- Consolidate and de-duplicate on a regular cadence (period in `references/curation.md`'s threshold table): merge synonymous entries, clear out stale ones; **consolidation and promotion-driven removal are the only two legitimate scenarios for rewriting the curated lane**, and both must leave a trace in the change log.
+- Once the main file crosses its size target, or the curated lane crosses its sink-down line (thresholds both in `references/curation.md`) → sink the details into `user/*.md` helper files, leaving an index line in the main file.
+## 3. Use: injection and routing
+### Injected every session
+The main file is kept small so the host can load what a task needs; whether it's loaded across projects, vendors, or models must be authorized by the user, following minimum necessary disclosure. Pick one wiring per host (details for each host are in the Compatibility section of the README):
+- **A reference line in the instructions file**: e.g. Claude Code adds one line, `@USER.md`, to CLAUDE.md;
+- **A host with no reference mechanism**: add a line to the system prompt or rules file, like "read USER.md before starting work each session";
+- **An optional session-start hook** (optional wiring — this skill bundles no runtime of its own, and the discipline holds without one). Example Claude Code config:
+```json
+{
+  "hooks": {
+    "SessionStart": [
+      { "hooks": [ { "type": "command", "command": "cat USER.md" } ] }
+    ]
+  }
+}
+```
+(A POSIX example; swap in the equivalent command for Windows or another host. Whether to wire it, and how, is entirely up to the user.)
+### Three-line routing test (a weak-model fallback — full detail in `references/layering.md`)
+1. Stops being true on a different machine → `USER.local.md`;
+2. Stops holding on a different project/role → that project's own local file;
+3. Everything else → `USER.md`.
+**Pointer discipline**: don't re-copy a fact that already has an authoritative source elsewhere — leave only a pointer line ("see such-and-such file for detail") to prevent the two sources from drifting apart. Worked examples in `references/layering.md`.
+## 4. Two-tier layout for multi-agent teams (one user × several role agents)
+A single `USER.md` is enough for one agent; when the same user runs **several role agents** (one each for orchestration / engineering / testing / design, each with its own working directory), the profile splits into **two tiers** — never one full profile copy per agent home:
+- **Shared tier (one file, team-shared location)**: preferences / red lines / decision standards that hold across roles — "lead with the conclusion", "human confirmation before publishing" — stored once, read by every agent.
+- **Role tier (0..n files, each agent's own directory)**: only entries that **stop being true when the role changes** — "use strict mode when reviewing code" belongs to the engineering agent's role tier; the design agent never needs it.
+**The single routing test**: *would this stop being true in another role?* Only then does it stay in the role tier; everything else goes to the shared tier. This is routing line 2 of the three-line test, made concrete for the multi-agent case.
+**Three companion rules**:
+1. **No file at birth**: a newly onboarded agent gets no pre-created role-tier file (an empty shell carries no information); create it when the first role-specific preference appears (YAGNI).
+2. **Same entry format**: both tiers use the curated-lane format (date + source + unverified tag); promotion and consolidation rules are unchanged.
+3. **Lift on consolidation**: during periodic consolidation, check each role tier — entries that actually hold across roles get **lifted to the shared tier and deleted from the role tier**, preventing the same preference from drifting across N copies (with N copies, one is always stale, and the agent reading the stale one works from an outdated preference).
+> Field provenance: converged from a 16-agent production team (sanitized) — initially every agent home grew its own profile, the same preference was copied in several places and versions drifted apart; converging to two tiers with "stops being true when the role changes" as the sole test ended the drift. Single-user single-agent readers can skip this section.
+## Privacy red lines
+- This file family is inherently full of PII — treat it as sensitively as the most sensitive thing you handle.
+- "Moves with your folder" is user portability, not automatic authorization for cross-project, cross-vendor, or cross-model sharing; confirm scope before loading, and inject only the part the current task actually needs.
+- `USER.local.md` **must be gitignored**, and must never be committed.
+- `USER.md` **must be manually reviewed** before pushing to any public repository: check real names, contact details, employer specifics, and credential pointers item by item.
+- Plaintext passwords / API keys / tokens **must never be written** into this file family — only a pointer to where the credential lives and how to call it.
+- Templates must never ship pre-filled with real content — every template this skill ships is placeholders plus fictional examples.
+## Quick examples
+- The user says, once again, "give me the conclusion first in replies," and the same-type unverified entry in the curated lane has crossed the promotion threshold (see `references/curation.md`) → draft a promotion proposal (fold it into one sentence, intended for How I work), and move it only once the user confirms.
+- The user says "use pnpm on this machine, not npm" → stops being true on a different machine → record it in `USER.local.md`.
+- The user says, inside a particular repo, "commit messages for this project are in Spanish" → doesn't hold outside this project → record it in that project's local file, don't write it into `USER.md`.
+- The user vents late at night, "I'm so done, don't show me any more long replies" → an emotional moment → curate it as a single observation with confidence down-weighted, don't touch the hand-written lane directly, and certainly don't treat it as a hard rule.
+- The user's hand-written lane says "I'm working on project X" but it hasn't come up in three months → the agent **suggests** updating the What I'm up to section, with a reason attached; not a single word changes before the user nods.
+## Version & update self-check
+- Current version: **1.1.0** (matches the root `plugin.json`, which is this release's package manifest).
+- Repository: `https://github.com/LucioLiu/user-md`
+- Self-check: for plugin installs, the host's update mechanism follows the version number; for manual clone / template-copy installs, diff this file against the repo's main branch to see if you're behind. Your own `USER.md` file family is unaffected by a skill update — that's your data, not part of the skill.
+---
+> Both language editions are maintained in sync. If they ever diverge, the **Chinese edition** is authoritative — please open an [issue](https://github.com/LucioLiu/user-md/issues) if you spot one.

