release 2026.09.09.01

Twenty-seven commits and ten days. One new table, plugin_ports, which is a plain create — nothing references it, so db:push has no ordering problem to get wrong. The marketplace's nullable release_tag is deliberately out of scope: db:push covers officer_db, and the marketplace runs only on officer-dev, which the rollout skips. @anthropic-ai/sdk is new, so bun install is a step. The lockfile is committed and matches; a host that cannot resolve from it is a finding, not something to force past.
anthropicViaProxy — the boundary for platform-side Claude calls
2026-09-09 23:17:12 +01:00 · 2026-09-09 22:48:05 +01:00 · 2026-09-08 22:10:18 +01:00 · 2026-09-08 15:16:44 +01:00 · 2026-09-07 14:06:46 +01:00 · 2026-09-06 15:25:13 +01:00
47 changed files with 1997 additions and 273 deletions
@@ -22,14 +22,14 @@ the owner's OS user and can never be granted. Indirection there really is accide

 ## Themes

-**108 open, 33 closed**, across 35 themes.
+**109 open, 33 closed**, across 35 themes.

 | Theme | Open | Closed |
 | --- | --- | --- |
 | [Affordances](todo/affordances/) | 2 | 0 |
 | [Agents coordinating agents](todo/agent-coordination/) | 5 | 0 |
 | [Speaking into a session from the backend](todo/agent-delivery/) | 1 | 1 |
-| [Backburner](todo/backburner/) | 3 | 0 |
+| [Backburner](todo/backburner/) | 4 | 0 |
 | [Background jobs](todo/background-jobs/) | 2 | 1 |
 | [Calendar and contacts](todo/calendar/) | 1 | 0 |
 | [Chat interface](todo/chat/) | 1 | 0 |
@@ -71,6 +71,7 @@ the owner's OS user and can never be granted. Indirection there really is accide

 ## Newest ten

+- `2026-09-04` **M3E Canvas as a plugin — sketch a UI, hand the prompt to the agent** — [backburner/m3e-canvas-as-a-plugin](todo/backburner/m3e-canvas-as-a-plugin.md)
 - `2026-08-29` **Officer blocks its own event loop for ~0.9s re-parsing a live transcript, and everything it serves stalls with it** — [known-bugs/officer-blocks-its-event-loop-parsing-transcripts](todo/known-bugs/officer-blocks-its-event-loop-parsing-transcripts.md)
 - `2026-08-27` **Turn an ordinary chat into a commander or a bot, carrying its context along** — [agent-coordination/turn-a-chat-into-a-commander-or-a-bot](todo/agent-coordination/turn-a-chat-into-a-commander-or-a-bot.md)
 - `2026-08-27` **The calendar API cannot express repeating events, reminders, invitees or a cancelled occurrence** — [calendar/the-json-api-cannot-express-recurrence-alarms](todo/calendar/the-json-api-cannot-express-recurrence-alarms.md)
@@ -80,4 +81,3 @@ the owner's OS user and can never be granted. Indirection there really is accide
 - `2026-08-27` **Setting up a phone for calendar and contacts should be one link inside Officer, not a manual account form** — [mobile-apps/setting-up-a-phone-should-be-a-link-in-the-app](todo/mobile-apps/setting-up-a-phone-should-be-a-link-in-the-app.md)
 - `2026-08-27` **Marketplace: browse by tag — filter list, card grid, detail with a back button** — [plugin-ecosystem/marketplace-browse-by-tag](todo/plugin-ecosystem/marketplace-browse-by-tag.md)
 - `2026-08-25` **An affordance: a scoped agent warns when a request lands outside its scope** — [affordances/an-affordance-a-scoped-agent-warns-on-the](todo/affordances/an-affordance-a-scoped-agent-warns-on-the.md)
- `2026-08-25` **Commit signing is inconsistent across harnesses — Claude names itself, opencode names nothing** — [agent-coordination/commit-signing-is-inconsistent-across-harnesses](todo/agent-coordination/commit-signing-is-inconsistent-across-harnesses.md)
@@ -6,6 +6,7 @@
      "name": "officer",
      "dependencies": {
        "@anthropic-ai/claude-agent-sdk": "^0.2.41",
+        "@anthropic-ai/sdk": "^0.71.0",
        "@bitcoinerlab/secp256k1": "1.2.0",
        "@getalby/sdk": "^8.0.3",
        "@hookform/resolvers": "^5.2.2",
@@ -324,6 +325,8 @@

    "@anthropic-ai/claude-agent-sdk": ["@anthropic-ai/claude-agent-sdk@0.2.59", "", { "optionalDependencies": { "@img/sharp-darwin-arm64": "^0.34.2", "@img/sharp-darwin-x64": "^0.34.2", "@img/sharp-linux-arm": "^0.34.2", "@img/sharp-linux-arm64": "^0.34.2", "@img/sharp-linux-x64": "^0.34.2", "@img/sharp-linuxmusl-arm64": "^0.34.2", "@img/sharp-linuxmusl-x64": "^0.34.2", "@img/sharp-win32-arm64": "^0.34.2", "@img/sharp-win32-x64": "^0.34.2" }, "peerDependencies": { "zod": "^4.0.0" } }, "sha512-xPOUZZimZI5ChaO791olWGXqaRvCwOfj9/1micu42EL9czdcwiDm0WK1OGsqb2mZ7LSCoYWBB0ZHVKOxehemDA=="],

+    "@anthropic-ai/sdk": ["@anthropic-ai/sdk@0.71.2", "", { "dependencies": { "json-schema-to-ts": "^3.1.1" }, "peerDependencies": { "zod": "^3.25.0 || ^4.0.0" }, "optionalPeers": ["zod"], "bin": { "anthropic-ai-sdk": "bin/cli" } }, "sha512-TGNDEUuEstk/DKu0/TflXAEt+p+p/WhTlFzEnoosvbaDU2LTjm42igSdlL0VijrKpWejtOKxX0b8A7uc+XiSAQ=="],
+
    "@babel/code-frame": ["@babel/code-frame@7.29.0", "", { "dependencies": { "@babel/helper-validator-identifier": "^7.28.5", "js-tokens": "^4.0.0", "picocolors": "^1.1.1" } }, "sha512-9NhCeYjq9+3uxgdtp20LSiJXJvN0FeCtNGpJxuMFZ1Kv3cWUNb6DOhJwUvcVCzKGR66cw4njwM6hrJLqgOwbcw=="],

    "@babel/compat-data": ["@babel/compat-data@7.29.0", "", {}, "sha512-T1NCJqT/j9+cn8fvkt7jtwbLBfLC/1y1c7NtCeXFRgzGTsafi68MRv8yzkYSapBnFA6L3U2VSc02ciDzoAJhJg=="],
@@ -1616,6 +1619,8 @@

    "json-bigint": ["json-bigint@1.0.0", "", { "dependencies": { "bignumber.js": "^9.0.0" } }, "sha512-SiPv/8VpZuWbvLSMtTDU8hEfrZWg/mH/nV/b4o0CYbSxu1UIQPLdwKOCIyLQX+VIPO5vrLX3i8qtqFyhdPSUSQ=="],

+    "json-schema-to-ts": ["json-schema-to-ts@3.1.1", "", { "dependencies": { "@babel/runtime": "^7.18.3", "ts-algebra": "^2.0.0" } }, "sha512-+DWg8jCJG2TEnpy7kOm/7/AxaYoaRbjVB4LFZLySZlWn8exGs3A4OLJR966cVvU26N7X9TWxl+Jsw7dzAqKT6g=="],
+
    "json-schema-traverse": ["json-schema-traverse@1.0.0", "", {}, "sha512-NM8/P9n3XjXhIZn1lLhkFaACTOURQXjWhV4BA/RnOv8xvgqtqpAX9IO4mRQxSx1Rlo4tqzeqb0sOlruaOy3dug=="],

    "json-schema-typed": ["json-schema-typed@8.0.2", "", {}, "sha512-fQhoXdcvc3V28x7C7BMs4P5+kNlgUURe2jmUT1T//oBRMDrqy1QPelJimwZGo7Hg9VPV3EQV5Bnq4hbFy2vetA=="],
@@ -2290,6 +2295,8 @@

    "trough": ["trough@2.2.0", "", {}, "sha512-tmMpK00BjZiUyVyvrBK7knerNgmgvcV/KLVyuma/SC+TQN167GrMRciANTz09+k3zW8L8t60jWO1GpfkZdjTaw=="],

+    "ts-algebra": ["ts-algebra@2.0.0", "", {}, "sha512-FPAhNPFMrkwz76P7cdjdmiShwMynZYN6SgOujD1urY4oNm80Ou9oMdmbR45LotcKOXoy7wSmHkRFE6Mxbrhefw=="],
+
    "ts-interface-checker": ["ts-interface-checker@0.1.13", "", {}, "sha512-Y/arvbn+rrz3JCKl9C4kVNfTfSm2/mEp5FSz5EsZSANGPSlQrpRI5M4PKF+mJnE52jOO90PnPSc3Ur3bTQw0gA=="],

    "tslib": ["tslib@2.8.1", "", {}, "sha512-oJFu94HQb+KVduSUQL7wnpmqnfmLsOA/nAh6b6EH0wCEoK0/mPeXU6c3wKDV83MkOuHPRHtSXKKU99IBazS/2w=="],
@@ -0,0 +1,55 @@
+# 2026.08.30.03 — report
+
+**7 of 7 target hosts succeeded.** `officer-dev` skipped entirely, as the host the rollout ran on.
+
+Every target went `816f8695` → `c97da029` by `git merge --ff-only deploy/2026.08.30.03`. No host was
+dirty in tracked files, none had to be skipped, and no merge was anything other than a fast-forward.
+
+**The release did what it was cut for.** `fofos` went 4/5 → **5/5, exit 0**, with `fofos-admin` reading
+`packages: installed and resolvable`. That was the one check that mattered and it passed.
+
+**No schema step and no `bun install` step** — the plan has neither, and neither was improvised. So there
+is no per-host schema query in this report, because this release introduced nothing to query for.
+
+## Per host
+
+| Host | HEAD before → after | `plugins:update` | `officer` restart_time | health |
+| --- | --- | --- | --- | --- |
+| `officer-dev` | — | — | — | **SKIPPED** — the host this ran on |
+| `alpha` | `816f8695` → `c97da029` | 15/15 | 7 → 8, online | `401` (9000) |
+| `fofos` | `816f8695` → `c97da029` | **5/5, exit 0** | 4 → 5, online | `401` (9000) |
+| `macbookpro` | `816f8695` → `c97da029` | no plugins installed | 7 → 8, online | `401` (9010) |
+| `macbruno` | `816f8695` → `c97da029` | 1/1 | 10 → 11, online | `401` (9000) |
+| `production-pertento-bridge` | `816f8695` → `c97da029` | 1/1 | 8 → 9, online | `401` (9010) |
+| `edge-pertento-bridge` | `816f8695` → `c97da029` | 2/2 | 9 → 10, online | `401` (9000) |
+| `deadpool-rezio-bridge` | `816f8695` → `c97da029` | 2/2 | 26 → 27, online | `401` (9000) |
+
+`401` on `/api/health` is healthy — the endpoint is behind auth.
+
+## What is worth reading
+
+- **`fofos` is fixed.** `plugins:update` there had failed 4/5 on every rollout since `@tiptap/pm` was
+  installed, and it now exits 0. The plugin step of the next deploy should be unremarkable on that host.
+- **`macbookpro` needed a fetch retry.** The first `git fetch` failed with
+  `Connection closed by 77.42.39.155 port 22` — gitea's public SSH, rate-limited at the proxy. That host
+  resolves `gitea.officer.dev` to the public address where `alpha` reaches it over the tailnet. The first
+  retry succeeded and the rollout continued normally. A later probe was refused again, so the window is
+  intermittent rather than closed. **Nothing to fix in the deploy; the remote on that host is the thing
+  that is wrong**, and pointing it at the tailnet address would remove a class of failure that costs a
+  retry on every release.
+- **`scripts/chk.ts` is still untracked on `fofos`.** Noted, not a skip, exactly as the plan said. It has
+  survived two rollouts now; somebody may want to commit it or delete it.
+- **`plugins:update` restarted sidecars on several hosts** before the `pm2 restart officer` the plan
+  asked for — `officer-music` on `alpha`, `officer-gitea` on `production-pertento-bridge`,
+  `officer-offscale` on `edge-pertento-bridge`, `officer-email` on `deadpool-rezio-bridge`. That is the
+  documented behaviour of the plugins step, not something this rollout chose.
+- **On `production-pertento-bridge` only `officer` was restarted.** Pertento's eight production processes
+  were not touched.
+
+## What was not done
+
+- `officer-dev` was not pulled, not updated and not restarted. It already holds `c97da029` and its
+  `officer` was restarted by hand at 11:23. **The owner restarts this host deliberately, as the last
+  step**, once happy with the rest.
+- No `bun db:push`, no `bun install`, no direct SQL. The plan called for none.
+- No file-browser check, per the plan.
@@ -0,0 +1,129 @@
+# 2026.09.09.01 — ten days of platform work, and one new table
+
+**Twenty-seven commits**, `deploy/2026.08.30.03..fafc1adb`. The fleet is ten days behind; this is a
+catch-up release rather than a fix for anything found in the last one.
+
+The larger threads in it, so a host that misbehaves has somewhere to point:
+
+- **Plugin ports** — the platform can allocate a plugin a stable port and remember it (`518d329c`,
+  `d9110c37`). This is the schema change; see below.
+- **Publishing is a tag** — the marketplace reads releases, not working trees (`78e89ebc`, `19bbbab9`).
+- **Plugin routing** — a plugin owns the routing under its mount and can declare its own routes
+  (`89820972`, `1fe767ea`), with `:section` given back to the fifteen plugins the splat took it from
+  (`eaa92738`).
+- **The code editor's tree and file browser** are exported for plugins to reuse (`fb530456`, `5fa3f3ec`,
+  `da07048f`).
+- **`anthropicViaProxy`** (`fafc1adb`) — the boundary through which platform-side code, including a
+  plugin's server routes, reaches Claude on the owner's own subscription.
+
+## The schema change — one new table, one nullable column
+
+Verified rather than assumed:
+
+```bash
+git diff --name-only deploy/2026.08.30.03..master | grep -iE 'schema|databases/'
+```
+
+Two files, and they are **not** equally in scope:
+
+1. **`officer_db` — `plugin_ports`, a brand-new table.** No foreign key onto it and no existing column
+   pointing at it, so this is not the create-then-constrain shape that `db:push` cannot do. A plain
+   `bun db:push` is enough, and it is a create on every host.
+
+2. **`marketplace_db` — `plugins.release_tag`, a nullable text column.** **Out of scope for this
+   rollout.** `bun db:push` covers `officer_db` only, and the marketplace runs on exactly one machine:
+
+   ```
+   officer-marketplace → officer-dev only    (verified 2026-09-09, pm2 jlist on all six)
+   ```
+
+   `officer-dev` is the host the rollout skips. So no target needs it, and nothing should improvise a
+   second push to reach it. It is named here so that when `officer-dev` is restarted deliberately, the
+   marketplace database is known to be the one outstanding piece.
+
+## Dependency change — `bun install` is required
+
+```bash
+git diff --name-only deploy/2026.08.30.03..master -- package.json bun.lock   # both changed
+```
+
+`@anthropic-ai/sdk@^0.71.0` was added (plus its `json-schema-to-ts` dependency). The lockfile is
+committed and matches, so the frozen-lockfile default resolves from it — **plain `bun install`, never
+`--no-frozen-lockfile`.** A host that cannot resolve from the lockfile is a finding, not something to
+force past.
+
+## Every host — the commands
+
+The standard steps in `prompts/deploy-commander.md` are the rollout (dirty check — TRACKED only,
+`git fetch origin --tags`, `git merge --ff-only deploy/2026.09.09.01`, `bun run plugins:update`), plus,
+in this order:
+
+```bash
+bun install
+bun db:push < /dev/null
+pm2 restart officer
+```
+
+`bun install` **before** `plugins:update`, because the plugin installer runs against the platform's own
+resolved modules. `db:push` after both, because `gen-plugin-schemas` runs inside it and reads the
+plugins as they now are.
+
+`< /dev/null` on the push is not decoration: stdin closed means a push that wants to ask a question
+fails instead of hanging forever in an unattended run. This change should ask nothing — a create of a
+table nothing else references — so **a prompt here is a finding.**
+
+`pm2 restart officer` only. `plugins:update` has already restarted the plugin sidecars itself.
+
+### Verify it came back — three checks, per host, by query
+
+Exit 0 from `db:push` is not evidence; `drizzle-kit push` can fail and still exit 0.
+
+1. **The table exists**, asked of that host's own database:
+
+   ```bash
+   psql "$DB_URL" -tAc "select to_regclass('public.plugin_ports')"    # → plugin_ports
+   ```
+
+   Read the URL from that checkout's `.env` — several hosts use `OFFICERDEV_POSTGRES_URL`, not
+   `POSTGRES_URL`.
+
+2. **`pm2 jlist`** — `officer` is `online` and its `restart_time` moved.
+
+3. **HTTP** — `/api/health` on the host's own port. `200` or `401` are both healthy. The port is in
+   `.env` and is **9010 on `macbookpro`**, 9000 elsewhere.
+
+## Per host
+
+| Host                    | Login          | State before                   | Note                                                      |
+| ----------------------- | -------------- | ------------------------------ | --------------------------------------------------------- |
+| `alpha`                 | `pastilhas`    | `19bbbab9`, 8 commits past the tag | Partially pulled at some point; the fast-forward is shorter than the others'. |
+| `fofos`                 | `pastilhas`    | `fafc1adb` — already at master  | Already carries every commit; the merge is a no-op that only moves the tag onto it. Still runs install/push/restart. One untracked file — a note, not a skip. |
+| `macbookpro`            | `pastilhas`    | `c97da029` (`2026.08.30.03`)   | Laptop. Asleep is a SKIP with a note. `export PATH=$PATH:/opt/homebrew/bin` for `bun`, and run `pm2` as a separate command. Port **9010**. |
+| `deadpool-rezio-bridge` | **`br`**       | `c97da029` (`2026.08.30.03`)   | Not `pastilhas`. |
+
+### Skipped, deliberately — not failures
+
+- **`officer-dev`** — the host the rollout runs on. Skipping it entirely is the rule; restarting
+  `officer` there ends the rollout mid-report. It is at `fafc1adb` and therefore **not behind** in code;
+  what it does still want is the marketplace column above, whenever the owner restarts it.
+- **`macbruno`** — unreachable at cut time (`connect to host macbruno port 22: Connection timed out`,
+  verified 2026-09-09). A Mac, so most likely asleep.
+- **Both Pertento hosts** — on the deliberately-skipped list in `docs/servers.md` since 2026-09-08.
+  Owner's call, no fault on either box.
+
+## What "done" looks like
+
+Four hosts attempted: `alpha`, `fofos`, `macbookpro`, `deadpool-rezio-bridge`. Each at
+`deploy/2026.09.09.01`, `plugin_ports` present by query, `officer` back online, health answering on the
+right port.
+
+A host that fails does not stop the rollout. It gets its own line saying which step failed and what
+never ran.
+
+## Known and carried, not introduced here
+
+`bunx tsgo` reports one pre-existing type error in `src/servers/sidecar/claude/claude-manager.ts:667`
+(`SdkUserMessage` vs `SDKUserMessage`, from the agent SDK's own types). It is present on `master`
+before this release's commits, it is type-only, and Bun strips types at run time — so it does not
+affect the rollout. Named here so it is not discovered mid-deploy and mistaken for something this
+release did.
@@ -29,17 +29,21 @@ Ranked by what actually blocks something.

