diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index b97e5b2..5d43296 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -1,6 +1,6 @@
 # Contributing to Klear-Team-Brain
 
-Tuser2s for your interest! Issues and pull requests are welcome.
+Thanks for your interest! Issues and pull requests are welcome.
 
 ## How this repo works
 
diff --git a/DEPLOY.md b/DEPLOY.md
index 658073b..40e8b6b 100644
--- a/DEPLOY.md
+++ b/DEPLOY.md
@@ -59,6 +59,8 @@ node server/admin.mjs add alice --name "Alice" --server-url https://brain.yourdo
 
 This appends Alice to `team.yaml` + `tokens.yaml`, restarts the service, and prints a `BRAIN-…` **invite code** plus a ready-to-send onboarding message. That invite code is exactly what `brain join` expects in [section 2](#2-clients-each-dev-machine).
 
+> **From your own machine:** once your laptop's `client.config.yaml` has an `admin: { ssh, dir }` block (pointing at the server's SSH target + code dir — see `client.config.example.yaml`), you can run the same thing locally: `brain admin add alice --name "Alice"`. It SSHes in and runs `server/admin.mjs` for you, so you don't need to log into the box. Same for `brain admin who` / `rm` / `org` / `repo` / `gitlab` / `gitea`.
+
 **Manual alternative.** Create the files by hand — copy `team.example.yaml` → `team.yaml`, and `tokens.example.yaml` → `tokens.yaml` with one `openssl rand -hex 24` per member. Members set up this way use the **manual client path** in §2 (raw token in `client.config.yaml`) — `brain join` takes an invite code, not a raw token.
 
 ### 1.4 Pick a durable, backed-up path for the truth store
@@ -67,6 +69,15 @@ This appends Alice to `team.yaml` + `tokens.yaml`, restarts the service, and pri
 export TRUTH_DIR=/var/lib/team-brain/truth   # the authoritative git truth store — back it up
 ```
 
+The truth store is a plain git repo — it's the whole value of the system and **can't be rebuilt**, so back it up. A simple cron-able snapshot:
+
+```bash
+git -C "$TRUTH_DIR" bundle create /backups/truth-$(date +%F).bundle --all   # single-file, restore with: git clone <bundle>
+# or just: tar czf /backups/truth-$(date +%F).tgz -C "$TRUTH_DIR" .
+```
+
+> On a low-RAM VPS, avoid `git gc --aggressive` on the truth store — it can OOM. Plain `git gc` (or letting git auto-gc) is fine.
+
 ### 1.5 HTTPS (easiest: Caddy auto-certs)
 
 `/etc/caddy/Caddyfile`:
@@ -102,22 +113,64 @@ WantedBy=multi-user.target
 
 ```bash
 sudo systemctl enable --now team-brain
-curl https://brain.yourdomain.com/health   # {"ok":true}
+curl https://brain.yourdomain.com/health   # {"ok":true,"version":"…","truth":{"dir":true,"git":true,"writable":true}}
 ```
 
+> `/health` does a cheap readiness check of the truth store (exists / is a git repo / writable) and returns **HTTP 503** with the failing field if any check fails — point your uptime monitor at it and alert on non-2xx.
+
 > The server binds `127.0.0.1` by default, so port 8787 is **not** reachable on the public IP — only the Caddy/HTTPS front door is. Keep it that way: if your reverse proxy runs on a different host (e.g. Docker), set `Environment=HOST=0.0.0.0` **and** firewall 8787 so only the proxy can reach it. Never expose 8787 to the internet directly (that bypasses TLS → tokens travel in plaintext).
 
 > Raw sessions are stored unredacted on disk (the redacted `.md` is what the query layer surfaces). Security rests on HTTPS + keeping this host open **only to your circle** + token auth. Don't push the truth store to any public remote.
 
-### 1.7 Optional: GitHub code state
+### 1.7 Optional: code state (GitHub / GitLab / Gitea)
+
+Create `registry.yaml` (copy from `registry.example.yaml`, **secret → gitignored**) to register the orgs/repos that should become first-class spaces, each with a read-only PAT. A global `GITHUB_TOKEN` (env or `GITHUB_TOKEN_FILE`) is the fallback. With either configured, `read_github` + the 4h code-state poll are enabled.
+
+Beyond GitHub, **self-hosted GitLab and Gitea** are first-class too — `registry.example.yaml` shows the multi-instance shape (host + base URL + token, with group/project or org/repo entries), and you can maintain them from your own machine with `brain admin gitlab …` / `brain admin gitea …`.
+
+> Code-state only includes branches pushed within the last **30 days**. Long-lived release branches that go quiet drop out of view — raise the window with `Environment=CODESTATE_ACTIVE_DAYS=365` to keep them. If `GITHUB_TOKEN_FILE` is set but unreadable, the server logs a warning at startup and runs as if no token were configured (so `read_github` would otherwise report a misleading "no permission").
+
+### 1.8 Optional: doc mirror (Feishu/Lark · Notion · Google Docs)
+
+Mirror human-written team docs into the truth store (one-way) so the asking agent can `grep`/`read` them alongside sessions and code. Each source is gated by its own secret yaml (leave it out → that layer stays quietly off), and all follow the same "share with the bot, then it mirrors" model:
+
+- **Feishu / Lark** — copy `feishu.example.yaml` → `feishu.yaml`, fill in your custom-app `app_id`/`app_secret`, restart. The full walkthrough (creating the app, scopes, and the **non-obvious whole-wiki authorization step** — you can't add the app directly; you add a *group containing the app's bot* as a wiki admin) is in the README under *Optional: mirror team docs (Lark / Feishu)*.
+- **Notion** — copy `notion.example.yaml` → `notion.yaml`, fill `api_token`, **share** the pages/databases with your integration, restart.
+- **Google Docs** — copy `google.example.yaml` → `google.yaml`, point it at a service-account JSON key, **share** the docs/folders with the service account's email, restart.
+
+All are **secret → gitignored**. Each polls every `poll_hours` (default 4).
+
+### 1.9 Optional: in-dashboard Q&A (`ASK_ENABLED`)
+
+The web dashboard can host a natural-language "ask one question" box that answers over the truth store. It's **off by default** and spawns a server-side `codex` process per question, so it has cost/latency implications and needs `codex` installed on the server. Enable + tune via env:
 
-Create `registry.yaml` (copy from `registry.example.yaml`, **secret → gitignored**) to register the GitHub orgs/repos that should become first-class spaces, each with a read-only PAT. A global `GITHUB_TOKEN` (env or `GITHUB_TOKEN_FILE`) is the fallback. With either configured, `read_github` + the 4h code-state poll are enabled.
+| Var | Default | What it does |
+|---|---|---|
+| `ASK_ENABLED` | (off) | Set `1` to enable the `/ask` endpoint + the dashboard Q&A box. |
+| `ASK_CODEX_BIN` | `codex` | Path to the codex binary the server spawns. |
+| `ASK_CODEX_ARGS` | `exec --skip-git-repo-check` | Args passed to codex. |
+| `ASK_CWD` | `TRUTH_DIR` | Working directory (defaults to the truth store, so codex can `grep`/`read` the `.md` directly). |
+| `ASK_TIMEOUT_MS` | `120000` | Per-question timeout. |
+| `ASK_MAX_CONCURRENT` | `2` | Max concurrent questions before returning `429`. |
 
-### 1.8 Optional: doc mirror (Lark / Feishu)
+### 1.10 Environment variables (reference)
 
-Mirror a Lark/Feishu **wiki** into the truth store (one-way) so the asking agent can `grep`/`read` your team docs alongside sessions and code. Copy `feishu.example.yaml` → `feishu.yaml` (**secret → gitignored**), fill in your custom-app credentials, and restart. Leave it out and the doc layer stays quietly off.
+Everything the server honors (all optional; sensible defaults shown):
 
-The full walkthrough — creating the app, which scopes to enable, and the **non-obvious whole-wiki authorization step** (you can't add the app directly; you add a *group containing the app's bot* as a wiki admin) — is in the README under *Setting up the doc mirror (Lark / Feishu)*.
+| Var | Default | Purpose |
+|---|---|---|
+| `TRUTH_DIR` | `<repo>/truth-server` | The authoritative git truth store. **Set this to a durable, backed-up path** (§1.4). |
+| `PORT` | `8787` | Listen port. |
+| `HOST` | `127.0.0.1` | Bind address. Keep loopback behind a proxy; only set `0.0.0.0` with a firewall (§1.6). |
+| `TEAM_FILE` | `<repo>/team.yaml` | Roster path. |
+| `TOKENS_FILE` | `<repo>/tokens.yaml` | Member tokens (secret). |
+| `REGISTRY_FILE` | `<repo>/registry.yaml` | Registered orgs/repos + PATs (secret). |
+| `GITHUB_TOKEN` / `GITHUB_TOKEN_FILE` | — | Global fallback GitHub token (§1.7). |
+| `CODESTATE_ACTIVE_DAYS` | `30` | Active-branch window for code-state (§1.7). |
+| `FEISHU_FILE` / `NOTION_FILE` / `GOOGLE_FILE` | `<repo>/<name>.yaml` | Doc-source credentials (§1.8). |
+| `NO_POLL` | (off) | `1` disables all background polling (code-state + doc sources) — useful for static/dev. Note: the Docker image defaults this to `1`. |
+| `PUBLIC_URL` | request `Host` | The base URL baked into the `curl …/get \| bash` install script. **Set it when behind a proxy** so members get the right install URL. |
+| `ASK_*` | — | In-dashboard Q&A (§1.9). |
 
 ---
 
@@ -161,9 +214,14 @@ upload_folders:             # the collection allowlist — the field you'll touc
   - /Users/you/Code/team-stuff   # a session uploads only if its cwd is under one of these
 exclude:                    # subdirectories to keep private
   - /Users/you/Code/team-stuff/secret
+# collect_all: true         # ignore upload_folders and collect ALL sessions on this machine
+                            # (what the installer sets when you pick no workspaces — don't use on machines with secrets/client data)
 codex: true                 # also collect ~/.codex/sessions
+session_history_md: true    # also collect session_history/**/*.md under upload_folders
+trae_memory: true           # also collect Trae's native session memory under upload_folders
 interval_sec: 60            # how often to scan, in seconds
 debounce_sec: 60            # seconds a session must be idle before it's "stable" enough to upload
+# auto_update: false        # default on: the resident self-updates from the server daily; set false to pin
 ```
 