SKILL.zh-CN.md ADDED Viewed

	@@ -0,0 +1,134 @@

+---
+name: user-md
+description: Keep your AI's understanding of you in a Markdown file you own - a five-section USER.md with a hand-written lane and an append-only agent-curated lane (dated, sourced, confidence-tagged), machine-local facts split into a gitignored USER.local.md. No silent rewrites; moves when your folder moves. Use when setting up a USER.md or user profile for an agent, recording user preferences or working style, deciding which file a fact about the user belongs in, curating agent-observed habits with confidence tags, onboarding a new agent to an existing user, or carrying personalization across machines and harnesses. 中文触发：用户画像、用户偏好沉淀、记住我的习惯、USER.md 建档养护。
+---
+[English](SKILL.md) | **简体中文**
+# user-md：用户画像文件纪律
+> 你的 AI 对你的理解，应当存在一个**你自己拥有的文件**里——不是锁在某家厂商的云端账号里。本文是画像文件的**纪律**（教 agent 怎么建、怎么养、怎么用这个文件），不是记忆系统（不含运行时与自动管道）。细则按需读取 `references/`：养成细则 `curation.md`、分层判据与宿主关系 `layering.md`。**全套文件的体量与频次阈值只在 `references/curation.md` 的阈值表定义一次**，本文与其他文件一律引用、不复抄数字。
+## 核心立场（四句）
+1. **文件你拥有**：明文 Markdown，任何编辑器可读、可改、可删，随文件夹迁移（moves when you move）。
+2. **双轨分明**：手写区归用户，agent 只可建议不可改写；沉淀区归 agent，只可追加、条条留痕。
+3. **不静默改写**：任何改写留痕（变更记录一行）；全部内容对用户可见、可编辑、可删除。
+4. **不编事实**：沉淀区只记真实观察，单次观察必须带待验证标；没观察到的不写。
+## 一、建：文件族与五节
+### 文件族三件
+| 文件 | 装什么 | 特性 |
+|---|---|---|
+| `USER.md` | 跨机器、跨项目仍成立的用户画像 | 可迁移；每会话全量注入，须保持小体量（阈值见 `references/curation.md`） |
+| `USER.local.md` | 换台机器就失效的内容：环境/路径/工具链/凭据指针 | 机器本地；**必须 gitignore** |
+| `user/*.md` 溢出辅文件 | 沉淀区超下沉线时移出的细节 | 主文件留一行索引，按需加载 |
+### USER.md 五节（定案结构，模板见 `templates/USER.md`）
+1. `## Who I am` — 身份事实：角色/背景/怎么称呼。**事实与偏好分开存**（两类信息分开是多家产品级记忆系统的共同做法）。
+2. `## How I work` — 偏好与风格：沟通方式/交付物标准/验收习惯/默认授权边界。
+3. `## Our relationship` — 关系与协作：什么该问、什么直接做、反馈怎么给。
+4. `## What I'm up to` — 动态认知层：当前目标与在做的事——**显式可修订假设**，随证据更新，过期即清。
+5. `## Learned by your agent` — agent 沉淀区（规则见下节双轨制）。
+**命名纪律**：节名沿 facts / preferences 谱系；**persona 一词全文禁用**——UX 语境指虚构原型用户、agent 语境指 agent 自身人格，双重占用，用了必歧义。
+**起步**：从 `templates/` 拷贝两份模板到工作区根目录，把 `USER.local.md` 加进 `.gitignore`，花五分钟填前四节。留空也能跑，但 agent 每次会话对你零认知。
+## 二、养：双轨制与沉淀纪律
+### 双轨制
+- **手写区（前四节）**：用户亲笔。agent 发现内容过期或矛盾时**只可建议**（给出建议文本+理由），用户自己动手或明确同意后才改，改完在文件尾变更记录留一行。
+- **沉淀区（第五节）**：agent 追加。每条格式：
+  `- 【YYYY-MM-DD · 来源事件】规则正文。【单次观察待验证】`
+  日期+来源是每条的出生证明；单次观察必须带待验证标。同型观察重现达到升区门槛（阈值见 `references/curation.md`）才可发起升区提议，**用户确认后才迁入手写区**。完整格式与流程见 `references/curation.md`。
+### 信号与误报
+- 可沉淀信号三类：**纠正**（用户纠正了你的做法或理解）、**重复偏好**（同类要求在不同场合再现）、**明确表达**（用户直说「记住」「以后都…」）。
+- **当场追加硬规则**：命中三类信号且通过误报过滤 → **当场追加**沉淀区——不攒批、不留到会话结束凭记忆补写。识别到了却不写，等于没有这条纪律。
+- 误报过滤：一次性语境不沉淀（本任务特有要求 ≠ 长期偏好）；极短消息默认忽略（过滤线见 `references/curation.md`）；情绪时刻的表达压低置信；与任务强绑定的指令归任务、不归画像。