 ## Staleness

- [ ] **Nothing refreshes the catalogue automatically.** Publishing a plugin update changes nothing until
-      somebody runs `bun --env-file=.env scripts/marketplace-sync.ts`. A Gitea webhook or a cron is the
-      obvious next step.
+- [x] **Nothing refreshes the catalogue automatically.** Done 2026-09-03: `scripts/marketplace-cron.sh`
+      every minute. A Gitea webhook would make it a second instead, at the price of a public write route
+      on the one process here with no login — not worth it for a minute.

- [ ] **The sync reads local clones, not the remotes.** The machine running it must have every plugin
-      checked out, so the catalogue describes that machine's copies. The version that clones each repo
-      into a temp directory is a small change.
+- [x] **The sync reads local clones, not the remotes.** Done 2026-09-03. Bare mirrors, read at the newest
+      tag, driven by the registry rather than by a directory listing. The catalogue no longer describes
+      any machine's working copies.

- [ ] **`rawUrl()` assumes Gitea and a `main` branch.** A GitHub-hosted plugin, or one whose default branch
-      is `master`, gets a 404 for its icon. The forge belongs in the registry rather than inferred from a
-      URL shape.
+- [ ] **`rawUrl()` assumes Gitea.** A GitHub-hosted plugin gets a 404 for its icon. The forge belongs in
+      the registry rather than inferred from a URL shape. (The `main`-vs-`master` half of this is gone —
+      URLs pin to the release tag now.)
+
+- [ ] **Four plugins have never been tagged**, so the store still shows rows the old sync took from a
+      working tree: av-club, kanban, openseo, fofos-admin. They are frozen until somebody tags them, and
+      the sync names them on every run somebody asks for.

 ## Scale

@@ -10,8 +10,9 @@ from. **The website installs nothing** — installing means cloning a repository
 starting a process on someone else's machine, so it happens entirely inside the owner's own Officer.

 ```
-gitea.officer.dev/plugins/*   the repositories (source of truth)
-        │  scripts/marketplace-sync.ts  (manual, reads manifest.ts + README.md)
+gitea.officer.dev/**          the repositories (source of truth)
+        │  bare mirrors, read AT THE NEWEST TAG — src/servers/marketplace/sync.ts
+        │  every minute, scripts/marketplace-cron.sh
        ▼
 officer_marketplace (postgres)          plugins + plugin_metadata
        │  src/servers/marketplace/server.tsx  :9100
@@ -20,6 +21,7 @@ marketplace.officer.dev                 GET /api/plugins, /api/plugins/:appName
        │  fetchCatalogue()  live → cached → bundled
        ▼
 an Officer install                      /marketplace → clone → push schema → mount
+                                        and the update badge — see "Publishing" below
 ```

 ## The parts
@@ -78,19 +80,87 @@ disappearing from the store.
 the clone, the plugin's own manifest is what the platform reads. A metadata row that disagrees with the
 repository is a marketplace bug, never a second source of truth.

-## Sync
+## Publishing
+
+**Push a tag. That is the whole procedure.** A timer on officer-dev picks it up within a minute and every
+Officer install sees it on the next page load.

 ```bash
-bun --env-file=.env scripts/marketplace-sync.ts
+bun --env-file=.env scripts/marketplace-sync.ts            # all of them, now
+bun --env-file=.env scripts/marketplace-sync.ts photos     # one
+bun --env-file=.env scripts/marketplace-sync.ts --add https://gitea.officer.dev/plugins/finances.git
+bun --env-file=.env scripts/marketplace-sync.ts --force    # re-publish rows this file's own logic changed
 ```

-Reads every directory under `plugins/`, imports its `manifest.ts`, derives `has` from the directory
-rather than from a declaration, reads `README.md`, and upserts. Re-running is always safe: it overwrites,
-never merges.
+### It reads releases, never a working tree

-It rewrites SSH remotes to HTTPS. `example` and `offscale` had `git@gitea…` origins, which would have
-handed every installer a URL that only works on the machine that published it — the install fails at
-authentication, which reads as "the plugin is broken" rather than "the catalogue is wrong".
+Every registered plugin is mirrored bare into `~/.cache/officer-marketplace/mirrors` and read at its
+newest tag — manifest, README and file list all out of `git cat-file`. There is no working tree in the
+path, so publishing something uncommitted is not a mistake that can be made.
+
+That is a change, and it was load-bearing. Until 2026-09-03 the sync imported `manifest.ts` off
+`platform/plugins/` — the directories THIS machine has installed — so publishing meant "get officer-dev's
+checkout to the release, then run the script", and nothing showed when that had not happened. On the day
+it was rewritten the live store was advertising four plugins from untagged working copies: av-club
+2026.08.29.01, kanban 2026.08.29.02, openseo 0.1.0 and fofos-admin 2026.08.29.26, none of which had ever
+been tagged. It is also why the job could not go on a timer before: a cron over it published whatever was
+in a working tree, on a schedule.
+
+### The registry decides which plugins, and where from
+
+The list comes from the `plugins` table — the half this document has always called authoritative, and
+which was in fact a by-product of what happened to be checked out. Now it decides. A plugin nobody has
+installed here can still be published, and a plugin lives wherever its row says: `fofos-admin` is under
+`fofos/`, not `plugins/`, which an org listing would have missed. Registering one is `--add`.
+
+### A plugin with no tag is left alone
+
+It has never made a release, so the sync has nothing true to say about it: the existing row stays and the
+run names it. Deleting it would drop somebody's plugin out of the store on the day the rule changed.
+Tagging it is the fix, and the report is the reminder.
+
+### `release_tag`, and why it is not `version`
+
+`version` is the manifest's own claim; `release_tag` is what was pushed. They match every time somebody
+tags correctly, and when they do not, the tag is the fact — a manifest whose version nobody bumped would
+otherwise make a real release invisible to anything comparing versions. The sync warns on a mismatch.
+
+Icon URLs pin to the tag too. `/raw/HEAD/…` served whatever was on the default branch, via a 303, so a
+plugin could show one icon in the store and ship another.
+
+### The timer
+
+`scripts/marketplace-cron.sh`, `* * * * *`, logging to `officer-dev/marketplace-sync.log`. One
+timestamped summary line per tick — always, including the ticks that do nothing, because a fresh
+timestamp is the only thing that distinguishes "nothing to publish" from "the timer died". A plugin-level
+line appears only when one actually moved.
+
+A tick costs about 2.7s with nothing to do: one query, one fetch per repository, no writes. Nothing is
+written unless the release commit differs from the one already published.
+
+A Gitea webhook would cut the delay from a minute to a second, at the price of a public write route and a
+shared secret. A minute was never the problem anybody had — the complaint that started this was a release
+that never appeared at all.
+
+### Remotes
+
+`--add` rewrites SSH to HTTPS. A `git@gitea…` origin in the catalogue hands every installer a URL that
+only works on the machine that published it, and the install fails at authentication — which reads as
+"the plugin is broken" rather than "the catalogue is wrong".
+
+## The update badge reads the catalogue
+
+`updateable` on `/api/plugins` compares the tag at the checkout's HEAD with the tag the catalogue says is
+newest. Difference, not ordering: parsing `2026.08.27.01` works right up until somebody tags `v2`, and
+then it fails silently.
+
+Before this, each host ran a rate-limited background `git fetch` against all twenty-one repositories to
+re-derive a fact one place already knew. That is gone. `releaseState()` still holds the git rule as a
+fallback for a plugin the store cannot speak for — an unregistered checkout, or a store that could not be
+reached — and in that case its refs are whatever the last `pullPlugin` left, which is the accepted cost.
+
+`fetchCatalogue()`'s cache dropped from five minutes to one when the badge started depending on it:
+otherwise the wait for a release to appear was the sync's minute plus the cache's five.

 ## The platform degrades, it does not break

@@ -111,13 +181,8 @@ The full list, ranked, is `docs/marketplace-todo.md`. The ones that shape the de
 - **`[blocker]` Third-party submissions cannot use this sync.** It reads a manifest by importing it, which
  is code execution. Fine for our own repositories; not a mechanism a stranger's plugin can go through.
  Submissions need either a sandbox or a published `manifest.json` alongside the TypeScript one.
- **`[open]` Sync reads local clones, not the remotes.** The machine running it must have the plugins
-  checked out. The version that clones each repo into a temp directory is a small change and was left out
-  of the first pass.
- **`[open]` `rawUrl()` assumes Gitea and a `main` branch.** A GitHub-hosted plugin or one defaulting to
-  `master` gets a 404 for its icon. The forge should become part of the registry rather than inferred.
+- **`[open]` `rawUrl()` assumes Gitea.** A GitHub-hosted plugin needs the other URL form. The forge should
+  become part of the registry rather than inferred from the URL.
 - **`[open]` No versions table.** `plugin_metadata` is one row per plugin — a snapshot, not a history. A
  store that can offer "install 1.2 instead of 1.3" needs a different shape.
- **`[open]` Nothing refreshes automatically.** Publishing a plugin update does not change the catalogue
-  until somebody runs the sync. A cron or a Gitea webhook is the obvious next step.
 - **`[open]` No search, no categories, no counts.** Three plugins do not need them.
@@ -21,7 +21,8 @@ ssh br@deadpool-rezio-bridge    # not implied

 ## Officer platforms

-Eight, all running. **Verified 2026-08-27** — every row below was read off the machine, not recalled.
+Six, all running. **Verified 2026-08-27**, minus the two Pertento hosts moved to "Deliberately skipped" on
+2026-09-08 — every remaining row below was read off the machine, not recalled.

 | Host                         | Tailnet IP  | Login                                     | Port     | Checkout                            | Domain                            |
 | ---------------------------- | ----------- | ----------------------------------------- | -------- | ----------------------------------- | --------------------------------- |
@@ -30,8 +31,6 @@ Eight, all running. **Verified 2026-08-27** — every row below was read off the
 | `fofos`                      | 100.64.0.17 | `pastilhas`                               | 9000     | `~/officerdev/platform`             | `officer.fofosfluffybastards.com` |
 | `macbookpro`                 | 100.64.0.9  | `pastilhas`                               | **9010** | `~/projects/officer-suite/platform` | `macbookpro.pastilhas.dev`        |
 | `macbruno`                   | 100.64.0.22 | **`brunoalcadapadezmonteirodaconceicao`** | 9000     | `~/officerdev/platform`             | — (none yet)                      |
-| `production-pertento-bridge` | 100.64.0.12 | `pastilhas`                               | **9010** | `~/officerdev/platform`             | `officer.pertento.ai`             |
-| `edge-pertento-bridge`       | 100.64.0.15 | `pastilhas`                               | 9000     | `~/officerdev/platform`             | `edge.pertento.ai`                |
 | `deadpool-rezio-bridge`      | 100.64.0.21 | **`br`**                                  | 9000     | `/home/br/officerdev/platform`      | `officer.rezio.dev` `[pending]`   |

 **THE PORT IS NOT ALWAYS 9000.** Two hosts run on 9010 and it is in `OFFICERDEV_PORT` in each checkout's
@@ -70,12 +69,21 @@ ssh host 'ls /opt/homebrew/bin/pm2 /usr/local/bin/pm2 ~/.nvm/versions/node/*/bin
 A host here is NOT a deploy target. A rollout names it as skipped and moves on — it is not a failure,
 and it must not be reported as one.

-None today. `edge-pertento-bridge` was here from 2026-08-23 and came back up on 2026-08-24; it is a
-deploy target again and its row is in the table above.
+**Both Pertento hosts, since 2026-09-08** — `production-pertento-bridge` and `edge-pertento-bridge`. Owner's
+call, not a fault found on either box. Details kept here rather than dropped, so putting them back is a
+one-line move, not a re-derivation:

-It was down long enough to matter, so expect it to be behind: when it returned it was 215 commits back at
-`2026.08.22.02`, with no `roles` table and no commanders. A host coming off this list needs its schema
-brought forward before it is treated as ordinary.
+| Host                         | Tailnet IP  | Login        | Port     | Checkout                 | Domain                 |
+| ----------------------------- | ----------- | ------------ | -------- | ------------------------- | ----------------------- |
+| `production-pertento-bridge` | 100.64.0.12 | `pastilhas`  | **9010** | `~/officerdev/platform`  | `officer.pertento.ai`  |
+| `edge-pertento-bridge`       | 100.64.0.15 | `pastilhas`  | 9000     | `~/officerdev/platform`  | `edge.pertento.ai`     |
+
+`edge-pertento-bridge` was skipped once before, from 2026-08-23 to 2026-08-24 — but that was downtime, not
+a choice, and it came back as a deploy target. This is a different reason and there is no return date.
+
+It was down long enough to matter last time, so expect it to be behind whenever it returns: when it came
+back on 2026-08-24 it was 215 commits back at `2026.08.22.02`, with no `roles` table and no commanders. A
+host coming off this list needs its schema brought forward before it is treated as ordinary.

 On 2026-08-23 the deploy agent spent four attempts and up to 60s per try on this host before reporting
 it unreachable — TCP on 22 accepted, then no SSH banner. That work was correct and entirely wasted, which
@@ -1,6 +1,6 @@
 {
  "name": "officer",
-  "version": "2026.08.30.03",
+  "version": "2026.09.09.01",
  "private": true,
  "type": "module",
  "workspaces": [
@@ -44,6 +44,7 @@
  },
  "dependencies": {
    "@anthropic-ai/claude-agent-sdk": "^0.2.41",
+    "@anthropic-ai/sdk": "^0.71.0",
    "@bitcoinerlab/secp256k1": "1.2.0",
    "@getalby/sdk": "^8.0.3",
    "@hookform/resolvers": "^5.2.2",
@@ -32,7 +32,7 @@ import { CATALOGUE } from '../src/servers/app-store/catalogue';
 import { PLATFORM_DIR } from '../src/servers/data-path';
 import { discoverPlugins } from '../src/servers/plugins/discover';
 import { pluginPm2Entry, type Pm2App } from '../src/servers/plugins/ecosystem';
-import { listPluginInstalls, listSidecarInstalls } from 'officerdb';
+import { listPluginInstalls, listPluginPorts, listSidecarInstalls } from 'officerdb';

 const done = (message: string): never => {
  console.log(`[gen:ecosystem] ${message}`);
@@ -92,7 +92,24 @@ const DISABLED = new Set(
 async function pluginEntries(): Promise<Pm2App[]> {
  const [installs, { plugins }] = await Promise.all([listPluginInstalls(), discoverPlugins()]);
  const enabled = new Set(installs.filter((i) => i.enabled).map((i) => i.appName));
-  return plugins.filter((p) => enabled.has(p.appName)).flatMap((p) => pluginPm2Entry(p) ?? []);
+  /*
+    Ports come from the database, not from a fresh allocation.
+
+    This script rebuilds the whole file from what is installed, and it must produce the SAME entries the
+    installer produced — allocating here would hand a plugin a different number than the one already
+    written into somebody's nginx configuration.
+  */
+  const allocations = await listPluginPorts();
+  const byPlugin = new Map<string, Record<string, number>>();
+  for (const row of allocations) {
+    const current = byPlugin.get(row.appName) ?? {};
+    current[row.name] = row.port;
+    byPlugin.set(row.appName, current);
+  }
+
+  return plugins
+    .filter((p) => enabled.has(p.appName))
+    .flatMap((p) => pluginPm2Entry(p, byPlugin.get(p.appName)) ?? []);
 }

 /** Installed AND enabled. Unlike a plugin, every catalogue entry has a `process` — none are filtered out. */
@@ -0,0 +1,46 @@
+#!/usr/bin/env bash
+# Publish every plugin release to the marketplace catalogue. Meant for cron, on the ONE machine that
+# publishes.
+#
+# ── Why a timer at all, when a webhook would be instant ──
+#
+# The sync reads bare mirrors at their newest tag, so running it on a schedule cannot publish anything
+# that was not pushed — which is exactly what made a timer unsafe before, and is no longer true. A Gitea
+# webhook would cut the delay from a minute to a second, and it would also mean a public write route, a
+# shared secret, and a second path into the same code. A minute is not the problem anybody had: the
+# complaint that started this was a release that never appeared at all.
+#
+# ── What it costs ──
+#
+# About 2.7s per tick with nothing to do — one query, one fetch per repository, no writes. Gitea is on
+# this same machine, so none of that leaves the box.
+#
+# ── The log ──
+#
+# One timestamped summary line per tick, ALWAYS, including the ticks that do nothing. A fresh timestamp
+# is the only thing that distinguishes "nothing to publish" from "the timer died". Plugin-level lines
+# appear only for a plugin that actually moved, which is what `--quiet` buys.
+
+set -uo pipefail
+
+PLATFORM="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+LOG="${MARKETPLACE_SYNC_LOG:-$(dirname "$PLATFORM")/marketplace-sync.log}"
+LOCK="/tmp/officer-marketplace-sync.lock"
+
+# Cron's PATH is three entries and bun is in none of them.
+export PATH="$HOME/.bun/bin:/usr/local/bin:/usr/bin:/bin"
+
+cd "$PLATFORM" || exit 1
+
+# Skip rather than queue. A tick that finds the previous one still running has nothing to add — the run
+# in progress will publish whatever this one would have.
+exec 9>"$LOCK"
+flock -n 9 || exit 0
+
+bun --env-file=.env scripts/marketplace-sync.ts --quiet >>"$LOG" 2>&1
+
+# Bounded without logrotate, because this is one file on one machine and a config file for it would be
+# more moving parts than the thing it manages.
+if [ "$(wc -l <"$LOG")" -gt 20000 ]; then
+  tail -n 5000 "$LOG" >"$LOG.tmp" && mv "$LOG.tmp" "$LOG"
+fi
@@ -1,43 +1,40 @@
 #!/usr/bin/env bun
-import { existsSync, readFileSync } from 'node:fs';
-import { basename, join, resolve } from 'node:path';
-import { upsertMetadata, upsertPlugin } from '../src/databases/marketplace_db/src/plugins/queries';
-import { discoverPlugins } from '../src/servers/plugins/discover';
+import { upsertPlugin } from '../src/databases/marketplace_db/src/plugins/queries';
+import { syncAll, syncPlugin } from '../src/servers/marketplace/sync';