 Day to day only two fields change: add a project to `upload_folders` when you pick it up (not in the list = never uploaded), or add a subdir to `exclude` to keep it private — `brain service restart` after either. `token` and `me.id` are your identity: changing `me.id` breaks per-person lookups, a wrong `token` means `401`. The file holds a secret and is gitignored — never commit it.
@@ -175,11 +233,13 @@ Start with 2 people, run a few days, and check: are searches accurate, is the up
 ## Daily ops
 
 ```bash
-brain status              # resident status + last sync
+brain status              # resident status + last sync + collection footprint (uploaded/skipped)
+brain viewer              # open the local footprint console (127.0.0.1): see/exclude/retract what this machine uploads
 brain logs -f             # collector logs
 brain update              # pull latest client from the server + restart resident
 brain service restart     # restart the resident after config/code changes
 brain uninstall           # stop resident + remove MCP + delete token config
+brain admin add|who|rm    # (admins only, needs the admin: block) manage the roster from your machine over SSH
 ```
 
 ## Troubleshooting
diff --git a/README.md b/README.md
index d849e4c..1768de8 100644
--- a/README.md
+++ b/README.md
@@ -95,11 +95,13 @@ With MCP wired up, ask in plain language inside CC / Codex (*"where did the auth
 | `sessions` | Find sessions by person + work time (who did what, in a given window). |
 | `stats` | Aggregate token usage / session counts / turns by day/week/person/repo/tool. |
 | `log` | Activity timeline (the git history; narrow by space/author/since). |
-| `read_github` | Reach out to GitHub for live code state or a file's current contents (code itself isn't stored). |
+| `read_github` | Reach out to GitHub / GitLab / Gitea (incl. self-hosted) for live code state or a file's current contents (code itself isn't stored). |
 
 Server-side queries run via `git grep` / `git ls-files` / `git log` / `fs` — **execFile, no shell, locked inside `TRUTH_DIR`, read-only** — with **zero server-side LLM calls**.
 
 > **Wiring other editors:** the MCP server is a stdio server; the command is always `<node> <install-dir>/mcp/server.mjs` (`brain mcp` prints your exact path). Add that as a stdio MCP server in Claude Code, Codex, or any MCP-capable client (Gemini CLI / Cursor / Cline / opencode…).
+>
+> **Remote / cloud agents (HTTP transport):** an agent that can't run the local stdio binary can mount the memory over HTTP instead — point it at `https://your-server/mcp` with your member token as a `Bearer` header. Same tools, no local install.
 
 ### ② Browse the web UI
 
@@ -123,7 +125,7 @@ each machine (client)                          your server (self-hosted)
 │ ① collector client/sync │  gzip+token │ ② server/server.mjs (HTTP, HTTPS  │
 │   resident, watches AI   │ ──────────▶ │   via a proxy like Caddy)         │
 │   sessions; gated to the │            │   /ingest → git truth store        │
-│   allowlisted circle     │            │   TRUTH_DIR + 4h GitHub poll →     │
+│   allowlisted circle     │            │   TRUTH_DIR + 4h repo poll →       │
 │                          │            │   code-state                       │
 │ query A: ask in editor   │  search+fetch│  /grep /find /read /ls /log       │
 │ query B: open the web UI │ ◀────────── │   + static dashboard hosted at /  │
@@ -135,8 +137,8 @@ The memory is a single **git repository** fusing three sources — **each captur
 | What you want to know | Lives mainly in | How it enters the memory |
 |---|---|---|
 | **Progress · reasoning** | CC/Codex sessions | distilled + redacted, stored as full-text transcript |
-| **Code state** | GitHub | not stored; fetched on demand + 4h poll into `code-state` |
-| **Goals · decisions (human-written)** | Team docs (Lark/Feishu wiki) | one-way mirror of doc bodies; grep/read it, edit at the source |
+| **Code state** | GitHub / GitLab / Gitea (incl. self-hosted) | not stored; fetched on demand + 4h poll into `code-state` |
+| **Goals · decisions (human-written)** | Team docs (Feishu/Lark wiki · Notion · Google Docs) | one-way mirror of doc bodies; grep/read it, edit at the source |
 
 > **Design trade-off:** only the "truth" (raw material + metadata) is kept clean and complete — because it's expensive and can't be rebuilt; "views" (queries / indexes / the dashboard) are swappable and throwaway. Sessions enter **distilled** (inline images stripped, giant tool outputs truncated — signal, not bytes); the byte-exact original stays on the producer's machine (`~/.codex` / `~/.claude`).
 