+### 整合与体量
+- 定期整合去重（周期见 `references/curation.md` 阈值表）：同义条目合并、过期条目清理；**整合与升区迁出是重写沉淀区仅有的两个合法场景**，均须在变更记录留痕。
+- 主文件超体量目标、或沉淀区超下沉线（阈值均见 `references/curation.md`）→ 细节下沉 `user/*.md` 辅文件，主文件留索引行。
+## 三、用：注入与分流
+### 每会话注入
+主文件保持小体量是为了让宿主能按任务加载所需内容；是否跨项目、跨供应商或跨模型加载，必须由用户授权，并遵守最小必要披露。接法按宿主选其一（各宿主详情见 README 的 Compatibility 节）：
+- **指令文件引用行**：如 Claude Code 在 CLAUDE.md 加一行 `@USER.md`；
+- **无引用机制的宿主**：系统提示或规则文件加一句「会话开始先读 USER.md」；
+- **可挂 session-start hook**（可选接法——本 skill 不自带任何运行时，不挂 hook 纪律照样成立）。Claude Code 示例配置：
+```json
+{
+  "hooks": {
+    "SessionStart": [
+      { "hooks": [ { "type": "command", "command": "cat USER.md" } ] }
+    ]
+  }
+}
+```
+（POSIX 示例；Windows 或其他宿主换等价命令。挂不挂、怎么挂由用户决定。）
+### 分流判据三句（弱模型兜底，详见 `references/layering.md`）
+1. 换台机器就失效 → `USER.local.md`；
+2. 换个项目/岗位就不成立 → 该项目自己的本地文件；
+3. 其余 → `USER.md`。
+**指针纪律**：已有权威源的事实不复抄，只放指针行（「详见某文件」）——防双源漂移。判例见 `references/layering.md`。
+## 四、多 agent 团队两层制（一个用户 × 多个分工 agent 时）
+单 agent 场景一份 `USER.md` 就够；当同一个用户配了**多个分工 agent**（编排 / 工程 / 测试 / 设计各一个、各有自己的工作目录）时，画像文件按**两层**放，不许每个 agent 家各养一份全量画像：
+- **通用层（一份，团队共享位置）**：跨岗位仍成立的偏好 / 雷区 / 拍板标准——"回复先给结论""发布前必须人工确认"这类换哪个 agent 都成立的条目，只在通用层存一份，所有 agent 都读它。
+- **岗位层（0..n 份，各 agent 自己目录）**：只装**换个岗位就不成立**的条目——"评审代码时用严格模式"归工程 agent 的岗位层，设计 agent 不需要知道。
+**唯一判据一句**：换个岗位就不成立的才留岗位层；其余一律进通用层。这是本文分流判据第 2 句在多 agent 形态下的具体化。
+**三条配套纪律**：
+1. **出生不建**：新 agent 上岗不预建岗位层文件（空壳无信息），首条岗位特有偏好出现时才立（YAGNI）。
+2. **条目格式照旧**：两层都沿用沉淀区格式（日期 + 来源 + 待验证标），升区 / 整合规则不变。
+3. **整合时上提**：定期整合时检查各岗位层——其实换岗也成立的条目**上提通用层并删岗位层原条**，防同一偏好在 N 个 agent 家复抄漂移（多份拷贝必然有一份过期，读到旧版的 agent 会拿过期偏好干活）。
+> 实践出处：来自一个 16 agent 生产团队的实战收敛（已脱敏）——起初各 agent 家各养一份画像，同一偏好多处复抄、版本互相漂移；后收敛为两层制并把"换岗即不成立"定为唯一判据，漂移即止。单用户单 agent 的读者可忽略本节。
+## 隐私红线
+- 本文件族天然全是 PII——按你手里最敏感的内容对待。
+- "随文件夹迁移"是用户可携带性，不是自动跨项目/跨供应商/跨模型共享授权；加载前确认作用域，只注入当前任务所需部分。
+- `USER.local.md` **必须 gitignore**，永不入库。
+- `USER.md` 推到公开仓库前**必须人工过一遍**：真名、联系方式、雇主内情、凭据指针逐项审。
+- 明文密码 / API key / token **永不写入**本文件族——只写「凭据在哪、怎么调用」的指针。
+- 模板永不预填真实内容——本 skill 自带模板全部为占位符+虚构示例。
+## 简例
+- 用户又一次说「回复先给结论」，沉淀区同型待验证条目已达升区门槛（见 `references/curation.md`）→ 起草升区提议（合并成一句，拟入 How I work），等用户确认后迁移。
+- 用户说「这台机器上用 pnpm 别用 npm」→ 换机器即失效 → 记 `USER.local.md`。
+- 用户在某个仓库里说「这个项目提交信息用西班牙语」→ 换项目不成立 → 记该项目本地文件，`USER.md` 不写。
+- 用户深夜发牢骚「烦死了，都别给我看长回复」→ 情绪时刻 → 按单次观察沉淀、压低置信，不直接改手写区，更不当铁律。
+- 用户手写区写了「我在做 X 项目」但三个月没提过 → agent **建议**更新 What I'm up to 节，附理由；用户点头前一个字不动。
+## 版本与更新自检
+- 当前版本：**1.1.0**（与根 `plugin.json` 一致——本次发行的包清单就是它）。
+- 仓库：`https://github.com/LucioLiu/user-md`
+- 自检：plugin 安装的，宿主更新机制会跟随版本号；手动 clone / 模板拷贝的，diff 本文件与仓库 main 分支即可知是否落后。你自己的 `USER.md` 文件族不受 skill 更新影响——那是你的数据，不是 skill 的一部分。
+---
+> 中英文双版同步维护；如有出入，以**中文版**为准——发现不一致请开 [issue](https://github.com/LucioLiu/user-md/issues)。