-// Fill the marketplace database from the plugin repositories.
-//
-// ── What this reads, and why it is not the marketplace's own opinion ──
-//
-// Every field except "where to clone from" comes out of the plugin's `manifest.ts` and its README. The
-// marketplace stores a COPY so the website can render without cloning three repositories per page load;
-// the manifest stays the source of truth, and the platform re-reads it after cloning. A row here that
-// disagrees with the repository is a stale cache, not a second opinion — which is why re-running this is
-// always safe and never merges: it overwrites.
-//
-// ── Reading the manifest by importing it ──
-//
-// A manifest is TypeScript, not JSON, so it is imported and evaluated rather than parsed. That is code
-// execution, and it is fine for OUR repositories and not fine as a general mechanism.
-//
-// `[blocker for third-party submissions]` A stranger's plugin cannot be imported by the machine that
-// serves the catalogue. Whoever builds submissions has to either evaluate manifests in a sandbox or make
-// authors publish a `manifest.json` alongside. Recorded rather than solved: nobody can submit yet.
-//
-// ── Local clones, for now ──
-//
-// It syncs from `plugins/` on this machine rather than by cloning each repo into a temp directory. That
-// keeps tonight's scope to "get real data into the database" — and it means the machine running the sync
-// must have the plugins checked out. See `docs/marketplace.md` → "Sync" for the version that clones.
+/**
+ * Fill the marketplace catalogue from the plugins' releases.
+ *
+ *   bun --env-file=.env scripts/marketplace-sync.ts                  every registered plugin
+ *   bun --env-file=.env scripts/marketplace-sync.ts photos           just one
+ *   bun --env-file=.env scripts/marketplace-sync.ts --add <url>      register a new one, then sync it
+ *   bun --env-file=.env scripts/marketplace-sync.ts --force          re-publish releases already published
+ *   bun --env-file=.env scripts/marketplace-sync.ts --quiet          print only what changed
+ *
+ * All the reasoning is in `src/servers/marketplace/sync.ts` — this file is the command line around it.
+ * It is a thin wrapper on purpose: the same functions are what the Gitea webhook calls, so a plugin
+ * published by hand and one published by pushing a tag go through exactly the same code.
+ *
+ * Re-running is always safe. It overwrites and never merges.
+ */

-type Sync = { appName: string; developer: string; repoUrl: string };
+const argv = process.argv.slice(2);
+/*
+  `--quiet` is what the timer runs: one line per plugin that actually moved, and a summary. Not silence —
+  the summary is always printed, because a log whose last line is old is the only way to tell "nothing to
+  publish" from "the timer died", and that question does come up.
+*/
+const quiet = argv.includes('--quiet');
+const force = argv.includes('--force');
+const args = argv.filter((a) => a !== '--quiet' && a !== '--force');

 /**
 * A clone URL the whole internet can use.
 *
- * `plugins/example` and `plugins/offscale` have `git@gitea.officer.dev:plugins/<name>.git` remotes,
- * because whoever set them up had a key on that host. Publishing that to a catalogue would hand every
- * installer a URL that only works on the machine it was published from — the install fails at
- * authentication, which reads as "the plugin is broken" rather than "the catalogue is wrong".
+ * Several plugin repos have `git@gitea.officer.dev:plugins/<name>.git` remotes, because whoever set them
+ * up had a key on that host. Registering one of those would hand every installer a URL that only works
+ * on the machine it was published from — the install fails at authentication, which reads as "the plugin
+ * is broken" rather than "the catalogue is wrong".
 */
 export function toHttpsRemote(remote: string): string | null {
  const trimmed = remote.trim();
@@ -60,142 +57,58 @@ export function toHttpsRemote(remote: string): string | null {
  return null;
 }

-const git = (dir: string, args: string[]): string | null => {
-  const result = Bun.spawnSync(['git', '-C', dir, ...args], { stdout: 'pipe', stderr: 'ignore' });
-  if (result.exitCode !== 0) return null;
-  return new TextDecoder().decode(result.stdout).trim() || null;
-};
-
-/*
-  Which plugins directory to publish FROM.
-
-  There is one: `platform/plugins/`, holding what this machine has INSTALLED — updated by
-  `plugins:update`, which is a git pull, so it sits at whatever version was last installed.
-
-  Publishing from it means the catalogue advertises the version this machine happens to be running
-  rather than the version that was released. That is exactly what happened on 2026-08-27: 21 plugins
-  were tagged and the store went on showing 1.0.0.
-
-  A second directory — `officer-dev/plugins/`, every repo in the org cloned side by side — was created
-  to fix that, and never did: its entries were SYMLINKS into this one, so both paths read the same
-  working copy and published the same version. It was removed on 2026-08-28.
-
-  **The problem is still open, and a second checkout was never the answer.** Each directory here is a
-  real git repo with its own tags, so the released manifest can be read out of the tag itself —
-  `git show <newest-tag>:manifest.ts` — without caring what is checked out. That is the fix; it is not
-  written yet, and until it is, this script advertises what is installed.
-*/
-const PLUGINS_DIR = process.env.PLUGINS_DIR ?? resolve(import.meta.dir, '..', 'plugins');
-
-async function syncOne(dir: string): Promise<Sync | null> {
-  const appName = basename(dir);
-  const manifestPath = join(dir, 'manifest.ts');
-  if (!existsSync(manifestPath)) {
-    console.warn(`[marketplace-sync] ${appName}: no manifest.ts — skipped`);
-    return null;
+if (args[0] === '--add') {
+  const raw = args[1];
+  if (!raw) {
+    console.error('usage: marketplace-sync.ts --add <clone url>');
+    process.exit(1);
  }
-
-  const remote = git(dir, ['remote', 'get-url', 'origin']);
-  const repoUrl = remote ? toHttpsRemote(remote) : null;
+  const repoUrl = toHttpsRemote(raw);
  if (!repoUrl) {
-    console.warn(`[marketplace-sync] ${appName}: no usable https remote (${remote ?? 'none'}) — skipped`);
-    return null;
+    console.error(`[marketplace-sync] ${raw} is not a clone URL this can turn into https`);
+    process.exit(1);
+  }
+  const appName = repoUrl.replace(/\.git$/, '').split('/').pop() ?? '';
+  if (!/^[a-z0-9][a-z0-9-]{0,63}$/.test(appName)) {
+    console.error(`[marketplace-sync] "${appName}" is not a usable app name`);
+    process.exit(1);
  }

-  const { manifest } = (await import(manifestPath)) as { manifest: Record<string, unknown> };
-  if (!manifest) {
-    console.warn(`[marketplace-sync] ${appName}: manifest.ts exports no \`manifest\` — skipped`);
-    return null;
+  // Registered with a placeholder developer, then immediately synced — which reads the real publisher
+  // out of the released manifest and overwrites it. A repository with no tag stays registered and
+  // unpublished, which is the correct state for one that has not made a release yet.
+  await upsertPlugin({ appName, developer: 'unknown', repoUrl });
+  console.log(`[marketplace-sync] registered ${appName} → ${repoUrl}`);
+  const one = await syncPlugin(appName, force);
+  report([one]);
+  process.exit(one.status === 'failed' ? 1 : 0);
+}
+
+const results = args[0] ? [await syncPlugin(args[0], force)] : await syncAll(force);
+report(results);
+process.exit(results.some((r) => r.status === 'failed') ? 1 : 0);
+
+function report(results: Awaited<ReturnType<typeof syncAll>>): void {
+  const stamp = new Date().toISOString();
+  for (const r of results) {
+    if (r.status === 'synced') console.log(`[marketplace-sync] ${r.appName} ← ${r.tag}`);
+    else if (r.status === 'failed') console.error(`[marketplace-sync] ${r.appName} failed: ${r.error}`);
+    else if (quiet) continue;
+    else if (r.status === 'unchanged') console.log(`[marketplace-sync] ${r.appName} = ${r.tag}`);
+    else console.warn(`[marketplace-sync] ${r.appName}: no tag — nothing published`);
  }

-  // What the plugin SHIPS, read from the directory rather than declared — the same rule discovery
-  // follows, so a catalogue entry cannot claim a sidecar the repository does not have.
-  const has = {
-    api: existsSync(join(dir, 'api', 'router.ts')),
-    schema: existsSync(join(dir, 'db', 'schema.ts')),
-    sidecar: existsSync(join(dir, 'sidecar', 'index.ts')) || existsSync(join(dir, 'sidecar', 'index.mjs')),
-    web: existsSync(join(dir, 'web')),
-  };
+  const synced = results.filter((r) => r.status === 'synced').length;
+  const unchanged = results.filter((r) => r.status === 'unchanged').length;
+  const unreleased = results.filter((r) => r.status === 'unreleased').map((r) => r.appName);
+  const failed = results.filter((r) => r.status === 'failed').map((r) => r.appName);

-  const readmePath = ['README.md', 'readme.md'].map((n) => join(dir, n)).find((p) => existsSync(p));
-  const readme = readmePath ? readFileSync(readmePath, 'utf-8') : null;
-
-  const developer = String(manifest.publisher ?? 'unknown');
-  const pluginId = await upsertPlugin({ appName, developer, repoUrl });
-
-  await upsertMetadata(pluginId, {
-    version: String(manifest.version ?? '0.0.0'),
-    platform: String(manifest.platform ?? '>=1.0.0'),
-    label: String(manifest.label ?? appName),
-    summary: String(manifest.summary ?? ''),
-    // Taken verbatim, NOT validated here. `manifestProblems` already refuses a tag outside the
-    // vocabulary at discovery, so anything reaching this point is either valid or belongs to a plugin
-    // the platform would not load anyway — and a sync that silently dropped a tag would be a catalogue
-    // quietly disagreeing with the repository it claims to mirror.
-    tags: Array.isArray(manifest.tags) ? manifest.tags : [],
-    color: (manifest.color as string) ?? null,
-    tile: (manifest.tile as string) ?? null,
-    // Artwork lives in the repo, so the URL is built from the remote rather than stored by hand. Only
-    // claimed when the file is actually there — a broken image is worse than a glyph.
-    iconUrl: existsSync(join(dir, 'assets', 'icon.png')) ? rawUrl(repoUrl, 'assets/icon.png') : null,
-    icon: (manifest.icon as string) ?? null,
-    permissions: manifest.permissions ?? [],
-    osDependencies: manifest.osDependencies ?? null,
-    has,
-    readme,
-    sourceCommit: git(dir, ['rev-parse', 'HEAD']),
-  });
-
-  console.log(
-    `[marketplace-sync] ${appName} ← ${repoUrl}${readme ? '' : ' (no README)'}${has.web ? '' : ' (no web/)'}`,
-  );
-  return { appName, developer, repoUrl };
+  console.log(`[marketplace-sync] ${stamp} — ${synced} published, ${unchanged} unchanged, ${unreleased.length} untagged`);
+  // Named rather than counted, and not under `--quiet`: "4 untagged" is a number nobody acts on, but a
+  // timer that reprints the same four names every minute is a log nobody reads. The names belong in a
+  // run somebody asked for.
+  if (unreleased.length && !quiet) {
+    console.warn(`[marketplace-sync] never tagged, so not updated: ${unreleased.join(', ')}`);
+  }
+  if (failed.length) console.error(`[marketplace-sync] failed: ${failed.join(', ')}`);
 }
-
-/**
- * A raw-file URL for a Gitea repository.
- *
- * `HEAD`, not a branch name. This said `branch/main` and the comment here warned that "a repo whose
- * default branch is `master` gets a 404" — which then happened: four plugin repositories were created
- * on `master` and every one of them served a broken image in the marketplace on any server that reads
- * the remote catalogue. A warning in a comment is not a guard.
- *
- * Gitea resolves `HEAD` to whatever the default branch is, so the question stops being asked. Verified
- * against both spellings.
- *
- * `[open]` Still Gitea-shaped. A GitHub-hosted plugin needs the other form, and that wants the forge to
- * be part of the registry rather than inferred from the URL.
- */
-function rawUrl(repoUrl: string, path: string): string {
-  return `${repoUrl.replace(/\.git$/, '')}/raw/HEAD/${path}`;
-}
-
-const dirs = existsSync(PLUGINS_DIR)
-  ? [...new Bun.Glob('*/').scanSync({ cwd: PLUGINS_DIR, onlyFiles: false })]
-      // Not a plugin: the shared install target for plugin npm packages, which lives here because bun
-      // resolves node_modules up the tree. `discoverPlugins` skips it for the same reason.
-      .filter((d) => d.replace(/\/$/, '') !== 'node_modules')
-      .map((d) => join(PLUGINS_DIR, d))
-  : [];
-
-if (dirs.length === 0) {
-  console.log(`[marketplace-sync] no plugin directories under ${PLUGINS_DIR} — nothing to sync`);
-  process.exit(0);
-}
-
-// Discovery is not used to FIND them — a plugin that fails discovery is exactly the kind of thing a
-// catalogue should still be able to describe — but its broken list is worth surfacing here.
-const { broken } = await discoverPlugins().catch(() => ({ broken: [] as { appName: string; error: string }[] }));
-for (const { appName, error } of broken) console.warn(`[marketplace-sync] ${appName} is broken on disk: ${error}`);
-
-const synced: Sync[] = [];
-for (const dir of dirs) {
-  const result = await syncOne(dir).catch((err) => {
-    console.error(`[marketplace-sync] ${basename(dir)} failed: ${err instanceof Error ? err.message : err}`);
-    return null;
-  });
-  if (result) synced.push(result);
-}
-
-console.log(`[marketplace-sync] ${synced.length}/${dirs.length} plugin(s) in the catalogue`);
-process.exit(0);
@@ -97,20 +97,34 @@ export function App() {
                404 on a route that worked yesterday reads as the feature having been removed. */}
            <Route path="/plugins" element={<PluginsRedirect />} />
            {/* Installed plugins. Core routes above stay hand-written; everything below is generated from
-                what is installed, because a bundler cannot follow a runtime import specifier. The wildcard
-                hands the whole subtree to the plugin's own router, which react-router nests natively. */}
-            {/* One pair per SCREEN, not per plugin — `screenId` is the key, because a plugin with two
+                what is installed, because a bundler cannot follow a runtime import specifier. */}
+            {/* One route per SCREEN, not per plugin — `screenId` is the key, because a plugin with two
                screens (caldav: Calendar and Contacts) contributes two entries sharing one `appName`. */}
+            {/*
+              Three routes, and the third is the only new one.
+
+              `/<route>` and `/<route>/:section` are what every plugin has always had, and fifteen of
+              them read `useParams().section`. A splat ALONE replaced that for a day and took the
+              parameter away from all of them at once — the screens still mounted and every one of them
+              lost its selection, which is a worse failure than a 404 because nothing says so.
+
+              The splat is added for depth rather than instead of them: `/fofos-admin/layout/header` was
+              not expressible before, and a plugin that wants its own subtree reads the location.
+              React-router ranks a static or dynamic segment above a splat, so a plugin with one segment
+              still gets `section` and only deeper paths fall through.
+            */}
            {installedPlugins.flatMap((plugin) => [
              <Route key={plugin.screenId} path={plugin.route} element={<Dashboard.PluginScreen {...plugin} />} />,
-              // The section pair, exactly as the core screens do it (`/soulseek/:section`): the plugin's
-              // panels read `useParams` themselves, so which section is open is the URL rather than state
-              // passed between them.
              <Route
                key={`${plugin.screenId}-section`}
                path={`${plugin.route}/:section`}
                element={<Dashboard.PluginScreen {...plugin} />}
              />,
+              <Route
+                key={`${plugin.screenId}-deep`}
+                path={`${plugin.route}/*`}
+                element={<Dashboard.PluginScreen {...plugin} />}
+              />,
            ])}
            <Route path="/system-monitor" element={<Dashboard.SystemMonitorScreen />} />
            <Route path="/system-monitor/:scope" element={<Dashboard.SystemMonitorScreen />} />
@@ -14,6 +14,14 @@ export type CatalogueEntry = {
  appName: string;
  publisher: string;
  version: string;
+  /**
+   * The git tag this entry was published from.
+   *
+   * Optional on the wire, not in the database, and the two mean different things. Here it is absent for
+   * a row that predates release-based publishing OR for a catalogue too old to send the field at all —
+   * so a consumer must treat "no tag" as "cannot tell" and fall back to whatever it knew before.
+   */
+  releaseTag?: string;
  platform: string;
  label: string;
  summary: string;
@@ -52,6 +60,7 @@ function toCatalogueEntry(row: Row): CatalogueEntry | null {
    appName: row.appName,
    publisher: row.developer,
    version: m.version,
+    ...(m.releaseTag ? { releaseTag: m.releaseTag } : {}),
    platform: m.platform,
    label: m.label,
    summary: m.summary,
@@ -106,6 +115,36 @@ export async function catalogueEntry(appName: string): Promise<CatalogueEntry |
  return { ...entry, ...(row.metadata.readme ? { readme: row.metadata.readme } : {}) };
 }

+/**
+ * A registry row, plus the one piece of metadata the sync needs before deciding to do any work.
+ *
+ * `sourceCommit` is what the last sync published, and it is here rather than fetched per plugin so a
+ * run that has nothing to do costs one query instead of twenty-five. Null means never synced.
+ */
+export type RegistryRow = { id: number; appName: string; repoUrl: string; sourceCommit: string | null };
+
+/**
+ * Every registered plugin — identity and clone URL, no metadata.
+ *
+ * This is what the sync iterates, and it is the reason the registry is called authoritative: the
+ * catalogue lists what somebody deliberately registered, not what happens to be checked out on the
+ * machine that runs the job. Unlisted rows are included; hiding a plugin from the public API is not the
+ * same as no longer tracking it, and one taken down for a bad release has to be able to come back
+ * current.
+ */
+export async function listRegistry(): Promise<RegistryRow[]> {
+  return marketplaceDb
+    .select({
+      id: plugins.id,
+      appName: plugins.appName,
+      repoUrl: plugins.repoUrl,
+      sourceCommit: pluginMetadata.sourceCommit,
+    })
+    .from(plugins)
+    .leftJoin(pluginMetadata, eq(pluginMetadata.pluginId, plugins.id))
+    .orderBy(plugins.appName);
+}
+
 // ── Writes. Used by the sync script, never by the public API ──

 type RegisterParams = { appName: string; developer: string; repoUrl: string };
@@ -73,6 +73,19 @@ export const pluginMetadata = pgTable(
      .references(() => plugins.id, { onDelete: 'cascade' }),

    version: text('version').notNull(),
+    /**
+     * The git tag this row was read from — what "published" actually means.
+     *
+     * `version` above is the manifest's own claim, and the two are the same string every time somebody
+     * tags correctly. They are kept apart because when they DISAGREE the tag is the fact: it is what was
+     * pushed, and a manifest whose version nobody bumped would otherwise make a real release invisible
+     * to anything comparing versions.
+     *
+     * Nullable because rows existed before releases were the unit of publishing. A null one is a
+     * snapshot taken from a working tree by the old sync, and every consumer should read it as "unknown"
+     * rather than "no release".
+     */
+    releaseTag: text('release_tag'),
    /** The platform range it declares support for, e.g. `>=1.0.0`. */
    platform: text('platform').notNull(),
    label: text('label').notNull(),
@@ -1,6 +1,6 @@
-import { eq } from 'drizzle-orm';
+import { and, eq } from 'drizzle-orm';
 import { db } from '../db';
-import { pluginInstalls } from './schema';
+import { pluginInstalls, pluginPorts } from './schema';

 export type PluginInstall = typeof pluginInstalls.$inferSelect;