@@ -145,7 +147,8 @@ The memory is a single **git repository** fusing three sources — **each captur
 ## Privacy & security
 
 - **Scope gate:** a session is uploaded only if its working directory is under your machine's `upload_folders` allowlist — inside the circle is shared by default, outside is private by default.
-- **Redaction:** secrets/tokens + home-directory paths are stripped **client-side before upload**, with second passes when the server projects the `.md` and at the `/read` exit.
+- **Redaction:** secrets/tokens + home-directory paths are stripped **client-side before upload**, with second passes when the server projects the `.md` and at the `/read` exit. Each producer can also keep a **personal redaction wordlist** (client names, code names…) that's stripped before anything leaves their machine.
+- **Producer transparency & control:** run `brain viewer` for a local console (127.0.0.1, only you can see it) that shows, per session, exactly what was uploaded vs skipped — and lets you exclude individual sessions, **retract** ones already in the shared store, or add personal redaction terms.
 - **Credentials never stored:** member tokens, GitHub PATs, and doc-source credentials all live server-side and are gitignored; the roster (no secrets) can be committed.
 - **The memory is the whole value — never push it to a public remote, and back it up.** Self-host on infrastructure open only to your circle.
 
@@ -166,7 +169,7 @@ If your team keeps human-written docs (goals, decisions, notes) in a Lark/Feishu
 
    The app then has read access to the *entire* wiki (propagation takes ~1–2 minutes).
 4. **Drop credentials on the server:** copy `feishu.example.yaml` → `feishu.yaml` (secret → gitignored), fill in `app_id` / `app_secret`, and restart. Leave it out entirely and the doc layer stays quietly off.
-5. **Verify:** after a poll cycle, the docs appear under `feishu/<wiki>/…` in the memory, searchable via `grep`.
+5. **Verify:** after a poll cycle, the docs appear under `feishu/<wiki>__<id>/…` in the memory, searchable via `grep`.
 
 > China Feishu (`open.feishu.cn`) and international Lark (`open.larksuite.com`) are isolated platforms — create the app on whichever one your team uses. Create a **new** wiki later? Repeat step 3 for it, or the brain won't see it.
 
diff --git a/README.zh-CN.md b/README.zh-CN.md
index 2a4b053..53a5eff 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -96,11 +96,13 @@ npm run sync -- --once    # 收一次（或 `npm run sync` 让它在后台持续
 | `sessions`    | 按人 + 工作时间检索 session（某人某段时间干了啥）。                          |
 | `stats`       | 按 天/周/人/repo/工具 聚合 token 用量 / session 数 / 对话轮次。           |
 | `log`         | 活动时间线（git 历史；可按 space/author/since 收窄）。                   |
-| `read_github` | 出网现拉某仓代码状态或文件最新内容（代码本体不入库）。                               |
+| `read_github` | 出网现拉 GitHub／GitLab／Gitea（含自建）某仓代码状态或文件最新内容（代码本体不入库）。       |
 
 服务器侧查询全走 `git grep` ／ `git ls-files` ／ `git log` ／ `fs`——**execFile 无 shell、锁死在 `TRUTH_DIR` 内、只读**——**零服务器端 LLM 调用**。
 
 > **接别的编辑器：** MCP 是个 stdio server，命令固定为 `<node> <安装目录>/mcp/server.mjs`（`brain mcp` 会打印你这台的实际路径）。任何支持 MCP 的客户端（Claude Code ／ Codex ／ Gemini CLI ／ Cursor ／ Cline ／ opencode…）加一个 stdio MCP server 即可。
+>
+> **远程／云端 agent（HTTP 传输）：** 跑不了本地 stdio 二进制的 agent，可改用 HTTP 挂载——指向 `https://你的服务器/mcp`，把个人 token 作为 `Bearer` 头带上即可。工具一样，无需本地安装。
 
 ### ② Web GUI 浏览
 
@@ -124,7 +126,7 @@ npm run sync -- --once    # 收一次（或 `npm run sync` 让它在后台持续
 │ ① 采集器 client/sync   │  gzip+token│ ② server/server.mjs(HTTP,前面    │
 │   常驻,盯 AI session    │ ─────────▶ │   可套 Caddy 等做 HTTPS)          │
 │   按白名单闸门只传圈内    │            │   /ingest → git 真相库 TRUTH_DIR   │
-│                        │            │   + 每 4h 拉 GitHub 出 code-state  │
+│                        │            │   + 每 4h 拉代码仓出 code-state    │
 │ 查询 A: 编辑器里问(MCP)  │  搜+拉,综合  │   /grep /find /read /ls /log       │
 │ 查询 B: 浏览器开 Web GUI │ ◀──────────│   + 在 / 托管纯静态看板            │
 └────────────────────────┘            └──────────────────────────────────┘
@@ -135,8 +137,8 @@ npm run sync -- --once    # 收一次（或 `npm run sync` 让它在后台持续
 | 你想知道            | 主要长在               | 怎么进记忆库                           |
 | --------------- | ------------------ | -------------------------------- |
 | **进展 · 思考过程**   | CC/Codex session   | 蒸馏 + 脱敏后存全文 transcript           |
-| **代码现状**        | GitHub             | 不存本体，查询时现拉 + 4h 轮询出 `code-state` |
-| **目标 · 决策（人写）** | 团队文档（Lark/飞书 wiki） | 正文单向镜像，可搜可读；改去源文档                |
+| **代码现状**        | GitHub／GitLab／Gitea（含自建） | 不存本体，查询时现拉 + 4h 轮询出 `code-state` |
+| **目标 · 决策（人写）** | 团队文档（飞书/Lark wiki · Notion · Google Docs） | 正文单向镜像，可搜可读；改去源文档                |
 
 > **设计取舍：** 只有“真相”（原料 + 元数据）被认真存、保持干净——因为它贵且不可重建；“视图”（查询／索引／看板）随时可换可丢。session 以**蒸馏**形态入库（剥内联图片、截巨型 tool 输出，存信号不存字节），字节精确的原文留在产出者本机（`~/.codex` ／ `~/.claude`）。
 
@@ -145,7 +147,8 @@ npm run sync -- --once    # 收一次（或 `npm run sync` 让它在后台持续
 ## 隐私与安全
 