SOP.md ADDED Viewed

	@@ -0,0 +1,22 @@

+**English** | [简体中文](SOP.zh-CN.md)
+# user-md · Manual SOP (runnable by a human, no AI required)
+> This is the discipline's **manual operation layer**: building / growing / using the profile file compressed into a card a human can run by hand — you (the user) can maintain the file yourself without an agent, and teams can use it as a "personal profile file maintenance procedure". The rule source remains `SKILL.md`; this card compresses, it does not add rules. Numeric thresholds live in the `references/curation.md` threshold table — never re-copied here.
+**Build (ten minutes)**
+1. Copy both templates from `templates/` to your workspace root: `USER.md` + `USER.local.md`.
+2. Add `USER.local.md` to `.gitignore` (machine-local facts never enter the repo).
+3. Spend five minutes filling the first four sections of `USER.md`: Who I am / How I work / Our relationship / What I'm up to. It runs even empty — but then your agent knows nothing about you.
+**Grow (day to day)**
+4. **Three routing lines**: invalid on another machine → `USER.local.md`; invalid in another project/role → that project's own local file; everything else → `USER.md`. Facts with an existing authoritative source get a pointer line, never a copy.
+5. **Curated-lane append format**: `- 【date · source event】rule text.【single observation, unverified】` — date + source are each entry's birth certificate; write it the moment you spot it, don't batch from memory later.
+6. **Promotion**: only when the same observation recurs past the threshold (see the table) do you merge curated entries into one line and move it into the hand-written lane — **only after you confirm it yourself**; emotional-moment statements get lowered confidence; one-off task demands are never curated.
+7. **Periodic consolidation**: merge synonyms, clear stale entries (cadence per the table); over the size target, sink details into `user/*.md` with an index line left behind; every rewrite leaves one line in the change log at the file's end.
+**Privacy self-check (before the file leaves this machine, every time)**
+8. Review real names / contact info / employer internals / credential pointers item by item; plaintext passwords / API keys **never** enter this file family; a manual pass is mandatory before pushing to any public repo; confirm authorization scope before loading across projects or vendors.
+---
+> Maintained in sync with the SKILL; recheck this card whenever the SKILL version bumps. If they diverge, `SKILL.zh-CN.md` is authoritative (Chinese is this repo's rule source).