@@ -47,3 +47,90 @@ export async function removePluginInstall(appName: string): Promise<boolean> {
  const rows = await db.delete(pluginInstalls).where(eq(pluginInstalls.appName, appName)).returning();
  return rows.length > 0;
 }
+
+// ── Ports the platform hands out and remembers ────────────────────────────────────────────────────
+
+export type PluginPort = typeof pluginPorts.$inferSelect;
+
+/** Every allocation, for the settings screen and for anyone asking what is spoken for. */
+export async function listPluginPorts(): Promise<PluginPort[]> {
+  return db.select().from(pluginPorts).orderBy(pluginPorts.port);
+}
+
+/** What this plugin has been given, as `{ name: port }`. Empty when it has asked for nothing. */
+export async function getPluginPorts(appName: string): Promise<Record<string, number>> {
+  const rows = await db.select().from(pluginPorts).where(eq(pluginPorts.appName, appName));
+  return Object.fromEntries(rows.map((r) => [r.name, r.port]));
+}
+
+/**
+ * Find a port nothing is using, by asking the kernel and then checking our own book.
+ *
+ * Binding `port: 0` is the only honest way to know a port is free — a list of what we have handed out
+ * says nothing about what else on this machine is listening, and a hardcoded range says nothing about
+ * either. So the kernel picks, we let go, and the caller binds it a moment later.
+ *
+ * ── The gap between letting go and binding ──
+ *
+ * Real, and unavoidable without holding the socket open across a process boundary. Something else can
+ * take the port in between. That is why allocation is not the last word: a sidecar that cannot bind its
+ * remembered port must say so loudly rather than quietly choosing another, because the number is written
+ * down in configuration this database cannot see.
+ */
+async function freePort(taken: Set<number>): Promise<number> {
+  for (let attempt = 0; attempt < 20; attempt++) {
+    const probe = Bun.serve({ port: 0, hostname: '127.0.0.1', fetch: () => new Response('') });
+    const port = probe.port;
+    probe.stop(true);
+    if (port && !taken.has(port)) return port;
+  }
+  throw new Error('could not find a free port after 20 attempts');
+}
+
+/**
+ * The port this plugin was given for `name`, allocating one the first time.
+ *
+ * Idempotent and stable: the same plugin and name always gets the same number back, which is the entire
+ * point — see the note on `pluginPorts`. Re-installing, upgrading and restarting all return what was
+ * given the first time.
+ */
+export async function allocatePluginPort(appName: string, name: string): Promise<number> {
+  const [existing] = await db
+    .select()
+    .from(pluginPorts)
+    .where(and(eq(pluginPorts.appName, appName), eq(pluginPorts.name, name)));
+  if (existing) return existing.port;
+
+  /*
+    Racing another install is possible and is handled by the unique index rather than by a lock: two
+    installs running at once can be handed the same number by two different kernel probes, and exactly
+    one of the inserts will win. The loser tries again with a fresh number rather than failing, because
+    "somebody else installed a plugin at the same moment" is not a reason to refuse.
+  */
+  for (let attempt = 0; attempt < 5; attempt++) {
+    const taken = new Set((await db.select().from(pluginPorts)).map((r) => r.port));
+    const port = await freePort(taken);
+    try {
+      const [row] = await db.insert(pluginPorts).values({ appName, name, port }).returning();
+      return row!.port;
+    } catch {
+      const [now] = await db
+        .select()
+        .from(pluginPorts)
+        .where(and(eq(pluginPorts.appName, appName), eq(pluginPorts.name, name)));
+      if (now) return now.port; // the other writer was allocating the same thing
+    }
+  }
+  throw new Error(`could not allocate a port for ${appName}/${name}`);
+}
+
+/**
+ * Give up a plugin's ports. NOT called on uninstall — see the note in schema.ts.
+ *
+ * The number lives in configuration this database cannot see, so handing it to something else because a
+ * plugin was briefly removed points that configuration at the wrong service. Only `purge` does this.
+ */
+export async function releasePluginPorts(appName: string): Promise<number> {
+  const rows = await db.delete(pluginPorts).where(eq(pluginPorts.appName, appName)).returning();
+  return rows.length;
+}
@@ -1,4 +1,4 @@
-import { pgTable, serial, text, boolean, timestamp, uniqueIndex } from 'drizzle-orm/pg-core';
+import { pgTable, serial, text, integer, boolean, timestamp, uniqueIndex } from 'drizzle-orm/pg-core';

 // Which plugins are installed on this machine, and whether they should be mounted.
 //
@@ -56,3 +56,57 @@ export const pluginInstalls = pgTable(
    uniqueIndex('uq_plugin_installs_app_name').on(t.appName),
  ],
 );
+
+/**
+ * Ports the platform has given out to plugins, and remembers.
+ *
+ * ── Why the platform allocates these when it allocates nothing else ──
+ *
+ * A sidecar's own port is not allocated: it binds `port: 0`, the kernel picks one, and it announces the
+ * number over the registration socket. That is right for what it is — a private detail between two
+ * processes, on loopback, re-chosen every restart, and nothing outside Officer ever needs to know it.
+ *
+ * A port something ELSE connects to is a different thing entirely. A plugin serving a site has to be
+ * reachable from nginx, which means the number must be stable across restarts and must be knowable by a
+ * human writing a proxy configuration. Ephemeral fails both.
+ *
+ * Doing it here rather than in each plugin's own settings buys the one thing per-plugin config cannot:
+ * **two plugins cannot be given the same number**, and the platform knows which are spoken for. That is
+ * `uq_plugin_ports_port` below, and it is the whole argument for this table.
+ *
+ * ── Kept when a plugin is uninstalled ──
+ *
+ * Deliberate. The number is written down somewhere this database cannot see — an nginx proxy host, a
+ * firewall rule, somebody's notes — so handing it to a different plugin because this one was briefly
+ * removed would point that configuration at the wrong service. Reinstalling gets the same port back.
+ * `purge` is the deliberate act that gives it up.
+ */
+export const pluginPorts = pgTable(
+  'plugin_ports',
+  {
+    id: serial('id').primaryKey(),
+    /** The plugin's app name — its directory, as everywhere else. */
+    appName: text('app_name').notNull(),
+    /**
+     * Which of the plugin's ports this is, as the manifest named it.
+     *
+     * A plugin may want more than one — a site and a preview of it — so the number alone is not an
+     * identity. Reaches the sidecar as `PORT_<NAME>` in its environment, uppercased.
+     */
+    name: text('name').notNull(),
+    port: integer('port').notNull(),
+    allocatedAt: timestamp('allocated_at', { withTimezone: true }).notNull().defaultNow(),
+  },
+  (t) => [
+    // One allocation per named port per plugin.
+    uniqueIndex('uq_plugin_ports_app_name_name').on(t.appName, t.name),
+    /*
+      No two plugins on the same number, said structurally.
+
+      This is the reason the table exists rather than a convention people follow. A collision here is not
+      a duplicate row, it is two services silently fighting over a listener, where the loser fails to bind
+      at startup and the winner is whichever happened to start first.
+    */
+    uniqueIndex('uq_plugin_ports_port').on(t.port),
+  ],
+);
@@ -0,0 +1,50 @@
+import Anthropic from '@anthropic-ai/sdk';
+import { ANTHROPIC_PROXY_URL } from './officer-url.mjs';
+import { readProxySecretFromDisk } from './sidecar/claude/state';
+
+/**
+ * The sanctioned way for PLATFORM-SIDE code — including a plugin's server routes — to call Claude.
+ *
+ * ── Why this module exists ──
+ *
+ * `officer-anthropic-proxy` swaps a local secret for the owner's OAuth bearer, and that secret lives
+ * in `sidecar/claude-state.json` — platform-internal state whose location, format and permissions are
+ * this repo's to change. The first caller outside the sidecar (fofos-website's translate route) read
+ * the file directly, which worked and meant the sidecar could no longer move its own state without
+ * breaking a plugin. This module is the boundary: plugins import IT, and only platform code touches
+ * the file.
+ *
+ * ── Why a module and not an HTTP route ──
+ *
+ * Every caller so far is server-side code in this same process, already behind its own gate. An HTTP
+ * endpoint that spends the owner's subscription would be NEW surface — reachable by anything holding a
+ * session, needing its own registry entry and permission — bought for no caller that exists. The day
+ * a browser-side caller appears, the route is a thin wrapper over this plus that registry entry.
+ *
+ * ── The secret is read per call ──
+ *
+ * A fresh install writes it through a 30s debounce, so anything that cached at boot could hold an
+ * empty string forever. Reading late costs one small file read and removes the failure mode.
+ */
+
+/**
+ * A client pointed at the proxy, or null when the secret is not on disk yet.
+ *
+ * Null rather than a throw, because "not yet" is an ordinary state on a fresh install — the caller
+ * says which process is missing, in its own words, with its own status code.
+ */
+/*
+  Re-exported so a caller's `instanceof` works.
+
+  A plugin that imported '@anthropic-ai/sdk' itself could resolve a SECOND copy from the plugins'
+  own node_modules — and an error thrown by the platform's copy is not an instance of the other
+  one's classes, so `catch (e) { e instanceof RateLimitError }` silently stops matching. One import
+  path, one copy, one class identity.
+*/
+export { default as Anthropic } from '@anthropic-ai/sdk';
+
+export function anthropicViaProxy(): Anthropic | null {
+  const secret = readProxySecretFromDisk();
+  if (!secret) return null;
+  return new Anthropic({ apiKey: secret, baseURL: ANTHROPIC_PROXY_URL });
+}
@@ -1,5 +1,5 @@
 import { describe, expect, test } from 'bun:test';
-import { proxyOverrideScript, rewriteAssetPaths, rewriteHtml } from './html-rewrite';
+import { proxyOverrideScript, rewriteAssetPaths, rewriteCss, rewriteHtml } from './html-rewrite';

 const BASE = '/api/dev-proxy/3001';
 const attrs = (html: string) => [...html.matchAll(/(?:src|href)="([^"]+)"/g)].map((m) => m[1]!);
@@ -58,6 +58,43 @@ describe('rewriteHtml', () => {
    expect(out.indexOf('<script>')).toBeLessThan(out.indexOf('<title>'));
  });

+  test('every candidate in a srcset is prefixed, descriptors intact', () => {
+    expect(rewriteHtml('<img srcset="/a.webp 96w, /b.webp 320w">', BASE)).toContain(
+      `srcset="${BASE}/a.webp 96w, ${BASE}/b.webp 320w"`,
+    );
+  });
+
+  /*
+    The failure this was found by. An <img> with a srcset also has a src, and the src WAS prefixed — but
+    a browser given a valid srcset picks from it and never reads src. So the logo was missing while the
+    markup looked right.
+  */
+  test('a srcset is chosen over src by the browser, so prefixing src alone is not enough', () => {
+    const out = rewriteHtml('<img src="/logo.webp" srcset="/logo.webp 96w">', BASE);
+    expect(out).toContain(`src="${BASE}/logo.webp"`);
+    expect(out).toContain(`srcset="${BASE}/logo.webp 96w"`);
+  });
+
+  test('a candidate with no descriptor is prefixed too', () => {
+    expect(rewriteHtml('<img srcset="/only.webp">', BASE)).toContain(`srcset="${BASE}/only.webp"`);
+  });
+
+  test('imagesrcset on a preload link is prefixed — the one that is easy to miss', () => {
+    expect(rewriteHtml('<link rel="preload" as="image" imagesrcset="/hero.webp 1280w">', BASE)).toContain(
+      `imagesrcset="${BASE}/hero.webp 1280w"`,
+    );
+  });
+
+  test('a protocol-relative candidate is left alone, like everywhere else here', () => {
+    expect(rewriteHtml('<img srcset="//cdn.example.com/a.webp 96w, /b.webp 320w">', BASE)).toContain(
+      `srcset="//cdn.example.com/a.webp 96w, ${BASE}/b.webp 320w"`,
+    );
+  });
+
+  test('`..` is collapsed in a srcset candidate as well', () => {
+    expect(rewriteHtml('<img srcset="/../../x.webp 96w">', BASE)).toContain(`srcset="${BASE}/x.webp 96w"`);
+  });
+
  test('a query string and hash survive normalisation', () => {
    expect(attrs(rewriteHtml('<script src="/../x.js?v=2#frag">', BASE))).toEqual(['/api/dev-proxy/3001/x.js?v=2#frag']);
  });
@@ -175,3 +212,42 @@ describe('the injected location reporter', () => {
    expect(r.posted).toHaveLength(2);
  });
 });
+
+describe('stylesheets', () => {
+  test('a quoted url is prefixed — the form every generator emits', () => {
+    expect(rewriteCss('@font-face{src:url("/fonts/Averta.woff2") format("woff2")}', BASE)).toContain(
+      `url("${BASE}/fonts/Averta.woff2")`,
+    );
+  });
+
+  test('single quotes and bare urls too', () => {
+    expect(rewriteCss("a{background:url('/bg.png')}", BASE)).toContain(`url('${BASE}/bg.png')`);
+    expect(rewriteCss('a{background:url(/bg.png)}', BASE)).toContain(`url(${BASE}/bg.png)`);
+  });
+
+  test('the quote style is preserved, not normalised', () => {
+    expect(rewriteCss('a{background:url("/bg.png")}', BASE)).not.toContain("url('");
+  });
+
+  test('a url that is already absolute is left alone', () => {
+    const external = 'a{background:url("https://cdn.example.com/bg.png")}';
+    expect(rewriteCss(external, BASE)).toBe(external);
+    const protocolRelative = 'a{background:url("//cdn.example.com/bg.png")}';
+    expect(rewriteCss(protocolRelative, BASE)).toBe(protocolRelative);
+  });
+
+  test('a relative url is left alone — it already resolves against the stylesheet', () => {
+    const relative = 'a{background:url("./bg.png")}';
+    expect(rewriteCss(relative, BASE)).toBe(relative);
+  });
+
+  test('a stylesheet still gets the /_bun/ rule', () => {
+    expect(rewriteCss('@import "/_bun/x.css";', BASE)).toContain(`${BASE}/_bun/x.css`);
+  });
+
+  test('inline CSS in HTML gets the same treatment, quotes included', () => {
+    expect(rewriteHtml('<div style=\'background:url("/a.png")\'></div>', BASE)).toContain(
+      `url("${BASE}/a.png")`,
+    );
+  });
+});
@@ -89,7 +89,38 @@ const normalisePath = (path: string): string => {
 /** `/x` → `<base>/x`, with `..` already collapsed. Never touches `//host` or a full URL. */
 const prefix = (base: string, path: string): string => `${base}${normalisePath(path)}`;

-// HTML: rewrite src/href/action attributes with absolute paths + inject fetch/XHR override
+/**
+ * A `srcset`, which is a LIST and cannot be prefixed as one value.
+ *
+ * `srcset="/a.webp 96w, /b.webp 2x"` — comma-separated candidates, each a URL plus an optional
+ * descriptor. The attribute rule below matches a whole attribute value, so it can only ever have
+ * rewritten the first URL and left the descriptor mangled onto it; it does not match `srcset` at all,
+ * which is why responsive images 404'd silently.
+ *
+ * Silently is the operative word. A `<picture>` or an `<img srcset>` still has a `src`, and that one WAS
+ * prefixed — but a browser given a valid srcset chooses from it and never looks at src. So the markup
+ * appeared correct, one attribute pointed at the platform root, and the image was simply missing.
+ *
+ * Split on commas rather than parsed: a URL cannot contain an unescaped comma in this position, and the
+ * descriptor never does.
+ */
+const prefixSrcset = (base: string, list: string): string =>
+  list
+    .split(',')
+    .map((candidate) => {
+      const trimmed = candidate.trim();
+      if (!trimmed) return '';
+      // The URL is everything up to the first space; the rest is `96w`, `2x`, or nothing.
+      const space = trimmed.search(/\s/);
+      const url = space === -1 ? trimmed : trimmed.slice(0, space);
+      const descriptor = space === -1 ? '' : trimmed.slice(space);
+      // Same rule as everywhere else here: root-absolute only, never protocol-relative.
+      return url.startsWith('/') && !url.startsWith('//') ? `${prefix(base, url)}${descriptor}` : trimmed;
+    })
+    .filter(Boolean)
+    .join(', ');
+
+// HTML: rewrite src/href/action/srcset attributes with absolute paths + inject fetch/XHR override
 export const rewriteHtml = (html: string, base: string) =>
  html
    .replace(/<head([^>]*)>/i, `<head$1>${proxyOverrideScript(base)}`)
@@ -97,12 +128,52 @@ export const rewriteHtml = (html: string, base: string) =>
      /(href|src|action)=("|')(\/(?!\/)[^"']*)\2/gi,
      (_match, attr: string, quote: string, path: string) => `${attr}=${quote}${prefix(base, path)}${quote}`,
    )
-    .replace(/url\((\/(?!\/)[^)"']*)\)/g, (_match, path: string) => `url(${prefix(base, path)})`);
+    /*
+      `srcset` and `imagesrcset`, which the attribute rule above cannot do — see prefixSrcset.

