feat: choose your own pool account; skills with "token" in a doc name upload; System Health off Linux

- Skill snapshot/upload: a key file's name (.env, *.pem, id_rsa, .netrc,
  .credentials.json…) is always refused, but a word like token/secret/
  password/credential only in a data file's name — design-tokens.md or
  count_tokens.py is skill content, not a key.
- System Health: off Linux the History, CPU threads and Busiest processes
  cards (all read from /proc) are left out instead of shown empty; the CPU
  bar without a sample and Windows' always-zero Load row go too.
- Users pick their own pool account per tool (/api/me/pool-accounts,
  Settings → AI tools), within their groups, refused while pinned.
- Admin user detail, 9router update, session replay and Antigravity
  catalog changes from parallel sessions.

docs/API.md

@@ -105,6 +105,8 @@ An admin (server mode) adds an account with `POST /api/accounts` `{tool: "claude
 | `PUT /api/me/prefs` `{prefs: {<key>: "<value>" \| null}}` | merged (`null` deletes a key), the same shape back; `400 EBADPREFS` for a key outside the list below or a value that is not a string of at most 8 KB |
 | `GET /api/me/instructions` | `{instructions, max, tools}` — `tools` lists the AI tool ids whose new panes receive them (`["claude","codex"]`; Cursor CLI and Antigravity have no per-session way to take them). The account's own agent instructions (Settings → AI tools → My agent instructions, `""` when none); `401 EAUTH` with no login |
 | `PUT /api/me/instructions` `{instructions: "<markdown>" \| null}` | replaced, the same shape back; blank or `null` clears; `400 EBADFIELD` not a string, `413 ETOOBIG` over 16 000 characters. Put BEFORE the workspace's own instructions in every new Claude/Codex pane of a workspace the account owns (local mode: every workspace) — see Workspace context below |
+| `GET /api/me/pool-accounts` | server mode: `{tools: [{tool, pin, choice, accounts: [{name, kind, email, expired, quotaPercent, full}]}]}` — per tool (`claude`, `codex`, `cursor`) with at least one account the account's [pool group](SERVER-MODE.md#a-users-own-pool-account) allows; never a token. Local mode: `{tools: []}` |
+| `PUT /api/me/pool-accounts` `{tool, account: "<name>" \| null}` | the account's new sessions of `tool` run on `account` (`null` = automatic): `{tool, choice, switched}` — `switched` as for an admin's pin (a Claude login goes into the home at once). Below an admin's pin (`409 ELOCKED`), only an account the group allows (`404 ENOACCOUNT`), `400 EBADTOOL`; login cookie only, audited |
 | `POST /api/me/onboarded` | the first-run setup is done or skipped (kept from the first call) |
 | `GET /api/me/account` | Settings → Account. Server mode: `{mode: "server", user: {id, username, role, disabled, displayName, email, avatarAt, createdAt}, workspaces, sessions, logins, lastLoginAt}`; local mode `{mode: "local"}` |
 | `PATCH /api/me/account` `{displayName?, email?}` | the account's own profile (`""`/`null` clears), the same shape back; `400 EBADFIELD` (display name over 100 characters or with control characters) / `EBADEMAIL` |
@@ -3010,7 +3012,7 @@ GET    /api/ai-tools/:id/config/:i    { path, format, text, sha }
 PUT    /api/ai-tools/:id/config/:i    Body: { text, baseSha } → { sha, backup }
 POST   /api/ai-tools/:id/install      Start the catalog's install job → 202 { job }
 POST   /api/ai-tools/:id/update       Start the catalog's update job → 202 { job }
-POST   /api/ai-tools/update-all       One job: every npm-managed tool with a newer version + every self updater
+POST   /api/ai-tools/update-all       One job: every npm-managed tool with a newer version + every self updater (any signed-in user; other /ai-tools writes are an admin's)
 GET    /api/ai-tools/job              The current or last job, or null
 GET    /api/ai-tools/versions         { [id]: { npmManaged, installed, latest, updateAvailable } }
 GET    /api/ai-tools/:id/icon         The tool's favicon, fetched once by the server; 204 → letter avatar
@@ -3038,7 +3040,7 @@ The login profiles of the AI tools other than Claude Code (`server/lib/tool-prof
 GET    /api/ai-proxies                Every saved profile, the built-in `native` one first
 GET    /api/ai-proxies/9router        `{ installed }` — is a `9router` binary on the server's PATH (nothing is run)
 GET    /api/ai-proxies/9router/update versions, running state, `blocker`, current job (`?fresh=1` re-asks npm)
-POST   /api/ai-proxies/9router/update shut down → npm update → start hidden; 202 + job (admin)
+POST   /api/ai-proxies/9router/update shut down → npm update → start hidden; 202 + job (any signed-in user)
 POST   /api/ai-proxies                Create — Body: { label, baseUrl, apiKey?, model, models?, behavesAs?, scope? }
 POST   /api/ai-proxies/test           Probe an endpoint — Body: { baseUrl, model, apiKey?, name? }
 PATCH  /api/ai-proxies/:name          Update the same fields, all optional
@@ -3061,7 +3063,7 @@ A **proxy profile** is an endpoint that stands in front of Claude Code or Codex
 