 - **范围闸门：** 只有 session 的 cwd 在你本机 `upload_folders` 白名单内才上传——圈内默认共享、圈外默认私有。
-- **脱敏：** 上传前在客户端就抹掉密钥/token + 家目录路径，服务器投影 `.md` 和 `/read` 出口再各兜底一次。
+- **脱敏：** 上传前在客户端就抹掉密钥/token + 家目录路径，服务器投影 `.md` 和 `/read` 出口再各兜底一次。每个生产者还能维护一份**个人脱敏词表**（客户名、代号…），上传前先抹掉。
+- **生产者透明可控：** 跑 `brain viewer` 打开本机控制台（127.0.0.1，仅你可见），逐条看每个 session 到底传了什么、跳过了什么——可逐条排除、把已进共享库的**撤回**、或加个人脱敏词。
 - **凭证不入库：** 成员 token、GitHub PAT、文档源凭证都住服务器、都 gitignore；花名册（不含密钥）可提交。
 - **记忆库就是全部价值——别 push 到任何公开 remote，并定期备份。** 自托管在只对圈内开放的基础设施上。
 
@@ -171,7 +174,7 @@ npm run sync -- --once    # 收一次（或 `npm run sync` 让它在后台持续
 
 4. **凭证放服务器：** 把 `feishu.example.yaml` 复制成 `feishu.yaml`（含密钥 → 已 gitignore），填好 `app_id` ／ `app_secret`，重启服务。不配则文档层安静关闭。
 
-5. **验证：** 一个轮询周期后，文档会出现在记忆库 `feishu/<知识库>/…` 下，可用 `grep` 搜到。
+5. **验证：** 一个轮询周期后，文档会出现在记忆库 `feishu/<知识库>__<id>/…` 下，可用 `grep` 搜到。
 
 > 国内飞书（`open.feishu.cn`）与国际版 Lark（`open.larksuite.com`）数据隔离——在你团队实际所在的平台建应用。以后每**新建**一个知识库，都要对它重复第 3 步，否则大脑看不见。
 
diff --git a/SECURITY.md b/SECURITY.md
index ccfafa2..a1d17dd 100644
--- a/SECURITY.md
+++ b/SECURITY.md
@@ -9,7 +9,7 @@ Klear-Team-Brain handles sensitive data: it aggregates redacted full-text transc
 Instead, report privately via one of:
 
 - GitHub's [private vulnerability reporting](https://github.com/Asklear/Klear-Team-Brain/security/advisories/new) (preferred), or
-- email the maintainers at **neiyouser2@gmail.com**.
+- email the maintainers at **neiyohank@gmail.com**.
 
 Please include: a description of the issue, steps to reproduce or a proof of concept, affected versions/components, and any suggested mitigation.
 
diff --git a/cli/brain.mjs b/cli/brain.mjs
index ba88faa..37a17ca 100644
--- a/cli/brain.mjs
+++ b/cli/brain.mjs
@@ -121,7 +121,9 @@ function writeConfig({ server, token, id, name, folders = [], consumer = false,
       ? `# 未指定工作空间 → 采集本机所有 session（含全部项目）。要收窄改 upload_folders。\ncollect_all: true\nupload_folders: []\n`
       : `upload_folders:\n${folders.map((f) => `  - ${f}`).join("\n")}\n`) +
     `exclude: []\n\n` +
-    `docs: true\ncodex: true\nagentdocs: true\ninterval_sec: 60\ndebounce_sec: 60\n`);
+    // 采集开关：codex / session_history / Trae 原生记忆默认都开（与 client.config.example.yaml 对齐）。
+    // 别再写 docs/agentdocs —— M3 起这两个键没有任何代码读，是历史残留。
+    `codex: true\nsession_history_md: true\ntrae_memory: true\ninterval_sec: 60\ndebounce_sec: 60\n`);
 }
 
 // 从本机 CC session 推测"你常在哪些 git 仓干活"，作为 upload_folders 建议
@@ -256,7 +258,8 @@ async function joinCmd(code) {
   const consume = !!inv.consumer || process.argv.includes("--consume-only");
 
   // 1. 可达 + 鉴权（分清三种失败：连不上服务器 / token 无效 / 网络抖动 —— 别一律说成「邀请码过期」误导排查）
-  const reach = await fetch(srv + "/health").then((r) => r.ok).catch(() => false);
+  // 任意 HTTP 响应都算「连得上」（含 /health 因真相库降级回的 503）—— 只有网络层抛错才是真不可达。
+  const reach = await fetch(srv + "/health").then(() => true).catch(() => false);
   if (!reach) die(`连不上 ${srv}（服务器没起 / 网络 / 代理？先确认地址能在浏览器打开）`);
   let who = null, whoErr = "";
   try {
@@ -299,7 +302,7 @@ async function joinCmd(code) {
   writeConfig({ server: srv, token, id, name, folders, consumer: consume, collectAll });
   console.log(c.ok(`✓ 配置写好 ${CFG}`));
   const mcpAttached = addMcp();
-  console.log(c.dim("→ 首次回填历史 session（跨境可能要一会，别关）…"));
+  console.log(c.dim("→ 首次回填历史 session（跨境可能要几分钟，别关）。下面会滚动日志，结束有一句「本机足迹」小结…"));
   spawnSync(NODE, [join(ROOT, "client", "sync.mjs"), "--once"], { cwd: ROOT, stdio: "inherit" });
   serviceInstall({ soft: true });                       // 装常驻失败不致命：配置/MCP/回填都已成 —— 再起服务（sync 启动时已能读到完整 config）
   if (mcpAttached.length)
@@ -463,18 +466,31 @@ function status() {
   }
   let running = false;
   if (IS_MAC) {
+    // print 退出码先分清「没装/没加载」与「装了」：失败=没加载（别再被正则兜成误导的状态字）。
     const r = sh("launchctl", ["print", `${gui()}/${LABEL}`]);
-    const st = (r.stdout.match(/state = (\w+)/) || [, "未装"])[1];
-    const pid = (r.stdout.match(/pid = (\d+)/) || [, "-"])[1];
-    running = st === "running";
-    console.log(`常驻:    ${running ? c.ok(`running (pid ${pid})`) : c.warn(st)}`);
+    if (r.status !== 0) {
+      console.log(`常驻:    ${c.warn("未装 / 未加载")}`);
+    } else {
+      const st = (r.stdout.match(/state = (\S+)/) || [, "unknown"])[1];
+      const pid = (r.stdout.match(/pid = (\d+)/) || [, null])[1];
+      running = st === "running";
+      console.log(`常驻:    ${running ? c.ok(`running (pid ${pid})`) : c.warn(`已加载但没在跑（state=${st}）—— 看 brain logs 找原因`)}`);
+    }
   } else {
     const a = sh("systemctl", ["--user", "is-active", "team-brain-sync"]).stdout.trim();
     running = a === "active";
-    console.log(`常驻:    ${running ? c.ok(a) : c.warn(a || "未装")}`);
+    console.log(`常驻:    ${running ? c.ok(a) : c.warn(a || "未装 / 未加载")}`);
   }
+  // 采集足迹（从结果账本）：让用户一眼看到「传了多少 / 跳过多少」，不必去 viewer
+  try {
+    const led = JSON.parse(readFileSync(join(ROOT, ".brain-ledger.json"), "utf8"));
+    const ss = led.sessions || [];
+    const up = ss.filter((s) => s.status === "uploaded").length;
+    const sk = ss.filter((s) => s.status === "skipped").length;
+    if (up || sk) console.log(`采集:    ${c.ok(`${up} 已传`)}${sk ? c.dim(` · ${sk} 跳过（brain viewer 看逐条原因）`) : ""}`);
+  } catch {}
   if (existsSync(LOG)) {
-    // 取最近一条 tick（logfmt：`<ts> INFO  tick cc_up=… …`），只显示时间 + 计数体，别再匹配老格式的「·」
+    // 一次读，既取最近一条 tick，也扫尾巴有没有失败/报错 —— 别让「同步坏了」时 status 还显得一切正常
     const lines = readFileSync(LOG, "utf8").trimEnd().split("\n");
     const last = [...lines].reverse().find((l) => / tick /.test(l));
     if (last) {
@@ -482,6 +498,8 @@ function status() {
       const body = last.split(" tick ")[1] || last;
       console.log(`最近同步: ${c.dim((ts.slice(11, 19) + " " + body).trim())}`);
     }
+    const fails = lines.slice(-300).filter((l) => /上传失败|tick 异常|出错|ERROR/.test(l)).length;
+    if (fails) console.log(c.warn(`⚠ 最近日志里有 ${fails} 条失败/报错 —— brain logs -f 看详情`));
   }
   try { const vi = JSON.parse(readFileSync(join(ROOT, ".brain-viewer.json"), "utf8")); if (vi.url) console.log(`查看器:  ${c.ok(vi.url)} ${c.dim("(brain viewer 打开)")}`); } catch {}
   if (!running && existsSync(CFG)) console.log(c.dim("\n起常驻：brain service install"));
diff --git a/client.config.example.yaml b/client.config.example.yaml
index 493537c..9cee487 100644
--- a/client.config.example.yaml
+++ b/client.config.example.yaml
@@ -23,6 +23,7 @@ trae_memory: true   # 是否采集 Trae 原生 ~/.trae-cn/memory/projects/<工
 #   - -Users-you-Documents-Code-your-team-stuff  # 也可直接填 slug；必须能对应某个 upload_folders
 interval_sec: 60    # 多久扫一次
 debounce_sec: 60    # session 空闲多少秒才算"稳定"、才上传
+# auto_update: false  # 默认开：常驻每天自检 /version，比本机高就拉新客户端代码并以新代码重启。设 false 关掉自动更新。
 # 注：M3 起只采集 session。文档迁飞书（不再有 docs/agentdocs/vault 配置）；
 #     space 身份由服务器按 registry 定（github__owner__repo 命中登记的 / 否则 local__<id> + folder 标签）。
 