-// JS/CSS: only `/_bun/` paths.
+      `imagesrcset` is the one on `<link rel="preload" as="image">`. It is easy to miss and expensive
+      to miss: a preload that 404s costs a request AND leaves the real image undiscovered until the
+      body is parsed, which is the opposite of what the preload was added for.
+    */
+    .replace(
+      /(imagesrcset|srcset)=("|')([^"']*)\2/gi,
+      (_match, attr: string, quote: string, list: string) => `${attr}=${quote}${prefixSrcset(base, list)}${quote}`,
+    )
+    .replace(CSS_URL, cssUrlPrefixer(base));
+
+/*
+  `url(/…)` — a font, a background image, an `@import`. CSS allows the target bare, single-quoted or
+  double-quoted, and all three mean the same thing, so all three have to match. The quote used is
+  captured and put back rather than normalised: rewriting is not the place to have an opinion about
+  someone else's stylesheet.
+
+  This rule missed every QUOTED url until 2026-08-31, which is a bad way to fail. `url(/x.png)` was
+  rewritten and `url("/x.png")` was not, so whether a stylesheet worked through the proxy came down to
+  the formatting preference of whatever generated it.
+*/
+const CSS_URL = /url\(\s*(["']?)(\/(?!\/)[^)"']*)\1\s*\)/g;
+
+const cssUrlPrefixer =
+  (base: string) =>
+  (_match: string, quote: string, path: string): string =>
+    `url(${quote}${prefix(base, path)}${quote})`;
+
+// JS: only `/_bun/` paths.
 //
 // Deliberately narrow. A blanket rewrite of every absolute path inside JavaScript hits strings that are
 // not URLs — a regex in the source, a route table, anything containing a slash — and corrupts the file it
 // is trying to fix. Bun's dev server emits its chunk imports under `/_bun/`, which is specific enough to
 // match safely. A Vite app instead wants `base` set in its own config; nothing here can guess that.
 export const rewriteAssetPaths = (text: string, base: string) => text.replace(/\/_bun\//g, `${base}/_bun/`);
+
+/*
+  A STYLESHEET gets more than that, because a stylesheet's absolute paths are unambiguous: `url(/…)` is
+  a URL and nothing else, so the argument for being narrow in JavaScript does not apply here.
+
+  This was the whole of it for a long time, and the symptom was specific: a site's own `@font-face` rules
+  kept pointing at `/fonts/…`, which inside the frame is the PLATFORM's root, so every face 404'd and the
+  preview quietly rendered in a system fallback. Nothing looked broken — it just looked like a different
+  site than the one being previewed, which is the one thing a preview must never do.
+*/
+export const rewriteCss = (text: string, base: string) =>
+  rewriteAssetPaths(text, base).replace(CSS_URL, cssUrlPrefixer(base));
@@ -1,7 +1,7 @@
 import { createRouter } from '../../create-router';
 import { resolveAuthToken } from '../../auth-token';
 import { isSuperAdmin } from '../../super-admin';
-import { rewriteAssetPaths, rewriteHtml } from './html-rewrite';
+import { rewriteAssetPaths, rewriteCss, rewriteHtml } from './html-rewrite';

 /**
 * See a dev server running on THIS machine's loopback, from a browser that is somewhere else.
@@ -145,7 +145,8 @@ devProxyRouter.all('/:port/*', async (ctx) => {

  const type = res.headers.get('content-type') ?? '';
  const isHtml = type.includes('text/html');
-  const rewritable = isHtml || type.includes('javascript') || type.includes('text/css');
+  const isCss = type.includes('text/css');
+  const rewritable = isHtml || isCss || type.includes('javascript');

  const headers = new Headers(res.headers);
  // Anything framed here is the owner's own dev server; the platform frames it deliberately.
@@ -169,7 +170,11 @@ devProxyRouter.all('/:port/*', async (ctx) => {
  }

  const text = await res.text();
-  const body = isHtml ? rewriteHtml(text, prefix) : rewriteAssetPaths(text, prefix);
+  const body = isHtml
+    ? rewriteHtml(text, prefix)
+    : isCss
+      ? rewriteCss(text, prefix)
+      : rewriteAssetPaths(text, prefix);
  // Both are wrong after a rewrite: the length changed, and the body was decoded on the way through.
  headers.delete('content-length');
  headers.delete('content-encoding');
@@ -1,6 +1,7 @@
 import { createRouter } from '@@/create-router';
 import { resolve, dirname, join, sep, parse as parsePath } from 'node:path';
-import { readdir, stat, mkdir, rm, rename, readFile, cp } from 'node:fs/promises';
+import { readdir, lstat, readlink, stat, mkdir, rm, rename, readFile, cp } from 'node:fs/promises';
+import type { Dirent } from 'node:fs';
 import { existsSync } from 'node:fs';
 import { getOwnerHomeDir, DATA_PATH } from '@@/data-path';
 import { resolveHomeDir } from '@@/user-home';
@@ -188,9 +189,17 @@ router.get('/ls', async (ctx) => {
    await cleanOldCacheDirs(rootDir);
  }

-  let names: string[];
+  /*
+    `withFileTypes`, so a symlink can be spotted without a second syscall per entry.
+
+    The kind comes off the directory entry itself on every filesystem that reports it, and where it
+    does not Node falls back to an lstat it would otherwise have had to make anyway. `stat` below
+    FOLLOWS the link — which is right, and is why it cannot answer this question: a symlink to a folder
+    must still say `directory` so that opening it works.
+  */
+  let names: Dirent[];
  try {
-    names = await readdir(absPath);
+    names = await readdir(absPath, { withFileTypes: true });
  } catch (ex) {
    // A missing directory resets the browser to the root, which is the right answer for a stale path.
    //
@@ -213,18 +222,44 @@ router.get('/ls', async (ctx) => {
  const claimed = await pluginFileTypes();

  const entries = await Promise.all(
-    names.map(async (name) => {
+    names.map(async (entry) => {
+      const name = entry.name;
      const fullPath = resolve(absPath, name);
      // Skip entries that escape the home dir (shouldn't happen but be safe)
      if (!isInside(rootDir, fullPath)) return null;
+
+      /*
+        Where a link POINTS, exactly as it was written.
+
+        Not resolved to an absolute path: a relative target is a fact about the link, and rewriting it
+        would hide that `../originals` moves with the folder while `/home/…/originals` does not. Only
+        asked for the entries that are links, which in an ordinary folder is none of them.
+      */
+      const target = entry.isSymbolicLink() ? await readlink(fullPath).catch(() => null) : null;
+
      const s = await stat(fullPath).catch(() => null);
-      if (!s) return null;
-      const isDirectory = s.isDirectory();
+
+      /*
+        A BROKEN link still gets a row, and did not before.
+
+        `stat` follows, so a link whose target is gone fails here and the entry returned null — the file
+        vanished from the listing entirely. Nothing said it was there, nothing could delete it, and the
+        folder simply had fewer things in it than `ls` showed. `lstat` describes the link itself, which
+        exists whatever it points at.
+      */
+      const own = s ?? (target !== null ? await lstat(fullPath).catch(() => null) : null);
+      if (!own) return null;
+
+      const isDirectory = s?.isDirectory() ?? false;
      return {
        name,
        type: isDirectory ? 'directory' : 'file',
-        size: s.size,
-        modifiedAt: s.mtimeMs,
+        size: own.size,
+        modifiedAt: own.mtimeMs,
+        /** Where it points, or absent when it is an ordinary file. See the note above. */
+        symlink: target,
+        /** A link whose target is gone. `type` is `file` for want of anything truer to say. */
+        broken: target !== null && s === null ? true : undefined,
        /*
          Whether `/thumb` will produce something, decided HERE so there is one list of extensions
          rather than one on each side of the wire.
@@ -4,7 +4,7 @@ import { isSuperAdmin } from '../../super-admin';
 import { mountPrefix } from '../../plugins/manifest';
 import { snapshotPlugins } from '../../plugins/mount';
 import { manualInstallHint, reportDependencies } from '../../plugins/os-deps';
-import { fetchCatalogue, fetchReadme } from '../../plugins/marketplace';
+import { fetchCatalogue, fetchReadme, publishedTags } from '../../plugins/marketplace';
 import { findOrphanTables } from '../../plugins/orphans';
 import { fetchPlugin } from '../../plugins/fetch';
 import { discoverPlugins } from '../../plugins/discover';
@@ -57,7 +57,9 @@ pluginsRouter.get('/catalogue/:appName/readme', async (ctx) => {
 });

 pluginsRouter.get('/', async (ctx) => {
-  const { states, broken } = await snapshotPlugins();
+  // The catalogue, for the update badge — cached, never throws, and falls back to the git rule per
+  // plugin when the store has nothing to say. See `plugins/release.ts`.
+  const { states, broken } = await snapshotPlugins(await publishedTags());
  // Asked per plugin rather than once, because `pm2 jlist` is a fork and most plugins have no sidecar to
  // ask about. A plugin that is installed and enabled but whose process is not online is the state worth
  // rendering differently — it is the difference between "off" and "broken".
@@ -59,7 +59,23 @@ export async function isKnownToPm2(name: string, cwd: string): Promise<boolean>
 */
 export async function startProcess(name: string, cwd: string): Promise<Pm2Result> {
  const known = await isKnownToPm2(name, cwd);
-  const args = known ? ['start', name] : ['start', ECOSYSTEM, '--only', name];
+  /*
+    Through the ECOSYSTEM FILE either way, and that is the whole point of this line.
+
+    `pm2 start <name>` restarts a known process with the environment PM2 already holds for it, and never
+    looks at the file again. That was fine while an entry was just a script path — it is not now that an
+    entry carries `env`, because a plugin whose allocated port changed would have its new number written
+    to the file and go on running with the old one.
+
+    Seen exactly once, on the day ports were added: the ecosystem said PORT_SITE=34507, the process was
+    serving 3011, and nothing anywhere reported a disagreement.
+
+    `restart <file> --only <name>` re-reads the entry; `--update-env` makes PM2 replace the stored
+    environment rather than merge into it, so a variable REMOVED from the entry actually goes away.
+  */
+  const args = known
+    ? ['restart', ECOSYSTEM, '--only', name, '--update-env']
+    : ['start', ECOSYSTEM, '--only', name];

  const { code, out } = await pm2(args, cwd);
  if (code !== 0) return { ok: false, error: out || `pm2 ${args.join(' ')} exited ${code}` };
@@ -18,9 +18,14 @@ import { catalogueEntry, listCatalogue } from '../../databases/marketplace_db/sr
 *
 * ── Read-only, and structurally so ──
 *
- * There are two GET routes and no others. Registering a plugin is `scripts/marketplace-sync.ts`, run by
- * hand against the database. That is not a temporary shortcut standing in for a write API: while
- * submissions do not exist, a public write surface is all risk and no feature.
+ * There are two GET routes and no others. Everything that WRITES the catalogue — registering a plugin,
+ * publishing a release — is `scripts/marketplace-sync.ts` against the database, on the one machine that
+ * publishes. That is not a temporary shortcut standing in for a write API: while submissions do not
+ * exist, a public write surface is all risk and no feature.
+ *
+ * It stayed that way when publishing went onto a timer, which is the moment it would have been easy to
+ * add a `POST /api/sync` for a Gitea webhook. A minute of delay is cheaper than a write route on the one
+ * process here that has no login.
 *
 * ── CORS is wide open, deliberately ──
 *
@@ -0,0 +1,301 @@
+import { existsSync, mkdirSync } from 'node:fs';
+import { homedir, tmpdir } from 'node:os';
+import { join } from 'node:path';
+import {
+  listRegistry,
+  type RegistryRow,
+  upsertMetadata,
+  upsertPlugin,
+} from '../../databases/marketplace_db/src/plugins/queries';
+
+/**
+ * Fill the catalogue from what each plugin has RELEASED.
+ *
+ * ── The bug this file exists to remove ──
+ *
+ * Until 2026-09-03 the sync read `platform/plugins/` — the directories this machine has INSTALLED — and
+ * imported each `manifest.ts` off the working tree. So "publish" meant "make sure officer-dev's checkout
+ * happens to be at the release, then run the script", and nobody could see when it was not. On the day
+ * this was rewritten the live catalogue was advertising four plugins from untagged working copies:
+ * av-club 2026.08.29.01, kanban 2026.08.29.02, openseo 0.1.0 and fofos-admin 2026.08.29.26, none of
+ * which had ever been tagged. The store was describing somebody's afternoon.
+ *
+ * That is also why the old script could not be put on a timer: a cron over it would have published
+ * whatever was sitting in a working tree, on a schedule.
+ *
+ * ── What replaces it: a bare mirror, read at a tag ──
+ *
+ * Every plugin in the registry is mirrored (`git clone --mirror`) into a cache directory, refreshed, and
+ * read AT ITS NEWEST TAG — the manifest, the README, the file list, all out of `git cat-file`. There is
+ * no working tree anywhere in this path, so publishing something uncommitted is not a mistake that can
+ * be made. A mirror is cheap: the whole set clones in a couple of seconds and refreshes in well under
+ * one.
+ *
+ * `docs/marketplace.md` proposed instead reading manifests over Gitea's API, which needed either a
+ * sandbox to evaluate a fetched string or a committed `manifest.json` for every plugin. Git gives the
+ * same guarantee — content at a named release, independent of any checkout — with neither, and the trust
+ * model is unchanged from today: we import our own repositories' code.
+ *
+ * `[still a blocker for third-party submissions]` It is unchanged BECAUSE it is unchanged. Importing a
+ * stranger's manifest is code execution no matter which ref it came from. Submissions still need a
+ * sandbox or a published `manifest.json`.
+ *
+ * ── Which plugins, and where from: the registry, and only the registry ──
+ *
+ * The list comes from the `plugins` table, which `docs/marketplace.md` has always called the
+ * authoritative, hand-curated half. It was not actually being used that way — the old sync discovered
+ * plugins by listing a directory, so the registry was a by-product of whatever officer-dev had
+ * installed. Now it decides, which means:
+ *
+ *   · a plugin nobody has installed here can still be published (`finances` is in exactly that state),
+ *   · a plugin lives wherever its registry row says, across orgs — `fofos-admin` is under `fofos/`,
+ *     not `plugins/`, so an org listing would have missed it,
+ *   · and adding one is a deliberate act: `marketplace-sync.ts --add <https clone url>`.
+ *
+ * ── A plugin with no tag is left ALONE, not removed ──
+ *
+ * It has never made a release, so this file has nothing true to say about it and says nothing: the
+ * existing metadata row stays, and the run reports it. Deleting the row would drop it out of the store
+ * on the day the rule changed, which is a decision about somebody's plugin dressed up as a refresh.
+ * Tagging it is the fix, and the report is there to ask for one.
+ */
+
+export type SyncOutcome =
+  | { appName: string; status: 'synced'; tag: string; version: string }
+  /** The release on the remote is the one already published. Nothing was written. */
+  | { appName: string; status: 'unchanged'; tag: string }
+  | { appName: string; status: 'unreleased' }
+  | { appName: string; status: 'failed'; error: string };
+
+/*
+  Where the mirrors live.
+
+  A fixed path under the user's cache directory, NOT one derived from the environment. This runs three
+  ways — by hand with `--env-file=.env`, from the webhook inside `officer-marketplace`, and from cron —
+  and those three do not have the same environment. A path that moved with `DATA_PATH` would have each
+  of them re-cloning twenty-five repositories into its own copy, and the "is it already mirrored" check
+  would answer differently depending on who asked.
+
+  Not in the source tree, because it is a cache of remote repositories and a `.gitignore` entry is a
+  worse way to keep several hundred megabytes out of a repo than not putting them there.
+*/
+const MIRROR_DIR =
+  process.env.MARKETPLACE_MIRROR_DIR ?? join(homedir(), '.cache', 'officer-marketplace', 'mirrors');
+
+const GIT_TIMEOUT_MS = 60_000;
+
+type Git = { ok: true; out: string } | { ok: false; err: string };
+
+async function git(args: string[], cwd?: string): Promise<Git> {
+  // A missing `cwd` makes posix_spawn fail with ENOENT, which reads as "git is not installed" and sent
+  // a first run of this looking for git for ten minutes. It means the mirror is not there yet.
+  if (cwd && !existsSync(cwd)) return { ok: false, err: `${cwd} does not exist` };
+
+  const proc = Bun.spawn(['git', ...args], {
+    cwd,
+    stdout: 'pipe',
+    stderr: 'pipe',
+    // Never inherited: a mirror fetch that hits a host asking for a password would otherwise sit on a
+    // terminal prompt forever, and in the webhook server there is no terminal to sit on.
+    env: { ...process.env, GIT_TERMINAL_PROMPT: '0', GIT_ASKPASS: '/bin/true' },
+    signal: AbortSignal.timeout(GIT_TIMEOUT_MS),
+  });
+  const [out, err] = await Promise.all([new Response(proc.stdout).text(), new Response(proc.stderr).text()]);
+  return (await proc.exited) === 0 ? { ok: true, out: out.trim() } : { ok: false, err: err.trim() || 'git failed' };
+}
+
+/** The mirror for a plugin, cloned if this is the first time and refreshed if it is not. */
+async function mirror(appName: string, repoUrl: string): Promise<string> {
+  mkdirSync(MIRROR_DIR, { recursive: true });
+  const dir = join(MIRROR_DIR, `${appName}.git`);
+
+  // Re-cloned rather than repaired when it is not a healthy bare repo: an interrupted clone leaves a
+  // directory that every later command fails against, and there is nothing in it worth keeping.
+  const existing = await git(['rev-parse', '--is-bare-repository'], dir);
+  if (!existing.ok || existing.out !== 'true') {
+    if (existsSync(dir)) await Bun.$`rm -rf ${dir}`.quiet().nothrow();
+    const cloned = await git(['clone', '--mirror', '--quiet', repoUrl, dir]);
+    if (!cloned.ok) throw new Error(`clone failed: ${cloned.err}`);
+    return dir;
+  }
+
+  // The registry is allowed to move a plugin. Setting the URL every time is one command and removes the
+  // failure where a repo is relocated and the mirror keeps refreshing from where it used to be.
+  await git(['remote', 'set-url', 'origin', repoUrl], dir);
+
+  // `--prune --prune-tags`: a deleted or MOVED tag has to disappear here too. Without it, a tag that was
+  // pushed by mistake and deleted would stay in the mirror forever and could go on being published as
+  // the newest release — a store advertising a version its repository no longer has.
+  const fetched = await git(['fetch', '--quiet', '--prune', '--prune-tags', '--tags', 'origin'], dir);
+  if (!fetched.ok) throw new Error(`fetch failed: ${fetched.err}`);
+  return dir;
+}
+
+/**
+ * The newest release, or null for a plugin that has never made one.
+ *
+ * Asked of git rather than by sorting version strings, which is the same rule `servers/plugins/release.ts`
+ * follows and for the same reason: parsing `2026.08.27.01` works right up until somebody tags `v2`, and
+ * then fails silently. `--abbrev=0` gives the closest tag reachable from HEAD, and a mirror's HEAD is the
+ * remote's default branch, so this is "the newest release on the branch the repository points at".
+ */
+async function newestTag(dir: string): Promise<string | null> {
+  const described = await git(['describe', '--tags', '--abbrev=0', 'HEAD'], dir);
+  return described.ok ? described.out : null;
+}
+
+/** A file at a ref, or null when it is not in that tree. */
+async function fileAt(dir: string, ref: string, path: string): Promise<string | null> {
+  const shown = await git(['cat-file', '-p', `${ref}:${path}`], dir);
+  return shown.ok ? shown.out : null;
+}
+
+/**
+ * The manifest at a release, as an object.
+ *
+ * Written to a temp file and imported, because a manifest is TypeScript. Keyed by the COMMIT, so the
+ * module cache is correct rather than merely convenient: the same commit is the same file, and a tag
+ * moved to a different commit gets a different path and a fresh import. That matters in the webhook
+ * server, which is long-lived and would otherwise serve the first version of a manifest it ever read.
+ *
+ * The `import type { PluginManifest }` at the top of every manifest is erased by the transpiler, so the
+ * file has no runtime dependency on the platform's path aliases and imports fine from anywhere.
+ */
+async function manifestAt(dir: string, ref: string, appName: string, commit: string): Promise<Record<string, unknown>> {
+  const source = await fileAt(dir, ref, 'manifest.ts');
+  if (source === null) throw new Error(`no manifest.ts at ${ref}`);
+
+  const scratch = join(tmpdir(), 'officer-marketplace-manifests');
+  mkdirSync(scratch, { recursive: true });
+  const path = join(scratch, `${appName}-${commit}.ts`);
+  await Bun.write(path, source);
+
+  const mod = (await import(path)) as { manifest?: Record<string, unknown> };
+  if (!mod.manifest) throw new Error(`manifest.ts at ${ref} exports no \`manifest\``);
+  return mod.manifest;
+}
+
+/**
+ * A raw-file URL for a Gitea repository, PINNED TO THE RELEASE.
+ *
+ * It used to be `/raw/HEAD/…`, which works only because Gitea 303s it to `/raw/branch/HEAD/…`; the
+ * artwork it served was whatever was on the default branch, so a plugin could show one icon in the store
+ * and ship another. `/raw/tag/<tag>/…` answers 200 directly and shows the icon that belongs to the
+ * version being advertised.
+ *
+ * `[open]` Still Gitea-shaped. A GitHub-hosted plugin needs the other form, and that wants the forge to
+ * be part of the registry rather than inferred from the URL.
+ */
+function rawUrl(repoUrl: string, tag: string, path: string): string {
+  return `${repoUrl.replace(/\.git$/, '')}/raw/tag/${encodeURIComponent(tag)}/${path}`;
+}
+
+/*
+  `force` re-publishes a release that is already published.
+
+  Needed because "unchanged" is judged on the plugin's commit, and that is only half the input: the other
+  half is THIS file. Change how a field is derived — as the switch to tag-pinned icon URLs did — and every
+  row is stale while every commit still matches. Then the run to fix them all is `--force`.
+*/
+async function syncRegistered(row: RegistryRow, force: boolean): Promise<SyncOutcome> {
+  const { appName, repoUrl } = row;
+  const dir = await mirror(appName, repoUrl);
+
+  const tag = await newestTag(dir);
+  if (!tag) return { appName, status: 'unreleased' };
+
+  const commitAt = await git(['rev-list', '-n', '1', tag], dir);
+  if (!commitAt.ok) throw new Error(`cannot resolve ${tag}`);
+  const commit = commitAt.out;
+
+  // Before the manifest is read, not after: this is what makes a per-minute run affordable. Twenty-five
+  // repositories with nothing new cost one query, twenty-five fetches of nothing, and no writes at all.
+  if (!force && row.sourceCommit === commit) return { appName, status: 'unchanged', tag };
+
+  const manifest = await manifestAt(dir, tag, appName, commit);
+
+  // What the plugin SHIPS, read from the tree at the release rather than from a declaration — the same
+  // rule discovery follows on disk, so a catalogue entry cannot claim a sidecar the release does not
+  // have.
+  const listed = await git(['ls-tree', '-r', '--name-only', tag], dir);
+  const files = new Set(listed.ok ? listed.out.split('\n') : []);
+  const has = {
+    api: files.has('api/router.ts'),
+    schema: files.has('db/schema.ts'),
+    sidecar: files.has('sidecar/index.ts') || files.has('sidecar/index.mjs'),
+    web: [...files].some((f) => f.startsWith('web/')),
+  };
+
+  const readme = (await fileAt(dir, tag, 'README.md')) ?? (await fileAt(dir, tag, 'readme.md'));
+
+  const version = String(manifest.version ?? tag);
+  if (version !== tag) {
+    // Not an error and not corrected here. The tag is what was released and the manifest is what the
+    // plugin will report about itself once installed; when they disagree, an update badge that compares
+    // versions goes quiet. Saying so is the only way anyone finds out.
+    console.warn(`[marketplace-sync] ${appName}: tag ${tag} but manifest says ${version} — they should match`);
+  }
+
+  const developer = String(manifest.publisher ?? 'unknown');
+  const pluginId = await upsertPlugin({ appName, developer, repoUrl });
+
+  await upsertMetadata(pluginId, {
+    version,
+    releaseTag: tag,
+    platform: String(manifest.platform ?? '>=1.0.0'),
+    label: String(manifest.label ?? appName),
+    summary: String(manifest.summary ?? ''),
+    // Taken verbatim, NOT validated here. `manifestProblems` already refuses a tag outside the
+    // vocabulary at discovery, so anything reaching this point is either valid or belongs to a plugin
+    // the platform would not load anyway — and a sync that silently dropped a tag would be a catalogue
+    // quietly disagreeing with the repository it claims to mirror.
+    tags: Array.isArray(manifest.tags) ? manifest.tags : [],
+    color: (manifest.color as string) ?? null,
+    tile: (manifest.tile as string) ?? null,
+    iconUrl: files.has('assets/icon.png') ? rawUrl(repoUrl, tag, 'assets/icon.png') : null,
+    icon: (manifest.icon as string) ?? null,
+    permissions: manifest.permissions ?? [],
+    osDependencies: manifest.osDependencies ?? null,
+    has,
+    readme,
+    sourceCommit: commit,
+  });
+
+  return { appName, status: 'synced', tag, version };
+}
+
+/** One plugin by app name. Unknown to the registry is a failure, not a silent no-op. */
+export async function syncPlugin(appName: string, force = false): Promise<SyncOutcome> {
+  const row = (await listRegistry()).find((r) => r.appName === appName);
+  if (!row) return { appName, status: 'failed', error: 'not in the registry' };
+  return syncRegistered(row, force).catch((err) => ({
+    appName,
+    status: 'failed' as const,
+    error: err instanceof Error ? err.message : String(err),
+  }));
+}
+
+/**
+ * Every registered plugin, one at a time.
+ *
+ * Sequential on purpose. Twenty-four concurrent fetches against one Gitea is a thundering herd for a job
+ * nothing is waiting on, and the whole set finishes in about a second anyway.
+ *
+ * Unlisted rows are included: `listed` hides a plugin from the public API, usually after a bad release,
+ * and it has to be able to come back with current metadata rather than with whatever was true when it
+ * was pulled.
+ */
+export async function syncAll(force = false): Promise<SyncOutcome[]> {
+  const rows = await listRegistry();
+  const out: SyncOutcome[] = [];
+  for (const row of rows) {
+    out.push(
+      await syncRegistered(row, force).catch((err) => ({
+        appName: row.appName,
+        status: 'failed' as const,
+        error: err instanceof Error ? err.message : String(err),
+      })),
+    );
+  }
+  return out;
+}
@@ -30,8 +30,22 @@ import type { DiscoveredPlugin } from './manifest';