 **`GET /api/ai-proxies/9router`** answers `{ installed: boolean }` from a PATH lookup only (the server's and the login shell's PATH); the AI Proxy page shows a **Config 9router** link beside **Add profile**, opening 9router's own UI in a new tab on the dashboard's own host, port 20128 (`http://<the host the page was opened on>:20128` — e.g. `http://10.1.13.7:20128` from a LAN browser, so the viewer needs no 9router of their own; 9router must listen beyond loopback for that), only when it is `true`.
 
-**Update 9router** (`server/lib/nine-router-update.js`; the AI Proxy page's button beside Config 9router, an admin's only — `/ai-proxies/9router` and `/ai-routers/9router` are in `ADMIN_WRITE_PREFIXES`). `GET /api/ai-proxies/9router/update` answers `{ installed, version, latest, updateAvailable, running, managedBy: "systemd"|"process"|null, blocker: {code, message, command?}|null, job }`. `POST` starts one job (202; `EBUSY` while one runs, else 409 with the blocker's code — checked BEFORE anything is stopped) of three steps, each `pending|running|done|failed|skipped`:
+**Update 9router** (`server/lib/nine-router-update.js`; the AI Proxy page's button beside Config 9router, shown to every user — `/ai-proxies/9router` and `/ai-routers/9router` are in `ADMIN_WRITE_PREFIXES`, but their `POST …/update` is in `MEMBER_WRITE_EXCEPTIONS`). `GET /api/ai-proxies/9router/update` answers `{ installed, version, latest, updateAvailable, running, managedBy: "systemd"|"process"|null, blocker: {code, message, command?}|null, job }`. `POST` starts one job (202; `EBUSY` while one runs, else 409 with the blocker's code — checked BEFORE anything is stopped) of three steps, each `pending|running|done|failed|skipped`:
 1. **shutdown** — a running `9router.service` user unit is `systemctl --user stop`ped; otherwise this account's processes whose argv names the install's `cli.js` get SIGTERM, then they and their descendants SIGKILL after 5 s; the step waits until the port is free. Skipped when nothing ran.
 2. **update** — `npm install -g --prefix <the install's own prefix> --prefer-online 9router@latest`; a prefix this account cannot write runs it under `sudo -n` with `--cache /var/cache/agenthub-9router-npm` (root's HOME is inaccessible inside the AgentHub unit; the sudoers rule must name that exact argv), only when `sudo -n <npm> --version` succeeds (else `ENOPERM` with `command: "sudo npm install -g 9router@latest"`).
 3. **relaunch** — only when 9router was running before: `systemctl --user start 9router.service`, or `node cli.js --tray --no-browser --skip-update -p <old port> [-H <old host>]` detached in its own `systemd-run --user --scope` (plain detached spawn without one); done once the port answers. Runs even after a failed update.
@@ -3165,7 +3167,7 @@ What agents produced, per OWNER (`server/lib/artifacts.js`, plan: `docs/plans/ar
 