diff --git a/docs/CHANGELOG.en.md b/docs/CHANGELOG.en.md
index a8ba9ba..25b8a16 100644
--- a/docs/CHANGELOG.en.md
+++ b/docs/CHANGELOG.en.md
@@ -4,10 +4,42 @@ English | [中文](./CHANGELOG.zh-CN.md)
 
 This project follows [Semantic Versioning](https://semver.org/), formatted after [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). Check your local version with `brain version`.
 
-> Still pre-1.0 and early, incrementing by patch. Formal versioning began at `v0.1.11`; earlier versions are a retroactive split of the development history by milestone.
+> Still pre-1.0 and early, incrementing by patch. Formal versioning began at `v0.1.11`; earlier versions are a retroactive split of the development history by milestone. `v0.1.12`–`v0.1.13` were internal-only and never released publicly.
 
 ---
 
+## [0.1.22] - 2026-06-25 · Hardened read-only queries + observable status
+
+### Added
+- **`brain status` shows your capture footprint and warns on failures**: it reads the result ledger to show what's been uploaded vs skipped, and scans the log tail for errors, so a broken sync no longer looks healthy. It also tells "not installed / not loaded" apart from "loaded but not running" via launchctl exit codes instead of guessing from a regex.
+- **`/health` is now a real readiness check**: verifies the truth store exists, is a git repo, and is writable; degrades to `503` + version instead of a blind `200`.
+- Configurable active-branch window for code-state via `CODESTATE_ACTIVE_DAYS` (default 30).
+
+### Changed
+- `brain join` first-run backfill now streams a rolling log and ends with a local-footprint summary.
+- Overview / repo / code-state wording for un-pushed progress is now the neutral "activity since last push".
+
+### Fixed
+- **Invalid regex in `grep` returns a clear error** instead of being swallowed as "no matches".
+- **`GITHUB_TOKEN_FILE` set but unreadable now warns at startup** instead of silently running token-less.
+- **Large-file guard on `/read`**: oversized full reads are refused; reads without a limit default to a 5000-line truncation with a pagination hint.
+- `brain join` reachability probe decoupled from `/health` `503`: any HTTP response now counts as "reachable".
+
+## [0.1.21] - 2026-06-25 · Footprint viewer polish + English activity log
+
+### Fixed
+- Local footprint viewer: sidebar footer is now pinned full-height (previously only on the overview page), and the Activity log is fully in English.
+
+## [0.1.20] - 2026-06-25 · Login-free viewer + device token
+
+### Changed
+- **The local footprint viewer no longer asks for a token to log in** (accessing your own machine from your own machine needs no password); it shows your device token directly instead, and the hosted console gained a "Local console" entry that jumps straight to your on-machine viewer.
+
+## [0.1.19] - 2026-06-25 · Producer footprint viewer
+
+### Added
+- **Producer footprint viewer**: every producer can open a local web page to see exactly which of their sessions were captured and what was uploaded, exclude or retract individual items already in the memory, and maintain a personal redaction wordlist (custom sensitive terms masked before upload) — self-service transparency over what leaves your machine. The installer now bundles `web/viewer.*` (which previously 404'd on real installs), with zero new client dependencies.
+
 ## [0.1.18] - 2026-06-25 · Mount the memory from any agent (HTTP-transport MCP)
 
 ### Added
diff --git a/docs/CHANGELOG.zh-CN.md b/docs/CHANGELOG.zh-CN.md
index 9571424..59ad4dd 100644
--- a/docs/CHANGELOG.zh-CN.md
+++ b/docs/CHANGELOG.zh-CN.md
@@ -4,10 +4,42 @@
 
 本项目版本号遵循[语义化版本](https://semver.org/lang/zh-CN/)，格式参考 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.1.0/)。本机版本可用 `brain version` 查看。
 
-> 项目仍处于 1.0 前的早期阶段，按 patch 递增。版本管理在 `v0.1.11` 正式引入，更早的版本是按里程碑对历史开发的回溯切分。
+> 项目仍处于 1.0 前的早期阶段，按 patch 递增。版本管理在 `v0.1.11` 正式引入，更早的版本是按里程碑对历史开发的回溯切分。`v0.1.12`–`v0.1.13` 仅内部使用，未公开发布。
 