 const ECOSYSTEM_PATH = join(PLATFORM_DIR, 'ecosystem.config.cjs');

-/** One PM2 app entry. Matches what `services.sh` emits for the core processes. */
-export type Pm2App = { name: string; script: string; args?: string; cwd: string; watch: boolean };
+/**
+ * One PM2 app entry. Matches what `services.sh` emits for the core processes.
+ *
+ * `env` carries the ports the platform allocated for this plugin — `PORT_<NAME>`. It is written into the
+ * ecosystem file rather than sent over the registration socket because a socket answer arrives after the
+ * process has already had to decide what to bind, and pm2 is the only thing that can tell a process
+ * something before it starts.
+ */
+export type Pm2App = {
+  name: string;
+  script: string;
+  args?: string;
+  cwd: string;
+  watch: boolean;
+  env?: Record<string, string>;
+};

 /**
 * The header to write when there is no file to take one from.
@@ -81,12 +95,24 @@ function writeEcosystemApps(apps: Pm2App[]): void {
  const marker = existing.indexOf('module.exports');
  const header = marker > 0 ? existing.slice(0, marker) : FALLBACK_HEADER;

-  const lines = apps.map(
-    (app) =>
+  const lines = apps.map((app) => {
+    /*
+      Values are emitted as JSON rather than quoted by hand. A port is a number today, but this is the
+      seam where a path or a URL arrives eventually, and one apostrophe in a value would produce a file
+      PM2 cannot parse — which takes every process on the machine down, not just this plugin's.
+    */
+    const env =
+      app.env && Object.keys(app.env).length
+        ? `, env: { ${Object.entries(app.env)
+            .map(([k, v]) => `${JSON.stringify(k)}: ${JSON.stringify(v)}`)
+            .join(', ')} }`
+        : '';
+    return (
      `    { name: '${app.name}', script: '${app.script}'` +
      (app.args ? `, args: '${app.args}'` : '') +
-      `, cwd: '${app.cwd}', watch: ${app.watch} },`,
-  );
+      `, cwd: '${app.cwd}', watch: ${app.watch}${env} },`
+    );
+  });
  writeFileSync(ECOSYSTEM_PATH, `${header}module.exports = {\n  apps: [\n${lines.join('\n')}\n  ],\n};\n`);
 }

@@ -98,7 +124,7 @@ function writeEcosystemApps(apps: Pm2App[]): void {
 * the database says is installed and enabled. Neither should duplicate the other's idea of what a
 * plugin's entry looks like.
 */
-export function pluginPm2Entry(plugin: DiscoveredPlugin): Pm2App | null {
+export function pluginPm2Entry(plugin: DiscoveredPlugin, ports?: Record<string, number>): Pm2App | null {
  if (!plugin.sidecar) return null;

  const name = pluginProcessName(plugin.appName);
@@ -108,9 +134,16 @@ export function pluginPm2Entry(plugin: DiscoveredPlugin): Pm2App | null {
    ? plugin.sidecar.script.slice(PLATFORM_DIR.length + 1)
    : plugin.sidecar.script;

+  // `PORT_<NAME>`, uppercased. Absent entirely when a plugin asked for none, which is nearly all of them.
+  const env = Object.fromEntries(
+    Object.entries(ports ?? {}).map(([name, port]) => [`PORT_${name.toUpperCase().replace(/-/g, '_')}`, String(port)]),
+  );
+
+  const base = { name, cwd: PLATFORM_DIR, watch: false, ...(Object.keys(env).length ? { env } : {}) };
+
  return plugin.sidecar.runtime === 'node'
-    ? { name, script: 'node', args: script, cwd: PLATFORM_DIR, watch: false }
-    : { name, script: 'bun', args: `run ${script}`, cwd: PLATFORM_DIR, watch: false };
+    ? { ...base, script: 'node', args: script }
+    : { ...base, script: 'bun', args: `run ${script}` };
 }

 /**
@@ -119,8 +152,8 @@ export function pluginPm2Entry(plugin: DiscoveredPlugin): Pm2App | null {
 * Idempotent, because re-installing is how a plugin is upgraded and the script path may have moved.
 * Returns false when the plugin has no sidecar at all, which is not a failure — most plugins won't.
 */
-export function addPluginToEcosystem(plugin: DiscoveredPlugin): boolean {
-  const entry = pluginPm2Entry(plugin);
+export function addPluginToEcosystem(plugin: DiscoveredPlugin, ports?: Record<string, number>): boolean {
+  const entry = pluginPm2Entry(plugin, ports);
  if (!entry) return false;

  const apps = readEcosystemApps().filter((a) => a.name !== entry.name);
@@ -15,7 +15,7 @@ import { deleteProcess, processStatus, startProcess, stopProcess } from '../app-
 import { addPluginToEcosystem, pluginProcessName, removePluginFromEcosystem } from './ecosystem';
 import { discoverPlugins } from './discover';
 import { fetchPlugin } from './fetch';
-import { revokeGrantsForPermissions } from 'officerdb';
+import { allocatePluginPort, releasePluginPorts, revokeGrantsForPermissions } from 'officerdb';
 import { grantPluginToAllRoles } from './grant-on-install';
 import { publishAssets, unpublishAssets } from '../app-store/assets';
 import { refreshPluginMounts, snapshotPlugins } from './mount';
@@ -453,7 +453,29 @@ export async function installPlugin(
    );

    if (hasSidecar(plugin)) {
-      addPluginToEcosystem(plugin);
+      /*
+        Ports the manifest asked the platform to own, allocated before the entry is written because the
+        entry carries them.
+
+        Idempotent: a plugin that already has an allocation gets the same number back, so re-installing
+        and upgrading do not move a port somebody has written into an nginx proxy host. See
+        `pluginPorts` in officerdb for why the platform allocates these when it allocates nothing else.
+      */
+      const ports: Record<string, number> = {};
+      for (const declared of plugin.manifest.extraPorts ?? []) {
+        ports[declared.name] = await allocatePluginPort(appName, declared.name);
+      }
+      if (Object.keys(ports).length) {
+        await step(
+          steps,
+          onStep,
+          `ports: ${Object.entries(ports)
+            .map(([name, port]) => `${name}=${port}`)
+            .join(', ')}`,
+        );
+      }
+
+      addPluginToEcosystem(plugin, ports);
      await step(steps, onStep, `ecosystem: ${pluginProcessName(appName)} added`);

      const started = await startProcess(pluginProcessName(appName), PLATFORM_DIR);
@@ -609,6 +631,17 @@ export async function purgePlugin(
    }
  }

+  /*
+    Ports go here and NOT on uninstall, which is the deliberate asymmetry.
+
+    An allocated number is written down where this database cannot see it — an nginx proxy host, a
+    firewall rule, somebody's notes. Uninstalling is reversible and often temporary, so giving the number
+    up there would let a later plugin take it and point that configuration at the wrong service. Purge is
+    the act that says the plugin is gone for good, and that is when the number goes back in the pool.
+  */
+  const released = await releasePluginPorts(appName);
+  if (released) await step(steps, onStep, `ports released: ${released}`);
+
  await rm(plugin.dir, { recursive: true, force: true });
  await step(steps, onStep, `deleted ${plugin.dir}`);

@@ -405,6 +405,51 @@ export type PluginManifest = {
   */
  protocolHeaders?: string[];

+  /**
+   * Ports this plugin needs the PLATFORM to allocate and remember.
+   *
+   * ── Not the same thing as a sidecar's own port ──
+   *
+   * A sidecar already gets a port, and nothing here changes that: it binds `port: 0`, the kernel picks
+   * one, and it announces the number over the registration socket. That port is a private detail between
+   * two processes — loopback, re-chosen on every restart, and nothing outside Officer ever needs it.
+   *
+   * This is for a port something ELSE connects to. A plugin serving a website has to be reachable from
+   * nginx, which means the number must survive a restart and must be knowable by whoever writes the proxy
+   * configuration. Ephemeral fails both, which is why the first site plugin had to hardcode 3011 and hope.
+   *
+   * ── What the platform guarantees, and what it does not ──
+   *
+   * Guaranteed: the same plugin and name always get the same number back, and no two plugins are ever
+   * given the same one — that second part is a unique index, not a convention, because a collision is two
+   * services fighting over a listener rather than a duplicate row.
+   *
+   * NOT guaranteed: that the port is still free when the sidecar starts. Something else on the machine
+   * can take it while Officer is down. A sidecar that cannot bind its remembered port must fail loudly
+   * rather than quietly choose another, because the number is written down where this platform cannot
+   * see it.
+   *
+   * ── How it reaches the sidecar ──
+   *
+   * As `PORT_<NAME>` in its environment, uppercased, written into the pm2 entry at install. The sidecar
+   * reads it rather than being told over the socket, because a socket answer would arrive after the
+   * process has already had to decide what to bind.
+   *
+   * ```ts
+   * extraPorts: [{ name: 'site' }]   // → PORT_SITE=<n> in the sidecar's environment
+   * ```
+   *
+   * A manifest does NOT get to ask for a particular number, and that is deliberate. A preference sounds
+   * helpful and fails silently: when the wished-for port is taken the plugin is handed a different one,
+   * the install succeeds, and whatever was configured against the old number keeps pointing at nothing.
+   * Nobody is told. Drawing a fresh one every time makes the number always something you had to go and
+   * read, which is the state that keeps the configuration honest.
+   *
+   * Binding is still the sidecar's business, including which interface. A port that exists to be reached
+   * from a container cannot be loopback, and the platform has no way to know which a plugin means.
+   */
+  extraPorts?: { name: string }[];
+
  /**
   * Host binaries this plugin needs. Absent means it is self-sufficient, which offscale and example are.
   *
@@ -957,6 +1002,24 @@ export function manifestProblems(appName: string, manifest: Partial<PluginManife
      }
    }
  }
+  if (manifest.extraPorts !== undefined) {
+    if (!Array.isArray(manifest.extraPorts)) {
+      problems.push('extraPorts must be an array');
+    } else {
+      const seen = new Set<string>();
+      for (const [i, entry] of manifest.extraPorts.entries()) {
+        const name = (entry as { name?: unknown } | null)?.name;
+        if (typeof name !== 'string' || !/^[a-z][a-z0-9-]{0,30}$/.test(name)) {
+          problems.push(`extraPorts[${i}].name must be a lowercase name, e.g. "site"`);
+          continue;
+        }
+        // The name becomes an environment variable, so two that differ only by case would collide there
+        // while looking distinct here.
+        if (seen.has(name)) problems.push(`extraPorts[${i}] repeats the name "${name}"`);
+        seen.add(name);
+      }
+    }
+  }
  if (manifest.protocolHeaders !== undefined) {
    if (!Array.isArray(manifest.protocolHeaders)) {
      problems.push('protocolHeaders must be an array of header names');
@@ -39,6 +39,15 @@ export type CatalogueEntry = {
  appName: string;
  publisher: string;
  version: string;
+  /**
+   * The git tag this entry was published from — the newest RELEASE, as opposed to whatever a manifest
+   * claims about itself.
+   *
+   * Optional, and absent means "cannot tell" rather than "no release": an older catalogue does not send
+   * the field at all, and a plugin that has never been tagged has nothing to put in it. Both send the
+   * badge back to the git rule.
+   */
+  releaseTag?: string;
  platform: string;
  label: string;
  summary: string;
@@ -556,7 +565,15 @@ const MARKETPLACE_URL = process.env.MARKETPLACE_URL ?? 'https://marketplace.offi
 * plugin published this morning shows up this morning. The cache is per PROCESS and dies with it — a
 * restart is always a fresh look.
 */
-const CACHE_TTL_MS = 5 * 60 * 1000;
+/*
+  One minute, and the number changed when the badge started depending on it.
+
+  Five was fine while this only drew the store's tiles; it is not fine now that "is there an update" is
+  answered from the catalogue, because the wait for a release to appear became the sync's minute PLUS
+  this one. The cost is one small request per process per minute, and only while somebody is looking at
+  a page that lists plugins.
+*/
+const CACHE_TTL_MS = 60 * 1000;

 /**
 * How long to wait for the store before giving up on it.
@@ -639,6 +656,22 @@ export async function fetchCatalogue(): Promise<CatalogueEntry[]> {
  }
 }

+/**
+ * What the store says the newest release of each plugin is, keyed by app name.
+ *
+ * This is the fact the update badge is built on. It replaces every host in the fleet running `git fetch`
+ * against twenty-one repositories to re-derive something the catalogue already knows — see
+ * `plugins/release.ts` for the rule and for what happens to a plugin that has no entry here.
+ *
+ * A missing key and a null value mean the same thing to a caller and are kept apart for the reader: no
+ * entry at all (unregistered, or a store that could not be reached) versus an entry with no release yet.
+ * Both fall through to the git rule, which is the only honest answer when the store cannot speak.
+ */
+export async function publishedTags(): Promise<Map<string, string | null>> {
+  const entries = await fetchCatalogue().catch(() => []);
+  return new Map(entries.map((e) => [e.appName, e.releaseTag ?? null]));
+}
+
 /** One entry by app name, or null. Used by install to find where to clone from. */
 export async function catalogueEntry(appName: string): Promise<CatalogueEntry | null> {
  const all = await fetchCatalogue().catch(() => BUNDLED);
@@ -59,7 +59,13 @@ export type PluginsSnapshot = {
 * the tree while the row stayed, and there is nothing to mount, describe or offer. The row is left in the
 * database on purpose — deleting it here would turn "somebody moved the checkout" into silent data loss.
 */
-export async function snapshotPlugins(): Promise<PluginsSnapshot> {
+/**
+ * @param published Newest released tag per plugin, from the marketplace — `publishedTags()`. Passed by
+ *   the one caller that renders update badges, and by nothing else on purpose: every other caller wants
+ *   "what is on disk", and boot is among them. Making this function fetch the catalogue itself would put
+ *   a network request with a timeout in front of mounting plugins on a machine with no internet.
+ */
+export async function snapshotPlugins(published?: Map<string, string | null>): Promise<PluginsSnapshot> {
  const [{ plugins, broken }, installs] = await Promise.all([discoverPlugins(), listPluginInstalls()]);
  const byName = new Map(installs.map((row) => [row.appName, row]));