SOP.zh-CN.md ADDED Viewed

	@@ -0,0 +1,23 @@

+[English](SOP.md) | **简体中文**
+# user-md · 人工操作 SOP（不用 AI 也能照跑）
+> 这是本纪律的**人工操作层**：把画像文件的建/养/用压成一张人可以手动照跑的操作卡——你（用户本人）不靠 agent 也能自己建档养护；也适合给团队写"个人画像文件维护规程"。规则本体以 `SKILL.zh-CN.md` 为准，本卡只压缩、不另立规则；数字阈值以 `references/curation.md` 阈值表为准，本卡不复抄。
+**建（十分钟）**
+1. 从 `templates/` 拷两份模板到工作区根目录：`USER.md` + `USER.local.md`。
+2. 把 `USER.local.md` 加进 `.gitignore`（机器本地事实永不入库）。
+3. 花五分钟填 `USER.md` 前四节：我是谁 / 我怎么工作 / 协作关系 / 我最近在忙什么。留空也能跑，但 agent 对你零认知。
+**养（日常）**
+4. **分流三句**：换台机器就失效的 → `USER.local.md`；换个项目/岗位就不成立的 → 该项目自己的本地文件；其余 → `USER.md`。已有权威源的事实不复抄、只放指针行。
+5. **沉淀区追加格式**：`- 【日期 · 来源事件】规则正文。【单次观察待验证】`——日期+来源是每条的出生证明；识别到就当场记，别攒批凭记忆补。
+6. **升区**：同型观察重现达门槛（见阈值表）才把沉淀区条目合并成一句迁入手写区——**你自己确认后才迁**；情绪时刻的表达压低置信，一次性任务要求不沉淀。
+7. **定期整合**：同义条目合并、过期条目清理（周期见阈值表）；超体量就把细节下沉 `user/*.md`、主文件留索引行；每次改写在文件尾变更记录留一行。
+**隐私自查（每次把文件带出本机前过一遍）**
+8. 真名 / 联系方式 / 雇主内情 / 凭据指针逐项人工审；明文密码 / API key **永不写入**本文件族；推公开仓前必须人工过一遍；跨项目/跨供应商加载前确认授权范围。
+---
+> 本卡与 SKILL 同步维护，SKILL 升版时本卡跟随核对；如有出入，以 `SKILL.zh-CN.md` 为准。

plugin.json ADDED Viewed

	@@ -0,0 +1,10 @@

+{
+  "name": "user-md",
+  "version": "1.1.0",
+  "description": "Keep your AI's understanding of you in a Markdown file you own - a five-section USER.md with a hand-written lane and an append-only agent-curated lane (dated, sourced, confidence-tagged), machine-local facts split into a gitignored USER.local.md. No silent rewrites; moves when your folder moves. Use when setting up a USER.md or user profile for an agent, recording user preferences or working style, deciding which file a fact about the user belongs in, curating agent-observed habits with confidence tags, onboarding a new agent to an existing user, or carrying personalization across machines and harnesses. 中文触发：用户画像、用户偏好沉淀、记住我的习惯、USER.md 建档养护。",
+  "author": { "name": "LucioLiu", "url": "https://github.com/LucioLiu" },
+  "homepage": "https://github.com/LucioLiu/user-md",
+  "repository": "https://github.com/LucioLiu/user-md",
+  "keywords": ["claude-code", "agent-skills", "user-profile", "personalization", "agents-md", "openclaw"],
+  "license": "PolyForm-Noncommercial-1.0.0"
+}