 ---
 
+## [0.1.22] - 2026-06-25 · 只读查询加固 + status 看得见采集足迹
+
+### 新增
+- **`brain status` 显示采集足迹、同步坏了会告警**：从结果账本展示「已传/跳过」足迹，并扫日志尾巴对失败/报错告警，同步坏了不再显得一切正常；用 launchctl 退出码区分「未装/未加载」与「已加载但没跑」，不再被正则兜成误导状态。
+- **`/health` 改为真·就绪检查**：校验真相库存在、是 git 仓、可写；不就绪降级回 `503` + version，不再盲目 `200`。
+- code-state 活跃分支窗口可配 `CODESTATE_ACTIVE_DAYS`（默认 30）。
+
+### 改进
+- `brain join` 首次回填改为滚动日志、结尾给「本机足迹」小结。
+- 概览 / 仓库 / code-state 的「未推进度」措辞改为中性的「push 后有活动」。
+
+### 修复
+- **`grep` 非法正则回清晰报错**，不再吞成「无命中」。
+- **`GITHUB_TOKEN_FILE` 配了却读不到 → 启动告警**，不再静默以 token-less 跑。
+- **`/read` 大文件护栏**：超大整读拒绝；无 limit 时默认 5000 行截断带分页提示。
+- `brain join` 可达探测与 `/health` `503` 解耦：任意 HTTP 响应即「连得上」。
+
+## [0.1.21] - 2026-06-25 · 查看器打磨 + 日志英文化
+
+### 修复
+- 本机足迹查看器：侧边栏页脚固定满高（不再只在 overview 页显示），Activity 日志全部英文化。
+
+## [0.1.20] - 2026-06-25 · 查看器免登录 + 设备 token
+
+### 改进
+- **本机足迹查看器不再要求 token 登录**（本机访问本机无需口令），改为直接展示设备 token；线上控制台新增「本机控制台」入口，一键跳到本机查看器。
+
+## [0.1.19] - 2026-06-25 · 生产者本机足迹查看器
+
+### 新增
+- **生产者本机足迹查看器**：每个产出者都能打开本机网页，查看自己究竟有哪些 session 被采集、上传了哪些足迹，逐条排除或撤回已进记忆库的内容，并维护个人脱敏词表（自定义敏感词在上传前抹掉）——对「什么离开了你的机器」做自助式透明管控。装机包现已补带 `web/viewer.*`（此前真实安装会 404），且客户端零新增依赖。
+
 ## [0.1.18] - 2026-06-25 · 任意 agent 远程挂载（HTTP 传输 MCP）
 
 ### 新增
diff --git a/package.json b/package.json
index 286edf4..02874ff 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "klear-team-brain",
-  "version": "0.1.21",
+  "version": "0.1.22",
   "type": "module",
   "description": "Self-hosted team memory for AI coding agents — fuses Claude Code/Codex sessions, GitHub code state, and team docs into one git truth store, queryable from your editor over MCP.",
   "license": "Apache-2.0",
diff --git a/registry.example.yaml b/registry.example.yaml
index 3b5a8de..bb0b728 100644
--- a/registry.example.yaml
+++ b/registry.example.yaml
@@ -1,4 +1,4 @@
-# registry —— 登记的代码托管 org/repo + 各自的 PAT/token，决定"有哪些团队 space"（REDESIGN §3.4）。
+# registry —— 登记的代码托管 org/repo + 各自的 PAT/token，决定"有哪些团队 space"（space 模型见 README）。
 # 支持三家：github（顶层，单实例 github.com）/ gitlab / gitea（自建可多实例，见下）。
 # 复制成 registry.yaml 填真实值。registry.yaml 含 PAT → 已 gitignore，别提交（住服务器，跟 tokens.yaml 一样）。
 # 服务器启动加载、改完 systemctl restart 生效。可用 `brain admin org/repo add ... --pat <PAT>` 维护。
diff --git a/server/codestate.mjs b/server/codestate.mjs
index 7b001e8..794c6e6 100644
--- a/server/codestate.mjs
+++ b/server/codestate.mjs
@@ -12,6 +12,10 @@ import { writeSpaceMeta, readSpaceMeta } from "./space.mjs";
 const REPO_PREFIXES = ["github__", "gitlab__", "gitea__"];
 const isRepoKey = (k) => REPO_PREFIXES.some((p) => k.startsWith(p));
 
+// 「活跃分支」窗口：最后 push 超过这么多天的分支不进 code-state。默认 30 天，长命名 release 分支会被滤掉，
+// 用 CODESTATE_ACTIVE_DAYS 调大（如 365）即可纳入。
+const ACTIVE_DAYS = Number(process.env.CODESTATE_ACTIVE_DAYS) || 30;
+
 // 读不到该仓时，写一张显式告警 code-state，让管理员在 read_github / 搜索里一眼看到该补 token/权限。
 function renderNoAccess(spaceKey, provider, ref, status) {
   const why = status === 404 ? "仓库不存在，或该 token 无权访问它（私有仓未纳入授权）"
@@ -52,7 +56,7 @@ function render(spaceKey, ref, rows, pulls, prNote = "") {
   const lines = [`# code-state · ${spaceKey}`, `repo: ${ref}`, ``, `## 活跃分支`];
   for (const r of rows) {
     lines.push(`- **${r.name}** — \`${r.sha}\` ${r.msg}　(${(r.when || "").slice(0, 16)})` +
-      (r.leads ? `　⚠️ **有未推进度**（session ${r.sess.slice(0, 16)} 晚于最后 push，见 sessions/`
+      (r.leads ? `　· **push 后有活动**（session ${r.sess.slice(0, 16)} 晚于最后 push，见 sessions/`
         + `${r.name.replace(/\//g, "-")}/）` : ""));
   }
   lines.push(``, `## Open PR/MR`);
@@ -64,7 +68,7 @@ function render(spaceKey, ref, rows, pulls, prNote = "") {
 }
 
 // 刷新单个团队仓 space：从 space.yaml 取 provider/owner/repo + registry 取 token/baseUrl，按 provider 分发客户端。
-export async function refreshSpace(truthDir, spaceKey, registry, fallbackToken, { activeDays = 30 } = {}) {
+export async function refreshSpace(truthDir, spaceKey, registry, fallbackToken, { activeDays = ACTIVE_DAYS } = {}) {
   const spaceDir = join(truthDir, "spaces", spaceKey);
   const meta = readSpaceMeta(truthDir, spaceKey);
   const { provider, owner, repo, ref } = meta;
diff --git a/server/query.mjs b/server/query.mjs
index 0da9093..435793e 100644
--- a/server/query.mjs
+++ b/server/query.mjs
@@ -26,7 +26,10 @@ async function runGrep(cmd, args, strip) {
     return { matches: lines.slice(0, MAX_LINES).join("\n").trimEnd(), truncated: lines.length > MAX_LINES };
   } catch (e) {
     if (e.code === 1) return { matches: "", truncated: false };   // 退出码 1 = 无命中（非错误）
-    throw e;
+    // 退出码 ≥2 = grep 自己报错（最常见：正则非法，如未配对括号/悬挂反斜杠）。别把它吞成「无命中」
+    // 也别让 raw execa 错冒成 500 —— 把 stderr 首行清洗后抛出，端点据此回 400，agent 才知道是正则写错了。
+    const detail = (e.stderr || e.message || "").split("\n").map((l) => l.trim()).find(Boolean) || "grep failed";
+    throw new Error(`grep 失败（多半是正则非法，检查模式）：${detail.replace(/^fatal:\s*/i, "")}`);
   }
 }
 