@@ -72,7 +78,7 @@ export async function snapshotPlugins(): Promise<PluginsSnapshot> {
        plugin,
        install,
        outdated: !!install && install.version !== plugin.manifest.version,
-        release: await releaseState(plugin.dir),
+        release: await releaseState(plugin.dir, published?.get(plugin.appName) ?? null),
      };
    }),
  );
@@ -94,6 +94,61 @@ describe('releaseState', () => {
    expect(s.updateable).toBe(true);
  });

+  /*
+    The catalogue path. These are the ones that matter now: a host no longer fetches to find out what the
+    newest release is, so every assertion below is about the store's tag winning over local refs — and
+    about the fallback still being there when it has nothing to say.
+  */
+  test('the published tag decides, without the checkout having ever fetched it', async () => {
+    const { work, origin, clone } = pair('published-newer');
+
+    // The clone is at v1 and has NEVER seen v2 — no fetch, no remote ref for it. Exactly the state a
+    // host is in a minute after somebody else tagged a release.
+    publish(work, origin, 'two', 'v1');
+    git(clone, 'fetch', '-q', '--tags', 'origin');
+    git(clone, 'checkout', '-q', 'v1');
+
+    const s = await releaseState(clone, 'v2');
+    expect(s.tagAtHead).toBe('v1');
+    expect(s.newestTag).toBe('v2');
+    expect(s.updateable).toBe(true);
+    expect(s.reason).toBe('new-release');
+  });
+
+  test('the published tag matching HEAD is up to date', async () => {
+    const { work, origin, clone } = pair('published-same');
+    publish(work, origin, 'two', 'v1');
+    git(clone, 'fetch', '-q', '--tags', 'origin');
+    git(clone, 'checkout', '-q', 'v1');
+
+    const s = await releaseState(clone, 'v1');
+    expect(s.updateable).toBe(false);
+    expect(s.reason).toBe('at-latest');
+  });
+
+  test('an untagged checkout is updateable AND says what it would land on', async () => {
+    const { clone } = pair('published-untagged');
+    const s = await releaseState(clone, 'v9');
+    expect(s.tagAtHead).toBeNull();
+    // The old code returned null here, so the button read "Update" with nothing after it.
+    expect(s.newestTag).toBe('v9');
+    expect(s.updateable).toBe(true);
+    expect(s.reason).toBe('untagged');
+  });
+
+  test('no published tag falls back to the git rule rather than to a guess', async () => {
+    const { work, origin, clone } = pair('no-entry');
+    publish(work, origin, 'two', 'v1');
+    git(clone, 'fetch', '-q', '--tags', 'origin');
+    git(clone, 'checkout', '-q', 'v1');
+
+    // Null, as it is for a plugin nobody registered or a store that could not be reached.
+    const s = await releaseState(clone, null);
+    expect(s.tagAtHead).toBe('v1');
+    expect(s.newestTag).toBe('v1');
+    expect(s.updateable).toBe(false);
+  });
+
  test('a directory that is not a checkout is not updateable', async () => {
    const d = join(root, 'plain');
    mkdirSync(d, { recursive: true });
@@ -29,12 +29,32 @@ import { join } from 'node:path';
 * is that tag, this checkout is at the latest release. If it is a different one, the remote has moved on.
 * That holds for any tag names anyone ever chooses, including ones that do not sort.
 *
- * ── What this does NOT do ──
+ * ── Where "the newest tag" comes from: the catalogue, not this machine's refs ──
 *
- * It never fetches. Every answer comes from refs already on disk, because this is called once per plugin
- * on a request that renders a list — twenty-one network round trips would make the plugins screen feel
- * broken. Tags arrive at clone and at the fetch `pullPlugin` already does, so "check for updates" is a
- * consequence of updating rather than a separate poll.
+ * Every answer here is local, and this is called once per plugin on a request that renders a list —
+ * twenty-one network round trips would make the plugins screen feel broken. That left one hard problem:
+ * refs on disk go stale. Tags arrived at clone and at the fetch `pullPlugin` does, so you learned an
+ * update existed BY UPDATING. A machine that had not touched a plugin since August compared this
+ * morning's release against a snapshot from before it existed and said, truthfully and uselessly, that
+ * it was at the latest.
+ *
+ * That was patched with a rate-limited background `git fetch` per checkout, which worked and was the
+ * wrong shape: every host in the fleet polling twenty-one repositories to re-derive a fact one place
+ * already knows. The marketplace publishes releases now — see `servers/marketplace/sync.ts` — so the
+ * newest release is simply asked of the catalogue, which every host already fetches for the store.
+ *
+ *   published tag (catalogue)  vs  tag at HEAD (this checkout)   →  differ, so there is an update
+ *
+ * DIFFERENCE, not ordering. `2026.08.27.01` parses fine until somebody tags `v2`, and then it fails
+ * silently. Two strings that are not equal is a comparison that cannot rot.
+ *
+ * ── The git rule stays, for a plugin the store cannot speak for ──
+ *
+ * A checkout of something unpublished — a plugin being written, one never registered — has no catalogue
+ * entry, and the fallback below is the old rule against local remote-tracking refs. Those refs are no
+ * longer refreshed here, so for that case this can miss a tag until something else fetches. That is the
+ * accepted cost: the alternative was every machine polling every repository forever to cover a case that
+ * only arises on a developer's own box, where `git fetch` is a thing they were about to run anyway.
 */

 export type ReleaseState = {
@@ -53,7 +73,12 @@ const run = async (dir: string, args: string[]): Promise<string | null> => {
  return (await proc.exited) === 0 ? out.trim() || null : null;
 };

-export async function releaseState(dir: string): Promise<ReleaseState> {
+/**
+ * @param publishedTag The tag the marketplace says is this plugin's newest release, or null when the
+ *   catalogue has nothing to say — an unregistered plugin, one never tagged, or a store that could not
+ *   be reached. Null falls through to the git rule rather than to a guess.
+ */
+export async function releaseState(dir: string, publishedTag: string | null = null): Promise<ReleaseState> {
  const none = { tagAtHead: null, newestTag: null };
  if (!existsSync(join(dir, '.git'))) {
    // Copied in by hand or bundled. Nothing to compare, and nothing to pull either — `pullPlugin` says
@@ -64,6 +89,16 @@ export async function releaseState(dir: string): Promise<ReleaseState> {
  // `--exact-match`, so this is "the tag AT this commit", not "the last tag before it". Fails, and
  // therefore returns null, when HEAD is between releases — which is the first half of the rule.
  const tagAtHead = await run(dir, ['describe', '--exact-match', '--tags', 'HEAD']);
+
+  if (publishedTag) {
+    // Reported even when HEAD is untagged, so the UI can say what pressing Update lands on. The old
+    // code returned a bare null here and the button read "Update" with nothing after it.
+    if (!tagAtHead) return { tagAtHead: null, newestTag: publishedTag, updateable: true, reason: 'untagged' };
+    return publishedTag === tagAtHead
+      ? { tagAtHead, newestTag: publishedTag, updateable: false, reason: 'at-latest' }
+      : { tagAtHead, newestTag: publishedTag, updateable: true, reason: 'new-release' };
+  }
+
  if (!tagAtHead) return { ...none, updateable: true, reason: 'untagged' };

  // Asked of the REMOTE branch, not of HEAD: a tag pushed after this checkout was made is not reachable
@@ -1,4 +1,5 @@
 import { existsSync, readdirSync } from 'node:fs';
+import { publishedTags } from './marketplace';
 import { releaseState } from './release';
 import { join } from 'node:path';
 import { PLUGINS_DIR } from './discover';
@@ -144,7 +145,10 @@ export async function updatePlugin(

    Checked BEFORE the pull, because a pull is what would make the answer stale.
  */
-  const state = await releaseState(join(PLUGINS_DIR, appName));
+  // Against the CATALOGUE, like the badge. Without it this asked local remote-tracking refs, which are
+  // no longer refreshed in the background — so a plugin at the latest tag it knows about would have
+  // short-circuited here and told the owner it was up to date without checking anything.
+  const state = await releaseState(join(PLUGINS_DIR, appName), (await publishedTags()).get(appName) ?? null);
  if (!state.updateable && state.reason === 'at-latest') {
    steps.push(`already at ${state.tagAtHead}, the newest release`);
    onStep?.(steps[steps.length - 1]!);
@@ -0,0 +1,74 @@
+/**
+ * A screenshot, shrunk to the largest size the model will actually look at.
+ *
+ * ── Why there is a ceiling at all ──
+ *
+ * An attached image goes to the model as base64 INSIDE the message, and it stays there: every later
+ * turn in the conversation re-sends every earlier picture. So a screenshot is not paid for once, it is
+ * paid for on every turn for the rest of the session — in request bytes and in tokens.
+ *
+ * ── Why 1568 specifically ──
+ *
+ * It is the point past which Anthropic downscales server-side. A 2539px-wide screenshot is resized to
+ * 1568 before the model sees a pixel of it, so the extra width was never read — it was carried up the
+ * wire, counted, and thrown away. Sending it already small changes nothing about what the model can
+ * see and stops paying for what it cannot.
+ *
+ * Measured on one day of this session: 25 screenshots, 10 of them over the line, 53,300 tokens of
+ * image data — 23% of which bought nothing.
+ *
+ * ── The ORIGINAL is still uploaded ──
+ *
+ * This makes the copy that travels in the prompt. The file that lands on disk is untouched, so an agent
+ * that needs the detail can `Read` it at full size — which is the right split: the conversation carries
+ * what is worth re-reading every turn, and the disk keeps what was actually captured.
+ */
+const MAX_EDGE = 1568;
+
+/** JPEG for photographs would be smaller, and a screenshot of text is exactly where it looks worst. */
+const TYPE = 'image/png';
+
+export async function downscaleForPrompt(file: File): Promise<string> {
+  const bitmap = await createImageBitmap(file).catch(() => null);
+  if (!bitmap) return readAsDataUrl(file);
+
+  const scale = Math.min(1, MAX_EDGE / Math.max(bitmap.width, bitmap.height));
+  // Already small enough. Re-encoding it would cost quality and could cost SIZE — a PNG round-tripped
+  // through a canvas is often bigger than the one the screenshotter wrote.
+  if (scale === 1) {
+    bitmap.close();
+    return readAsDataUrl(file);
+  }
+
+  const width = Math.round(bitmap.width * scale);
+  const height = Math.round(bitmap.height * scale);
+
+  const canvas = document.createElement('canvas');
+  canvas.width = width;
+  canvas.height = height;
+
+  const context = canvas.getContext('2d');
+  if (!context) {
+    bitmap.close();
+    return readAsDataUrl(file);
+  }
+
+  // `high` because this is text: the default is a box filter, and screenshot text resampled with one
+  // comes out fringed in a way that is harder to read than the same text smaller.
+  context.imageSmoothingEnabled = true;
+  context.imageSmoothingQuality = 'high';
+  context.drawImage(bitmap, 0, 0, width, height);
+  bitmap.close();
+
+  return canvas.toDataURL(TYPE);
+}
+
+/** The file as it is, for anything this cannot or should not shrink. */
+function readAsDataUrl(file: File): Promise<string> {
+  return new Promise((resolve, reject) => {
+    const reader = new FileReader();
+    reader.onload = () => resolve(String(reader.result));
+    reader.onerror = () => reject(reader.error);
+    reader.readAsDataURL(file);
+  });
+}
@@ -1,4 +1,5 @@
 import { useState } from 'react';
+import { downscaleForPrompt } from './components/downscale';
 import { toast } from 'sonner';
 import { useClient } from 'hooks/useClient';
 import type { Attachment } from './types';
@@ -52,11 +53,22 @@ export function useAttachments(params?: UseAttachmentsParams) {
      if (params?.sessionId) formData.append('sessionId', params.sessionId);
      formData.append('provider', 'chat');

-      const res = await client.post<{ filename: string; dataUrl: string; attachmentId: string }>('/upload', formData);
+      /*
+        The original is uploaded and a SMALLER copy is what travels in the prompt.
+
+        The server's `dataUrl` is the file byte for byte, and that copy is re-sent on every turn for the
+        rest of the conversation. Past 1568px on the long edge the model never sees the extra pixels —
+        they are downscaled server-side before it looks — so the width was being carried and counted and
+        then thrown away. See downscale.ts.
+
+        Both, deliberately: the disk keeps what was captured, so `Read` can still open it at full size.
+      */
+      const [res, dataUrl] = await Promise.all([
+        client.post<{ filename: string; dataUrl: string; attachmentId: string }>('/upload', formData),
+        downscaleForPrompt(file),
+      ]);
      setAttachments((prev) =>
-        prev.map((a, i) =>
-          i === idx ? { ...a, dataUrl: res.dataUrl, attachmentId: res.attachmentId, loading: false } : a,
-        ),
+        prev.map((a, i) => (i === idx ? { ...a, dataUrl, attachmentId: res.attachmentId, loading: false } : a)),
      );
    } catch {
      setAttachments((prev) => prev.filter((_, i) => i !== idx));
@@ -21,6 +21,14 @@ type FileTreeProps = {
   * an agent had just written to showing the listing from before its turn.
   */
  refreshToken?: number;
+  /**
+   * A file to reveal: every directory on the way to it opens, loading as it goes.
+   *
+   * For a tree whose selection is driven from OUTSIDE — a nav beside it, a URL — where a folder left
+   * shut around the open file hides the one thing being looked at. Expanding is still disclosure and
+   * still remembered, so anything opened by hand stays open.
+   */
+  revealPath?: string | null;
 };

 type TreeNodeProps = {
@@ -31,6 +39,7 @@ type TreeNodeProps = {
  searchForFile?: ((path: string) => string) | null;
  depth: number;
  refreshToken?: number;
+  revealPath?: string | null;
 };

 const FileIcon = ({ name }: { name: string }) => {
@@ -44,7 +53,16 @@ const sortEntries = (entries: DirEntry[]) => {
  return [...dirs, ...files];
 };

-const TreeNode = ({ entry, parentPath, root, onOpenFile, searchForFile, depth, refreshToken }: TreeNodeProps) => {
+const TreeNode = ({
+  entry,
+  parentPath,
+  root,
+  onOpenFile,
+  searchForFile,
+  depth,
+  refreshToken,
+  revealPath,
+}: TreeNodeProps) => {
  const [expanded, setExpanded] = useState(false);
  const [children, setChildren] = useState<DirEntry[] | null>(null);
  const [loading, setLoading] = useState(false);
@@ -73,6 +91,32 @@ const TreeNode = ({ entry, parentPath, root, onOpenFile, searchForFile, depth, r
    };
  }, [refreshToken]);

+  /*
+    On the way to the revealed file: open, and load if this is the first time.
+
+    Only ever opens. A node that closes itself when the selection moves elsewhere would collapse the
+    tree under somebody as they clicked around it.
+  */
+  const onTheWay = !!revealPath && isDir && revealPath.startsWith(`${fullPath}/`);
+
+  useEffect(() => {
+    if (!onTheWay) return;
+    setExpanded(true);
+    if (children !== null) return;
+
+    let alive = true;
+    listDirRef
+      .current(fullPath)
+      .then((res) => alive && setChildren(sortEntries(res.entries)))
+      .catch(() => alive && setChildren([]));
+    return () => {
+      alive = false;
+    };
+    // `children` is deliberately absent: this must run when the PATH changes, not when the load it
+    // started completes.
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [onTheWay, fullPath]);
+
  const handleClick = async () => {
    if (!isDir) {
      onOpenFile(fullPath, entry.name);
@@ -140,6 +184,7 @@ const TreeNode = ({ entry, parentPath, root, onOpenFile, searchForFile, depth, r
              searchForFile={searchForFile}
              depth={depth + 1}
              refreshToken={refreshToken}
+              revealPath={revealPath}
            />
          ))}
        </div>
@@ -148,7 +193,7 @@ const TreeNode = ({ entry, parentPath, root, onOpenFile, searchForFile, depth, r
  );
 };

-export const FileTree = ({ root, basePath, onOpenFile, searchForFile, refreshToken }: FileTreeProps) => {
+export const FileTree = ({ root, basePath, onOpenFile, searchForFile, refreshToken, revealPath }: FileTreeProps) => {
  const { t } = useTranslation();
  const { listDir } = useFilesAPI(root);
  const [entries, setEntries] = useState<DirEntry[] | null>(null);
@@ -208,6 +253,7 @@ export const FileTree = ({ root, basePath, onOpenFile, searchForFile, refreshTok
          searchForFile={searchForFile}
          depth={0}
          refreshToken={refreshToken}
+          revealPath={revealPath}
        />
      ))}
    </div>
@@ -5,6 +5,9 @@ import { CodeEditorHeader } from './CodeEditorHeader';
 import { panelName } from '../../locales';

 export { CodeEditorView } from './CodeEditor';
+// The tree on its own, for a panel that wants the editor's file list against its own directory rather
+// than the whole editor. It only LISTS — opening is the caller's callback.
+export { FileTree } from './FileTree';

 export const appRegistryMetas: AppRegistryMeta[] = [
  {
@@ -1,3 +1,4 @@
+import { useEffect, useRef } from 'react';
 import { Breadcrumb } from './components/Breadcrumb';
 import { Toolbar } from './components/Toolbar';
 import { UploadProgress } from './components/UploadProgress';
@@ -35,6 +36,14 @@ type FileBrowserAppProps = {
   * handling reimplemented, all of it free to drift from the real browser.
   */
  onPickFile?: (file: PickedFile) => void;
+  /**
+   * The folder being shown, whenever it changes.
+   *
+   * For a chooser whose answer is a FOLDER rather than a file. Folders navigate rather than being
+   * pickable — getting to a file requires it — so the only way to say "this one" is to be standing in
+   * it, and the caller needs to know which one that is to put a button on it.
+   */
+  onPathChange?: (path: string) => void;
 };

 export const FileBrowserApp = ({
@@ -45,9 +54,18 @@ export const FileBrowserApp = ({
  panelId,
  pathParam,
  onPickFile,
+  onPathChange,
 }: FileBrowserAppProps) => {
  const fileBrowserManager = useFileBrowserApp(basePath, rootOverride, urlPath, panelId, pathParam, onPickFile);
-  const { handleNavigate } = fileBrowserManager;
+  const { handleNavigate, currentPath } = fileBrowserManager;
+
+  // Reported from here rather than threaded through the hook: the hook's job is the folder, and one
+  // more positional argument on a six-argument call is how a signature stops being readable.
+  const report = useRef(onPathChange);
+  report.current = onPathChange;
+  useEffect(() => {
+    report.current?.(currentPath);
+  }, [currentPath]);

  return (
    <div className="flex flex-col h-full overflow-hidden">
@@ -1,6 +1,7 @@
 import { useState, useEffect, useRef } from 'react';
 import {
  Folder,
+  Link2,
  Trash2,
  Pencil,
  MoreVertical,
@@ -631,6 +632,21 @@ export const FileItem = ({
  const clickTimer = useRef<ReturnType<typeof setTimeout> | null>(null);
  const isDir = entry.type === 'directory';