references/curation.md ADDED Viewed

	@@ -0,0 +1,67 @@

+# 养成细则（curation）
+> USER.md 沉淀区的完整操作规则：识别什么、怎么记、怎么升、怎么瘦身。**全套 skill 的阈值类数字只在本文「阈值表」定义一次**——其他文件（含 SKILL.md、README、模板）一律引用本表，不得另写数字。改阈值 = 改本表一处，全套生效。
+## 阈值表（唯一定义处）
+| 阈值 | 值 | 作用 |
+|---|---|---|
+| 主文件体量目标 | < 150 行 | `USER.md` 全文（含沉淀区）保持每会话全量注入可负担 |
+| 沉淀区下沉线 | 沉淀区 > 40 行，或主文件超体量目标 | 触发细节下沉 `user/*.md`，主文件留索引行 |
+| 升区门槛 | 同型观察 ≥ 2 次，且来自不同会话/事件 | 沉淀条目可发起升手写区提议 |
+| 整合周期 | 每月 1 次 | 沉淀区去重合并与过期清理 |
+| 极短消息过滤线 | 单条消息 < 10 个词 | 过短消息默认不作为偏好信号沉淀 |
+## 信号识别（什么值得沉淀）
+三类可沉淀信号：
+1. **纠正**：用户纠正了 agent 的做法或理解（「不是这样，我要的是…」）。
+2. **重复偏好**：同类选择或要求在不同场合再次出现。
+3. **明确表达**：用户直接说「记住」「以后都…」「我喜欢/别再…」。
+## 误报过滤（什么不沉淀）
+- **一次性语境**：只对当前任务成立的要求（「这份翻译用正式语气」≠「永远正式语气」）。
+- **极短消息**：低于过滤线（见阈值表）的消息默认忽略——信息量不足以支撑长期规则。
+- **情绪时刻**：激烈情绪下的表达，先按单次观察沉淀并压低置信，不直接当规则。
+- **任务强绑定指令**：属于任务规格的内容归任务，不归画像。
+## 条目格式
+每条沉淀 = 出生证明 + 正文 + 置信标：
+```
+- 【YYYY-MM-DD · 来源事件】规则正文。（可选：踩坑背景或用户原话锚）【单次观察待验证】
+```
+规则：
+- 日期与来源事件必填——没有出处的条目不许写。
+- 单次观察必须带【单次观察待验证】；同型观察达升区门槛后（见阈值表），在原条目追加后续日期与来源，标记改为【已重现，可提议升区】。
+- 置信标与日期括号的样式（【】或 [ ]）可随文件主语言取等价形式，但**同一文件内必须统一**——置信标要被检索与机检，两制并存会漂。
+- 多个 agent 共用一份 `USER.md` 时，条目可附观察者署名（如「(observed by coding-agent)」），便于交叉验证。
+- 虚构示例：
+  - 【2026-03-04 · 代码评审会话】Alex 长回复偏好先结论后过程。【单次观察待验证】
+  - 【2026-03-04 · 代码评审会话】【2026-03-19 · 周报草稿会话】Alex 长回复偏好先结论后过程。【已重现，可提议升区】
+## 升区流程（沉淀区 → 手写区）
+1. 同型观察达升区门槛（见阈值表）。
+2. agent 起草升区提议：合并后的一句成文 + 引用原始条目（日期/来源）。
+3. **用户确认后**才写入对应手写节；沉淀区原条目删除，变更记录留一行（日期 · 升了什么 · 来源条目）。
+4. 用户否决：条目保留沉淀区，追加「已提议未采纳 YYYY-MM-DD」，不再重复提议；用户日后改口可再议。
+## 整合去重（定期，周期见阈值表）
+- 同义条目合并：保留最早日期（资历）与最新表述（现行语义）。
+- 过期清理：**沉淀区中**与 What I'm up to 相关的已完成/已放弃**观察条目**，删除或移入变更记录；手写区的过期事项**只可向用户提出清理建议**，不可代删。
+- **整合与升区迁出是重写沉淀区仅有的两个合法场景**，均必须在变更记录留一行（日期 · 合并/清理/升迁了什么）。
+- 手写区在整合中依然只可建议、不可代改。
+## 膨胀与下沉
+- 触发：见阈值表下沉线。
+- 动作：**在你自己的工作区里**（`USER.md` 旁边，不是本 skill 目录里）建一个 `user` 子目录，按主题在里面新建辅文件——例如写作风格就叫 `writing-style.md`。这些是你下沉时才生成的文件，不是本 skill 随包附带的。把细节条目整体移过去，主文件沉淀区留一行索引，例如：「写作风格细则：见 user 目录下的 writing-style」。
+- 辅文件不做每会话注入，按需读取；索引行本身参与后续整合。**辅文件中的沉淀条目沿用沉淀区全部规则**（条目格式/升区/整合留痕）——下沉不是洗白，纪律跟着条目走。
+- 主文件仍超体量目标时，优先建议用户压缩手写区措辞（用户自己动手），不牺牲沉淀纪律。