diff --git a/server/server.mjs b/server/server.mjs
index a1e3b96..212e7b9 100644
--- a/server/server.mjs
+++ b/server/server.mjs
@@ -4,7 +4,7 @@
 import http from "node:http";
 import { fileURLToPath } from "node:url";
 import { dirname, join, extname } from "node:path";
-import { readFileSync, existsSync, createReadStream, statSync, writeFileSync, mkdtempSync, rmSync } from "node:fs";
+import { readFileSync, existsSync, createReadStream, statSync, writeFileSync, mkdtempSync, rmSync, accessSync, constants } from "node:fs";
 import { tmpdir } from "node:os";
 import { execFileSync, execFile } from "node:child_process";
 import { promisify } from "node:util";
@@ -39,6 +39,8 @@ const PORT = Number(process.env.PORT) || 8787;
 // 容器/反代在别的网段时用 HOST=0.0.0.0 显式放开（此时务必靠防火墙/安全组封住该端口）。
 const HOST = process.env.HOST || "127.0.0.1";
 const MAX_BODY = 64 * 1024 * 1024;
+const READ_HARD_MAX = 24 * 1024 * 1024;   // /read 整读护栏：超此值又没带 offset/limit 就拒，免得把巨文件整个读进内存
+const READ_MAX_LINES = 5000;              // /read 没带 limit 时的默认行上限：超了截断并告知总行数，免撑爆 agent 上下文
 // 安全响应头：挂在所有响应上。CSP 锁死外部加载面（脚本只许自身、无 inline）——给渲染不可信内容的 GUI 兜底。
 // style 仍放行 inline（GUI 大量用 style=""）；img 放行 data:（favicon 是内联 svg）。
 const SEC_HEADERS = {
@@ -61,9 +63,15 @@ const WEB_TYPES = {
 const roster = loadRoster(process.env.TEAM_FILE || join(ROOT, "team.yaml"));  // 默认 ROOT/team.yaml；docker 用 TEAM_FILE 指到持久卷
 const tokens = tokenIndex(roster, loadTokens(process.env.TOKENS_FILE || join(ROOT, "tokens.yaml")));
 const registry = loadRegistry(process.env.REGISTRY_FILE || join(ROOT, "registry.yaml"));  // 登记的 github org/repo（启动加载、restart 生效）
-const GITHUB_TOKEN = process.env.GITHUB_TOKEN
-  || (process.env.GITHUB_TOKEN_FILE && existsSync(process.env.GITHUB_TOKEN_FILE)
-      ? readFileSync(process.env.GITHUB_TOKEN_FILE, "utf8").trim() : "");
+// GITHUB_TOKEN：直给优先；否则读 GITHUB_TOKEN_FILE。文件配了却读不到不能静默吞成空——
+// 否则后面 read_github / code-state 会以「无 token」跑、报「无权限」误导排查，启动就警一声。
+const GITHUB_TOKEN = (() => {
+  if (process.env.GITHUB_TOKEN) return process.env.GITHUB_TOKEN;
+  const f = process.env.GITHUB_TOKEN_FILE;
+  if (!f) return "";
+  try { return readFileSync(f, "utf8").trim(); }
+  catch (e) { log.warn("GITHUB_TOKEN_FILE 配了但读不到 → 将以「无 GitHub token」运行（read_github/code-state 受影响）", { file: f, err: e.message }); return ""; }
+})();
 const FEISHU = loadFeishu(process.env.FEISHU_FILE || join(ROOT, "feishu.yaml"));  // 文档层（飞书）凭证：缺则不启用
 const NOTION = loadNotion(process.env.NOTION_FILE || join(ROOT, "notion.yaml"));  // 文档层（Notion）凭证：缺则不启用
 const GOOGLE = loadGoogle(process.env.GOOGLE_FILE || join(ROOT, "google.yaml"));  // 文档层（Google Docs）凭证：缺则不启用
@@ -258,7 +266,16 @@ const server = http.createServer((req, res) => {
 
 async function handle(req, res, u) {
   // --- 健康检查 ---
-  if (req.method === "GET" && u.pathname === "/health") return json(res, 200, { ok: true });
+  // 深一点：光回 200 没意义——真相库不存在 / 不是 git 仓 / 不可写，都该让监控（和 brain join 的可达探测）看见。
+  // 全是 statSync/accessSync 级廉价检查，不扫库，安全频繁打。
+  if (req.method === "GET" && u.pathname === "/health") {
+    const truth = { dir: false, git: false, writable: false };
+    try { truth.dir = existsSync(TRUTH) && statSync(TRUTH).isDirectory(); } catch {}
+    try { truth.git = existsSync(join(TRUTH, ".git")); } catch {}
+    try { accessSync(TRUTH, constants.W_OK); truth.writable = true; } catch {}
+    const ok = truth.dir && truth.git && truth.writable;
+    return json(res, ok ? 200 : 503, { ok, version: CLIENT_VERSION, truth });
+  }
 
   // --- 最新客户端版本（无需鉴权，版本号非机密）：采集器每天自检，比本机高就自动更新 ---
   if (req.method === "GET" && u.pathname === "/version") return json(res, 200, { version: CLIENT_VERSION });
@@ -355,11 +372,19 @@ async function handle(req, res, u) {
     let abs;
     try { abs = safeRelPath(TRUTH, path, "path"); } catch { return json(res, 400, { error: "bad path" }); }
     if (!existsSync(abs)) return json(res, 404, { error: "not found" });
-    if (statSync(abs).isDirectory()) return json(res, 400, { error: "是目录，请用 ls" });
-    let text = redactReadable(readFileSync(abs, "utf8"));
+    const st = statSync(abs);
+    if (st.isDirectory()) return json(res, 400, { error: "是目录，请用 ls" });
     const offset = Math.max(0, Number(u.searchParams.get("offset")) || 0);
     const limit = Number(u.searchParams.get("limit")) || 0;
-    if (offset || limit) text = text.split("\n").slice(offset, limit ? offset + limit : undefined).join("\n");
+    // 大文件护栏：没带 offset/limit 又超硬上限 → 别把整文件读进内存，回提示让 agent 分页或先 grep/find 定位。
+    if (!offset && !limit && st.size > READ_HARD_MAX)
+      return json(res, 413, { error: `文件太大（${(st.size / 1048576).toFixed(1)}MB）：带 offset/limit 分页读，或先用 grep/find 定位再 read 具体段。`, size: st.size });
+    let text = redactReadable(readFileSync(abs, "utf8"));
+    if (offset || limit) { text = text.split("\n").slice(offset, limit ? offset + limit : undefined).join("\n"); return json(res, 200, { path, text }); }
+    // 没带 limit 但行数巨多 → 默认截断并告知总行数，免得一次回几万行撑爆 agent 上下文（带 offset 接着读）。
+    const lines = text.split("\n");
+    if (lines.length > READ_MAX_LINES)
+      return json(res, 200, { path, text: lines.slice(0, READ_MAX_LINES).join("\n"), truncated: true, total_lines: lines.length, hint: `仅回前 ${READ_MAX_LINES} 行；要后面的加 offset=${READ_MAX_LINES}（可继续递增）` });
     return json(res, 200, { path, text });
   }
 