+  /*
+    A link, and where it goes.
+
+    Drawn as an arrow beside the name rather than as a badge or a different icon: the icon already says
+    what the thing IS — a folder, a photograph — and a symlink to a folder is still a folder to anyone
+    using it. What the arrow adds is that there is an indirection in the way, which is the thing that
+    explains why deleting it did not free any space, or why two places show the same file.
+
+    The target goes in `title`. It is often long and absolute, so it belongs in a tooltip rather than in
+    a row that has to fit a name, a type, a size and a date.
+  */
+  const linkLabel = entry.symlink
+    ? `${entry.broken ? 'Broken symlink' : 'Symlink'} → ${entry.symlink}`
+    : undefined;
+
  const icon = isDir ? (
    <Folder className="h-5 w-5 text-duck-yellow fill-duck-yellow/30" />
  ) : (
@@ -820,7 +836,18 @@ export const FileItem = ({
                  onCancel={() => setRenaming(false)}
                />
              ) : (
-                <span className="text-sm font-medium text-duck-dark truncate block">{entry.name}</span>
+                <span className="flex min-w-0 items-center gap-1">
+                  <span className="truncate text-sm font-medium text-duck-dark">{entry.name}</span>
+                  {linkLabel && (
+                    <span
+                      title={linkLabel}
+                      aria-label={linkLabel}
+                      className={`shrink-0 leading-none ${entry.broken ? 'text-destructive' : 'text-duck-dark/45'}`}
+                    >
+                      <Link2 className="h-3.5 w-3.5" />
+                    </span>
+                  )}
+                </span>
              )}
            </div>
            <span className="hidden md:inline text-xs text-duck-dark/50 shrink-0 w-20 text-right">
@@ -886,7 +913,18 @@ export const FileItem = ({
                onCancel={() => setRenaming(false)}
              />
            ) : (
-              <span className="w-full truncate text-center text-sm font-medium text-duck-dark">{entry.name}</span>
+              <span className="flex w-full min-w-0 items-center justify-center gap-1">
+                <span className="truncate text-center text-sm font-medium text-duck-dark">{entry.name}</span>
+                {linkLabel && (
+                  <span
+                    title={linkLabel}
+                    aria-label={linkLabel}
+                    className={`shrink-0 leading-none ${entry.broken ? 'text-destructive' : 'text-duck-dark/45'}`}
+                  >
+                    <Link2 className="h-3.5 w-3.5" />
+                  </span>
+                )}
+              </span>
            )}

            <span className="text-xs text-duck-dark/50">
@@ -213,7 +213,11 @@ export const VideoDownloadDialog = ({ fileBrowserManager }: VideoDownloadDialogP
            </div>
          </form>
        ) : (
-          <div className="flex flex-col gap-3">
+          // `min-w-0` on the DIRECT grid child. DialogContent is a grid, and a grid item defaults to
+          // `min-width: auto` — so this is the element that grows to fit a long title and takes the
+          // card out of the dialog with it. The constraint has to be here; on the scroll container
+          // inside it was one level too low to stop anything.
+          <div className="flex min-w-0 flex-col gap-3">
            <div className="flex items-center justify-between gap-2">
              <button
                type="button"
@@ -248,7 +252,19 @@ export const VideoDownloadDialog = ({ fileBrowserManager }: VideoDownloadDialogP
              </div>
            )}

-            <div className="flex max-h-[52vh] flex-col gap-2 overflow-y-auto pr-1">
+            {/*
+              `min-w-0` is the one that makes `truncate` work, three levels down.
+
+              DialogContent is a GRID, and a grid item defaults to `min-width: auto` — it refuses to
+              shrink below its content. So a long title pushed this container wider than the dialog's
+              own `max-w-lg`, the card grew with it, and the `truncate` on the title never had a width
+              to truncate against. The text simply ran out of the modal.
+
+              Making the card a column fixed a different half of this and was not enough: a column still
+              cannot be narrower than the widest thing a grid will let it be. The constraint has to come
+              from above, and this is the highest point where it was missing.
+            */}
+            <div className="flex max-h-[52vh] min-w-0 flex-col gap-2 overflow-y-auto pr-1">
              {entries.map((e, i) => (
                <div
                  key={`${e.url}-${i}`}
@@ -534,9 +534,11 @@ export const PluginDetail = () => {
                how you teach someone to stop pressing it. */}
            {plugin.updateable && (
              <Button disabled={busy} onClick={() => run('update', plugin.appName)}>
-                {plugin.releaseReason === 'new-release' && plugin.newestTag
-                  ? `Update to ${plugin.newestTag}`
-                  : 'Update'}
+                {/* Named whenever a release is known, not only on `new-release`. An untagged checkout
+                    reports `untagged`, and it used to lose the destination with it — the button said
+                    "Update" and the owner had to guess where it landed. The marketplace knows the newest
+                    release even for a checkout that has never fetched it. */}
+                {plugin.newestTag ? `Update to ${plugin.newestTag}` : 'Update'}
              </Button>
            )}
            {/* Said out loud rather than shown as an absent button. "No Update button" and "up to date"
@@ -205,6 +205,20 @@ export type DirEntry = {
   * response, which simply means no thumbnail — the file icon, exactly as before.
   */
  thumb?: 'image' | 'video' | null;
+  /**
+   * Where this points, when it is a symlink — exactly as the link was written.
+   *
+   * Relative targets stay relative. `../originals` and `/home/…/originals` behave differently when the
+   * folder moves, and resolving one into the other would hide the difference at the moment somebody is
+   * looking to understand it.
+   *
+   * `type` still describes the TARGET, so a link to a folder opens like a folder. This says only that
+   * there is a link in the way. Absent on an older response, which reads as "not a link" — the same
+   * thing every entry said before.
+   */
+  symlink?: string | null;
+  /** The target is gone. The row is drawn anyway, because the link itself is really there. */
+  broken?: boolean;
 };

 type ListDirResponse = {
@@ -36,7 +36,7 @@ export { SessionList, ChatDetailPanel } from './apps/ChatHistory';
 export type { SelectedSession } from './apps/ChatHistory';
 // The chat URL vocabulary, so the /chat screen, the panels and the server all spell a group the same way.
 export { chatListPath, chatNewPath, chatSessionPath, cwdFromSplat, GROUP_SEGMENT } from './apps/ChatHistory';
-export { CodeEditorView } from './apps/CodeEditor';
+export { CodeEditorView, FileTree } from './apps/CodeEditor';

 // Same for /system-monitor, where the param is the scope rather than a section.
 export { DEFAULT_MONITOR_SCOPE, monitorScopePath, isMonitorScope } from './apps/SystemMonitor/shared';
@@ -2,6 +2,15 @@ export { useSettings, DEFAULT_SETTINGS } from './useSettings';
 export type { UseSettingsType, UserSettings, UserState } from './useSettings';
 export { useUserState } from './useUserState';
 export { useDashboardState } from './useDashboardState';
+/*
+  Exported for a workspace nested INSIDE a panel — see plugins/fofos-admin/web/WorkPanel.tsx.
+
+  `useDashboardState` persists the whole tree, which is right for a dashboard somebody arranged and
+  wrong for a layout its author ships: the structure would be frozen at whatever it was the first time
+  a screen was opened. `useAuthoredLayout` keeps the structure in source and persists only sizes and
+  zoom, which is what a nested authored layout wants for the same reason `PluginScreen` does.
+*/
+export { useAuthoredLayout } from './useAuthoredLayout';
 export { useModels, useVisibleModels, useUserVisibleModels, useEnabledModels, modelKey } from './useModels';
 export type { ModelOption } from './useModels';
 export { useAccessPolicy } from './useAccessPolicy';
@@ -204,6 +204,51 @@
    background-attachment: fixed;
    background-size: cover;
  }
+
+  /*
+    NATIVE CONTROLS FOLLOW THE THEME.
+    ------------------------------------------------------------------------------------------
+
+    Nothing here set `color-scheme`, so the browser drew every control it renders itself as if the page
+    were light — on a dark theme, a `<select>`'s dropdown opened as a white sheet with the page's own
+    pale text on it. Unreadable, and unreachable by any utility class, because that list is painted by
+    the browser and is not in the document.
+
+    `color-scheme` is the only thing that reaches it. It also fixes, for free, the other controls with
+    no styleable interior: date and colour pickers, the spinner on a number input, and default
+    scrollbars where a theme has not replaced them.
+
+    ── Why the `dark` class decides it ──
+
+    It is the signal that exists. 203 of the 231 themes set `applyDarkClass`, and the provider puts the
+    class on <html> beside `data-theme`, so the two are always in agreement about the same element.
+    Both are specificity 0-1-0 and themes are imported after this file, so a theme that disagrees can
+    simply say so — this is a default, not a decision taken away.
+
+    ── And the explicit colours underneath ──
+
+    `color-scheme` gives the browser's generic dark, not the theme's. Where a browser honours colours on
+    `<option>` — Firefox, and Chrome on Linux and Windows — these make the list the theme's own popover
+    surface instead. Where it does not (Safari, Chrome on macOS, which use a genuinely native menu), the
+    scheme above is what carries it, and generic dark is a great deal better than white-on-white.
+
+    A utility still wins: these are base-layer rules, so `bg-transparent` on a select behaves exactly as
+    it did.
+  */
+  :root {
+    color-scheme: light;
+  }
+
+  .dark {
+    color-scheme: dark;
+  }
+
+  select,
+  option,
+  optgroup {
+    background-color: var(--popover);
+    color: var(--popover-foreground);
+  }
 }

 /*
@@ -149,7 +149,20 @@ export function useThemeKeys(options: ThemeKeysOptions = {}) {
        would be indefensible.
      */
      const target = event.target as HTMLElement | null;
-      const typing = Boolean(target && (target.isContentEditable || /^(input|textarea|select)$/i.test(target.tagName)));
+      /*
+        A CODE EDITOR is typing too, and it does not look like it.
+
+        Monaco's modern input surface is neither a textarea nor contenteditable — it is a plain div the
+        EditContext API writes into — so a tagName test misses it entirely and `ctrl+shift+←/→` cycled
+        the theme while somebody was selecting words in a file. `.monaco-editor` is the class its root
+        always carries; `closest` because the event lands on whatever is inside it.
+      */
+      const typing = Boolean(
+        target &&
+          (target.isContentEditable ||
+            /^(input|textarea|select)$/i.test(target.tagName) ||
+            target.closest?.('.monaco-editor, [data-typing]')),
+      );

      /*
        Star or unstar the CURRENT theme, from anywhere, list open or not.
@@ -0,0 +1,159 @@
+---
+id: backburner/m3e-canvas-as-a-plugin
+title: "M3E Canvas as a plugin — sketch a UI, hand the prompt to the agent"
+theme: backburner
+status: open
+created: 2026-09-04
+date_source: stated
+order: 4
+---
+
+**M3E Canvas as a plugin.** [github.com/lnkiai/m3e-canvas](https://github.com/lnkiai/m3e-canvas) — a
+browser design tool for Material 3 Expressive screens that exports the design as a natural-language
+prompt for an AI coding tool. MIT.
+
+The upstream question is now **answered** — see below; vendoring is the path. The rest is still open.
+This card holds the measurements so the decision is made from numbers rather than from a README. Everything below was read from the code at **`cfde698`, 2026-09-04**.
+
+## Why it is interesting
+
+Its output is *a prompt you paste into Claude*. Officer is the thing that removes the paste — the export
+button could open a chat session with the prompt already in it, pointed at a project directory on this
+machine. **That last step is the only part Officer uniquely adds, and it is the whole reason to bother.**
+The canvas already works standalone today.
+
+## What it actually is
+
+| | |
+| --- | --- |
+| size | 65 files, **14,822 lines** of TS/TSX |
+| deps | `next` 16.3.4, `react` 19.2.8, `motion` ^13.1.1, `html-to-image` ^1.11.13 |
+| licence | MIT |
+| storage | `localStorage` only, no backend |
+
+**Next.js is not actually there.** The entire codebase imports it **once**, and it is a type-only import:
+`import type { Metadata, Viewport } from "next"` in `app/layout.tsx`. Eighteen files carry `'use client'`.
+So it is a client-side React SPA that happens to be scaffolded in Next — there is nothing to strip, and
+porting means giving it a different mount point.
+
+That is the finding that changes the estimate. An earlier read of the README concluded "weeks of work,
+own a fork forever", and that was wrong.
+
+## The fork line falls on a file boundary
+
+`app/page.tsx` imports about **twenty modules, every one a named export** through the `@/` alias, and
+reaches into no internals. `lib/tokens.ts` is the shared spine with 16 importers. `app/` holds exactly
+one file that matters.
+
+| | lines | |
+| --- | --- | --- |
+| `app/page.tsx` | 3,433 | the whole editor: 15 components, 103 hook calls. **Yours.** |
+| `lib/ai.ts` | ~200 | **Delete it** — stores provider API keys in `localStorage`. Officer has a secret store and the agent. |
+| `lib/tokens.ts` | 1,614 | Material 3 tokens. Data. Vendor. |
+| `components/M3Node.tsx` | 1,233 | the component renderer. Vendor. |
+| `components/Inspector.tsx` | 1,171 | Vendor. |
+| `lib/prompt.ts` | 1,151 | the export engine. Vendor. |
+| `components/Preview.tsx` | 899 | Vendor. |
+| `components/ui.tsx` | 882 | Vendor. |
+| rest | ~4,200 | `Mobile`, `i18n`, `tidy`, `Toolbar`, `color`, `shapes`, `theme`, `project`. Vendor. |
+
+**~3,600 lines owned, ~11,200 vendored** — and the split means something. What is vendored is *"what a
+Material 3 component looks like and how to describe it in words"*, where upstream's value accumulates and
+we have no wish to innovate. What is owned is *"where does this live in the workspace, whose is it, and
+what happens on export"*, where upstream can never help because it does not know Officer exists.
+
+**So deep integration and tracking upstream are not in tension.** The integration work lives entirely in
+the one file that was always going to be rewritten.
+
+## The export point is clean
+
+```ts
+buildPrompt(doc: Doc, widths: Record<string, number>, onlyFrameId?: string, lang: Lang): string
+effectivePrompt(doc, widths, lang): string
+```
+
+Two pure functions in `lib/prompt.ts`. `Doc` in, string out, no side effects. No `postMessage` bridge and
+no clipboard interception needed — **call the function.** Today it ends in
+`navigator.clipboard.writeText` in three places; that is what gets replaced.
+
+`Doc` is plain serialisable JSON (`lib/project.ts` already downloads and reloads it), so persisting to a
+plugin table instead of `localStorage` is nearly free. Current keys: `DOC_KEY` and `UI_KEY` in
+`page.tsx`, `m3e:ai`, and `m3e:sec:<id>`.
+
+## The upstream problem, measured
+
+This is the part that makes it a backburner card rather than a build.
+
+- **37 commits, 3 days old** — first 2026-09-02, and **22 of them landed on 2026-09-04**, the day this
+  was read.
+- **Two human authors** (`lnkiai` 25, `CQMHV` 10) plus dependabot.
+
+It is under heavy active development *right now*. The export surface will churn, and no amount of
+discipline on our side changes that. "Vendor it and track upstream" assumes an upstream that has settled,
+and this one has not started.
+
+### If it is vendored anyway
+
+`web/vendor/m3e/`, committed with upstream URL, commit SHA and date in a README beside it. **The
+precedent is already in the tree** — `plugins/jellyfin/web/vendor/` and `public/vendor/pako/`.
+
+The piece that makes it survivable is **an adapter**: Officer's panels import from one `adapter.ts` that
+re-exports from `vendor/`, never from `vendor/` directly. An upstream rename then costs one file instead
+of twenty. Updating is: re-copy, read the diff, fix the seam.
+
+## Asked upstream, 2026-09-04. Declined — vendoring is the path
+
+[Issue #55](https://github.com/lnkiai/m3e-canvas/issues/55) asked for the seam;
+[PR #56](https://github.com/lnkiai/m3e-canvas/pull/56) implemented the smaller half of it. Both closed
+unmerged on 2026-09-05. The reply is worth reading in full, but the two reasons were:
+
+1. **The PR did not actually unblock embedding.** It wired one of the three clipboard writes —
+   `app/page.tsx` — and left `components/Inspector.tsx:443` and `components/PromptPanel.tsx:35`
+   writing directly. So it "adds a public surface I'd have to keep stable" without delivering the
+   thing it was for. **That was our error, not a judgement call**: the issue text named all three
+   sites and the PR wired one.
+2. **Embedding is not on their roadmap, and it is a single-user need so far.** Fair on both counts.
+
+**What they said they would want** if that changes, quoted because it is the specification for any
+future attempt:
+
+> If I decide to support embedding I'd want to do it in one piece: the `Canvas` extraction plus the
+> adapter wired into every export path. Your design for the adapter is sound and I'd start from it
+> then.
+
+So the adapter design survives — it is validated by the person who would have to maintain it — and the
+work it needs is known: **the file move plus all three export paths**, not one of them.
+
+The branch is on the fork at [andrepadez/m3e-canvas `embeddable-canvas`](https://github.com/andrepadez/m3e-canvas/tree/embeddable-canvas),
+two commits, typecheck and build green. Worth keeping: if this is vendored, that is the diff to apply
+on top, extended to the other two call sites.
+
+### What this settles
+
+**Vendor, do not depend.** The upstream door is open in principle and shut in practice, so the fork is
+the only route to the integration — which is what the split below was already designed for. Nothing
+about the file-boundary analysis changes; only the hope of upstreaming the shell does.
+
+## If it is picked up anyway
+
+**Open an issue asking upstream to make the seam explicit.** If `page.tsx` grew an entry point taking the
+persistence layer and the export sink — `<Canvas storage={…} onExport={…} />` — we would **depend rather
+than fork**, and the 3,600 owned lines shrink to a few hundred.
+
+**Answered: no, for now.** Recorded above. Asking cost an afternoon and bought a specification for
+what upstream would accept, plus confirmation that the adapter shape is right — which is a good return
+on a declined PR.
+
+## Open questions
+
+1. **Is this a design tool that lives in Officer, or the front end of "sketch it and the agent builds
+   it"?** If the second, `page.tsx` is where we deliberately diverge, and adopting the fork is the right
+   frame from day one rather than something discovered in six months.
+2. **Whose designs are they?** `Doc` is per-user data. Multi-user makes that a permissions question, and
+   the plugin table answers it in a way `localStorage` cannot.
+3. **Does the phone editor survive the port?** `components/Mobile.tsx` (534 lines) is a separate
+   phone-optimised editor. Officer's mobile story is a native app, so this may be redundant or may be the
+   best part.
+4. **An iframe was considered and rejected.** It buys a day and costs the function call, the theming, the
+   persistence, and forces back the clipboard round-trip the whole idea exists to remove. Recorded so it
+   is not re-proposed as the cheap option.