references/layering.md ADDED Viewed

	@@ -0,0 +1,56 @@

+# 分层判据与宿主关系（layering）
+> 本文回答两个问题：关于用户的一条信息该落进哪个文件；user-md 和你所在宿主已有的记忆机制是什么关系。阈值类数字一律见 `curation.md` 阈值表，本文不重复定义。
+## 一、分流判据三句（弱模型兜底）
+按顺序问，命中即停：
+1. **换台机器就失效吗？**（路径/工具链/硬件/本机登录态）→ `USER.local.md`。机器本地事实是**动态事实**：agent 使用前应验证仍成立（探一下路径/工具还在不在），失效即更新本文件——不把一次成功路径固化成常识。
+2. **换个项目或岗位就不成立吗？**（某仓库的提交规范、某项目专用语言要求）→ 该项目自己的本地上下文文件，`USER.md` 不写。
+3. 都不是 → `USER.md`。
+## 二、归类防混三判据
+分节归类拿不准时：
+1. **按信号的语义作用域分，不按谁观察到分。**agent 观察到的身份事实，升区后进 Who I am——不因为「是 agent 发现的」就永远留在沉淀区。
+2. **动态 ≠ 不落盘。**持久成立的事实进对应节；只有会话级的短暂状态（今天的心情、本次任务的临时约束）留在会话里，不写文件。
+3. **正文是现行语义；历史原文只作审计证据。**变更记录与旧条目不反向覆盖现行条文。
+## 三、状态与偏好为何必须分节
+实践中反复观察到：把「用户当前状态/在做的事」和「用户偏好」混进同一个「偏好」文件时，执行侧（尤其较弱的模型）会按文件名的语义拒绝写入状态类信息——文件叫什么名，弱模型就只往里写什么。把 What I'm up to 与 How I work 拆成两节后，这种执行歧义消失。这是结构性解法，不靠提示词补丁。
+## 四、指针纪律（防双源漂移）
+已有权威源的事实，`USER.md` 不复抄正文，只放指针行。
+- 反例：把某项目 README 里的架构说明抄进 `USER.md` → 项目演进后两处各说各话。
+- 正例：`- 我的开源项目概况：见各仓库 README，勿在本文件复述。`
+- 判据一句：这条信息若在别处有**被维护的原件**，这里只写指针；这里就是原件的，才写正文。
+## 五、与各宿主记忆机制的关系（不抢位）
+user-md 只占**用户画像**这一个轴。它不替代、不覆盖宿主自己的记忆机制；边界按下面分工。
+### OpenClaw workspace
+- `USER.md` 本来就是其用户轴文件，user-md 直接沿用文件名与 every-session 加载惯例，即插即用。
+- agent 轴（SOUL.md / IDENTITY.md）与记忆轴（MEMORY.md 及日记文件）不动——那是 agent 的人格与工作记忆，不是用户画像。
+### Claude Code
+- `CLAUDE.md` 管「怎么干活」（项目与全局指令）；`USER.md` 管「为谁干活」（用户画像）——互补不重叠。接法：CLAUDE.md 里加一行 `@USER.md`。
+- `CLAUDE.local.md` 与 `USER.local.md` 同构（machine-specific + gitignore），分工：前者装项目的本机差异，后者装用户的本机环境。
+- 宿主若有 agent 侧自动记忆，那是 agent 的工作记忆；`USER.md` 是用户侧画像，两者并存、互不覆盖。用户画像类内容出现在工作记忆里时，宜引导沉淀进 `USER.md`（单一真源），工作记忆留指针。
+### 云端记忆产品（自动记忆 + 手写指令类）
+- 此类产品的「手写指令区 + 自动记忆区」与本 skill 的双轨同构；根本区别在数据归属——锁在账号里 vs 躺在你文件夹里。
+- 可共存：把云端记忆里长期成立的条目手工搬进 `USER.md`，让文件成为跨产品的权威层；云端侧继续用，不冲突。
+### 记忆层产品（服务 + 数据库形态）
+- 强在自动召回与规模，弱在人读直编与整包迁移。
+- 可共存：`USER.md` 作人可读、可直编的权威画像层；记忆层作检索加速层。方向纪律：记忆层可从 `USER.md` 初始化；反向写回必须走沉淀区规则（追加+留痕），不得静默改写手写区。