diff --git a/team.example.yaml b/team.example.yaml
new file mode 100644
index 0000000..4c9a489
--- /dev/null
+++ b/team.example.yaml
@@ -0,0 +1,15 @@
+# 团队花名册（公开可提交，不含密钥）。复制成 team.yaml 填真实成员。
+# 身份归一：把每个人本机 git user.name 的各种写法都列进 git_names，
+# 大脑里就能把 session 归到同一个 canonical 成员（作者显示、按人检索靠它）。
+# 每人的接入 token 另放 tokens.yaml（机密、已 gitignore），key = 这里的 id。
+# 加人也可不手写：管理员机器上 `brain admin add <id> --name <显示名> [--email][--git-name]`（经 SSH 落到服务器）。
+members:
+  - id: alice                      # 唯一短 id（= tokens.yaml 的 key、邀请码里的 id）
+    name: Alice Chen               # 显示名
+    email: alice@example.com       # 可选
+    git_names:                     # 本机 git user.name 的各种写法（可留空）
+      - alice
+      - Alice Chen
+  - id: bob
+    name: Bob
+    git_names: [bob]
diff --git a/web/app.js b/web/app.js
index 67599b5..cdb5bfa 100644
--- a/web/app.js
+++ b/web/app.js
@@ -68,7 +68,7 @@ const I18N = {
     "stat.attention": "Needs attention",
     "ov.recentActivity": "Recent activity",
     "ov.allSessions": "All sessions →",
-    "ov.attentionHead": "⚠️ Needs attention · unpushed",
+    "ov.attentionHead": "Active after last push",
     "ov.allRepos": "All repos →",
     "ov.recentDocs": "Recent docs",
     "ov.feishuDocs": "Docs →",
@@ -77,8 +77,8 @@ const I18N = {
     "load.docs": "Loading docs",
     "ov.noActivity": "No activity yet",
     "ov.noActiveRepos": "No active repos.",
-    "ov.branchesUnpushed": (n) => `${n} branches unpushed`,
-    "ov.allPushed": "All pushed ✓",
+    "ov.branchesUnpushed": (n) => `${n} active after push`,
+    "ov.allPushed": "Nothing newer than last push ✓",
     "ov.noMirror": "No doc mirror yet.",
     // 仓库
     "load.repos": "Loading repos",
@@ -87,7 +87,7 @@ const I18N = {
     "repos.active": (n) => `Active · ${n}`,
     "repos.noActive": "No active repos",
     "repos.registered": (n) => `Registered only · ${n}`,
-    "repos.unpushedBadge": (n) => `${n} unpushed`,
+    "repos.unpushedBadge": (n) => `${n} after push`,
     "card.registeredOnly": "Registered only",
     "card.meta": (n, agoStr, people) => `Active · ${n} sessions · ${agoStr}${people > 1 ? ` · ${people} people` : ""}`,
     // code-state
@@ -95,8 +95,8 @@ const I18N = {
     "cs.noPR": "No open PRs.",
     "cs.activeBranches": "Active branches",
     "cs.openPR": "Open PRs",
-    "cs.unpushed": "unpushed",
-    "cs.unpushedTitle": "Local session is newer than the last push",
+    "cs.unpushed": "after push",
+    "cs.unpushedTitle": "A session on this branch is newer than its last push — post-push activity, not necessarily unpushed commits",
     "cs.none": "No code-state for this repo yet (registered only, no sessions yet, or awaiting the first 4h poll).",
     // 单仓
     "load.repo": "Loading repo",
@@ -271,7 +271,7 @@ const I18N = {
     "stat.attention": "待关注",
     "ov.recentActivity": "最近活动",
     "ov.allSessions": "全部会话 →",
-    "ov.attentionHead": "⚠️ 待关注 · 未推进度",
+    "ov.attentionHead": "push 后有活动",
     "ov.allRepos": "全部仓库 →",
     "ov.recentDocs": "最近文档",
     "ov.feishuDocs": "文档 →",
@@ -280,8 +280,8 @@ const I18N = {
     "load.docs": "加载文档",
     "ov.noActivity": "暂无活动",
     "ov.noActiveRepos": "没有活跃仓库。",
-    "ov.branchesUnpushed": (n) => `${n} 分支未推`,
-    "ov.allPushed": "全部已推送 ✓",
+    "ov.branchesUnpushed": (n) => `${n} 个分支 push 后有活动`,
+    "ov.allPushed": "无晚于 push 的活动 ✓",
     "ov.noMirror": "还没有文档镜像。",
     "load.repos": "加载仓库",
     "repos.title": "GitHub 仓库",
@@ -289,15 +289,15 @@ const I18N = {
     "repos.active": (n) => `活跃 · ${n}`,
     "repos.noActive": "暂无活跃仓库",
     "repos.registered": (n) => `仅登记 · ${n}`,
-    "repos.unpushedBadge": (n) => `${n} 未推`,
+    "repos.unpushedBadge": (n) => `${n} 晚于推送`,
     "card.registeredOnly": "仅登记",
     "card.meta": (n, agoStr, people) => `活跃 · ${n} 会话 · ${agoStr}${people > 1 ? ` · ${people} 人` : ""}`,
     "cs.noBranches": "无活跃分支（30 天内）。",
     "cs.noPR": "无 Open PR。",
     "cs.activeBranches": "活跃分支",
     "cs.openPR": "Open PR",
-    "cs.unpushed": "未推进度",
-    "cs.unpushedTitle": "本地 session 比最后一次 push 新",
+    "cs.unpushed": "晚于推送",
+    "cs.unpushedTitle": "本分支有比最后一次 push 更晚的 session —— push 后的活动，不一定有未推 commit",
     "cs.none": "该仓尚无 code-state（仅登记、还没 session，或等首次 4h 轮询）。",
     "load.repo": "加载仓库",
     "repo.defaultBranch": "默认分支",
diff --git a/web/viewer.js b/web/viewer.js
index f8dcb05..5cfdb0a 100644
--- a/web/viewer.js
+++ b/web/viewer.js
@@ -18,7 +18,7 @@ const T = {
   "foot.sync": ["synced {t}", "同步 {t}"], "foot.noSync": ["no sync yet", "尚无同步"],
   "who": ["{name}'s local footprint", "{name} 的本机足迹"],
   "ov.title": ["Overview", "概览"],
-  "ov.sub": ["What this machine has published to the team's shared library — at a glance, retractable anytime.", "team-brain 从这台机器往全队共享库发布了什么 —— 一眼看清，随时可撤。"],
+  "ov.sub": ["What this machine has published to the team's shared library — at a glance, retractable anytime.", "Klear-Team-Brain 从这台机器往全队共享库发布了什么 —— 一眼看清，随时可撤。"],
   "ov.what": ["This page lives <b>only on your machine</b> (127.0.0.1) and <b>only you can see it</b>. It governs <span class=\"g\">what the collector uploads from this computer to the team's shared library</span>. Preview and exclude happen locally; only <b>Retract</b> reaches the shared library to delete an entry.",
     "这个页面<b>只在你本机</b>（127.0.0.1）、<b>只有你能看</b>。它管的是「<span class=\"g\">采集常驻从这台电脑上传到全队共享库的内容</span>」。预览、排除都在本地发生；点「撤回」才会去线上库删掉对应那条。"],
   "ov.recent": ["Recent activity", "最近活动"],