 - **page** — a claude.ai link (`https://claude.ai/…` only, `400 EBADURL`), one row per owner and link. A session's `Artifact` tool publish — or a Claude Docs `batch` that creates a doc (`container.create`; its title is the doc's name, later batches are skipped) — is captured from its PostToolUse hook (a doc's own `frame.artifactUrl`; for a publish the first `https://claude.ai/…artifact/…` in `tool_response`), owned by the session's workspace owner — or, for a session no workspace owns (a member's `claude` started from their own terminal under S3), by the member whose account sent the hook (by its connection's uid, never the body); the host account names nobody, so such a page from the host stays unowned, with `suggested_owner_id` set to the owner of the workspace whose folder holds the session's `cwd` (else, under S3, the member whose home it is).
 - **file** — `candidates` are the documents (html, md, pdf, images, csv, office…) the workspace's sessions wrote with Write/Edit/MultiEdit/NotebookEdit, inside its folder. **Captured at Stop:** when a session a workspace owns sends `Stop` or `SessionEnd`, each `.pptx`/`.docx`/`.xlsx`/`.pdf`/`.html` in the workspace it created or changed since it started (a repo's untracked or modified files — an HTML file only when untracked — or a plain folder's files; at most 20 per stop, ≤ 25 MB) is snapshotted into the workspace owner's library as `auto: true` (shown **Auto**), off the hook's response path. The same file at the mtime its snapshot recorded is skipped; a changed one replaces that row's snapshot; a file pinned by hand is left alone. A pin is read under the download route's guards (`lanePath`, `resolveInLane`, `ESECRET`, ≤ 25 MB, `413 ETOOBIG`) and SNAPSHOTTED to `<data dir>/artifacts/<random>/`.
-- **skill** — the caller's own `~/.claude/skills/<name>` (their member home under S3; the host's for an admin or in local mode, else `409 ENOACCOUNT`) or a workspace's `.claude/skills/<name>`. Snapshotted whole: ≤ 300 files, ≤ 5 MB, a symlink anywhere is `ESYMLINK`, a secret-looking file name (`.env`, `*.pem`, `*token*`…) `403 ESECRET`.
+- **skill** — the caller's own `~/.claude/skills/<name>` (their member home under S3; the host's for an admin or in local mode, else `409 ENOACCOUNT`) or a workspace's `.claude/skills/<name>`. Snapshotted whole: ≤ 300 files, ≤ 5 MB, a symlink anywhere is `ESYMLINK`, a secret-looking file name (`.env`, `*.pem`, `id_rsa`, `.netrc`…, or a data file named `*token*`/`*secret*`/`*password*`/`*credential*` — the word alone does not count in Markdown, source code or images, e.g. `design-tokens.md`) `403 ESECRET`.
 - **file upload** — one file from the user's computer (≤ 25 MB, `413 ETOOBIG`), snapshotted under its own base name (no folder kept).
 - **skill upload / new** — a skill made from files rather than picked: one Markdown/text file of any name becomes `SKILL.md`; several files or a picked/dropped folder (`paths` keeps the layout) need a `SKILL.md` (at the top or in one folder, which is stripped — `400 ENOSKILLMD` otherwise); a single `.zip`/`.skill` is unpacked (sizes checked from its directory first, a symlink entry `ESYMLINK`, `__MACOSX`/`.DS_Store`/`.git`/`node_modules` skipped); an upload holding several skill folders makes one artifact per skill, named by its folder, and stores NOTHING if any of them fails. The `name` the user typed (only for a single skill; else the frontmatter `name:`, the folder or the file name) is made a folder name (`skillSlug`: marks dropped, other characters → `-`; `400 EBADNAME` if nothing is left) and written into `SKILL.md`'s frontmatter as `name:`; a `description` given replaces the frontmatter's, and a skill with none anywhere is `400 ENODESCRIPTION`. Paths are relative with no `..` (`EBADPATH`); the same limits and `ESECRET` as a picked skill. `new` writes `SKILL.md` from `content` the same way.
 - **plugin** — only the spec and its marketplace source (`EBADPLUGIN`, `EBADMARKET`: `owner/repo`, a `validateRemoteUrl` URL, or an https link to a `marketplace.json`); no files. A GitHub web link becomes `owner/repo`, and a forge's `…/src/…`, `…/tree/…`, `…/blob/…` suffix or trailing slash is trimmed. The same `key` added again by its owner updates that row. **inspect** reads the marketplace's `.claude-plugin/marketplace.json` to list its plugins: a git link is cloned shallow, blob-less and without checkout into a temp dir that is removed after, with every credential helper switched off and no prompt (a repository that needs a login is `502 ENEEDLOGIN`; type the key by hand — installing then uses each user's own git login), https only (an ssh link is `400 ESSHMARKET`); a repo with only a `plugin.json` is `400 ENOTMARKET`; unreachable `502 EMARKETFETCH`. It reaches any https host a signed-in user names, like `publish` does — no login is ever sent with it.