templates/USER.local.md ADDED Viewed

	@@ -0,0 +1,43 @@

+# USER.local.md
+<!--
+  Machine-local layer: anything that stops being true when you switch machines.
+  MUST be gitignored — add `USER.local.md` to .gitignore before your first commit.
+  NEVER store plaintext passwords, API keys, or tokens here — write pointers to where credentials live instead.
+  Facts here are DYNAMIC: verify they still hold before relying on them (probe the path, check the tool), and update this file when they don't.
+  All examples below are fictional — replace everything with your own content.
+-->
+## Machine
+<!-- Example:
+- Framework 13 laptop, Windows 11 + WSL2 (Ubuntu 24.04).
+- 32 GB RAM; heavy builds are fine, local LLMs are not.
+-->
+## Toolchain
+<!-- Example:
+- Node 22 via nvm; pnpm over npm on this machine.
+- Editor: VS Code with vim keybindings.
+-->
+## Paths & layout
+<!-- Example:
+- Projects live under ~/dev/, one folder per repo.
+- Scratch space: ~/dev/_scratch/ — safe to write temp files there.
+-->
+## Credential pointers (never the credentials themselves)
+<!-- Example:
+- GitHub: `gh` CLI is already logged in — use it instead of asking for a token.
+- Cloud credentials: system keychain — never ask me to paste them into chat.
+-->
+---
+## Changelog
+<!-- One line per notable change: date · what changed on this machine. -->

templates/USER.md ADDED Viewed

	@@ -0,0 +1,59 @@

+# USER.md
+<!--
+  Who you are and how your agent should work with you. Loaded every session — keep it small.
+  The first four sections are YOURS: hand-written; the agent may suggest edits and applies one only after your explicit confirmation.
+  The last section belongs to your agent: append-only, dated, confidence-tagged.
+  Size and curation thresholds: see references/curation.md in the user-md skill.
+  All examples below are fictional — replace everything with your own content.
+-->
+## Who I am
+<!-- Identity facts: name, role, background, how to address you. Facts here, preferences below.
+Example:
+- Alex Rivera. Call me Alex.
+- Indie game developer; day job: backend engineer at a mid-size logistics company.
+- Based in Lisbon; works in English and Portuguese.
+-->
+## How I work
+<!-- Preferences and style: communication, deliverable standards, review habits, default authorization boundaries.
+Example:
+- Conclusions first, details on request; anything long starts with a TL;DR.
+- Code deliverables need a runnable example, not just a diff.
+- Refactors are fine without asking; deleting files is not.
+-->
+## Our relationship
+<!-- How you two collaborate: what to ask about vs. just do, how to deliver feedback.
+Example:
+- Disagree openly — never agree just to be nice.
+- Ask before touching anything under deploy/; everything else, act first and report after.
+- When I'm vague, ask one clarifying question, then proceed on your best guess.
+-->
+## What I'm up to
+<!-- Current goals and active work. Explicit, revisable hypotheses — update as evidence changes, clear when done.
+Example:
+- Shipping the v0.3 demo of my roguelike by end of quarter.
+- Learning Rust on weekends; prefer Rust examples where reasonable.
+-->
+## Learned by your agent
+<!-- Agent-curated lane. Append-only. Every entry: [date · source event] + rule + confidence tag.
+     Promotion into the sections above requires repeated observation AND the user's explicit OK.
+     Entry format, promotion rules, and overflow thresholds: references/curation.md.
+Example:
+- [2026-03-04 · code review session] Prefers conclusions before reasoning in long replies. [single observation — unverified]
+-->
+---
+## Changelog
+<!-- One line per notable change: date · what was edited / merged / promoted / sunk to user/*.md. -->

中文用户看这里.md ADDED Viewed

	@@ -0,0 +1,16 @@

+# 中文用户看这里 👋
+**会长大的用户画像** —— 一份你自己拥有的 USER.md：AI 对你的了解（习惯/偏好/环境）存在本地文件里，带日期和出处，随文件夹搬家，不锁在任何平台云端。
+这个仓库的文件名都是英文（技术仓惯例），但中文内容都有，对照着找：
+| 你想要 | 打开这个文件 |
+|---|---|
+| **中文说明书**（这是什么/怎么装/怎么用） | [`README.zh-CN.md`](README.zh-CN.md) |
+| **中文版 skill 规则本体**（给 AI 读的） | [`SKILL.zh-CN.md`](SKILL.zh-CN.md) |
+| **人工操作卡**（不用 AI 也能照跑的手动版） | [`SOP.zh-CN.md`](SOP.zh-CN.md) |
+| 英文说明书 / 英文规则 | `README.md` / `SKILL.md` |
+**最快装法（中文用户）**：把 [`SKILL.zh-CN.md`](SKILL.zh-CN.md) 下载后改名为 `SKILL.md`，放进你 AI 工具的 skills 目录（Claude Code 是 `~/.claude/skills/user-md/SKILL.md`）——细节见中文说明书。
+> 本 skill 属于 [Nuwa 女娲](https://github.com/LucioLiu/nuwa) 数字员工生态的能力积木，可单独用，也可通过整合包 [agent-modpack](https://github.com/LucioLiu/agent-modpack) 一键连团队一起装。