Fix findings from the Crew merge QA campaign

.github/workflows/frontend.yml

@@ -196,15 +196,25 @@ jobs:
       # already encodes it — but nothing ran the linter, so two PRs merged in one
       # afternoon each added a file loose at `docs/` root.
       #
-      # ⚠ This is DELIBERATELY three rules, not the whole linter. The tree carries a
+      # ⚠ This is DELIBERATELY four rules, not the whole linter. The tree carries a
       # real backlog (156 findings across ten rules, measured 2026-09-22), and a gate
-      # that fails on arrival gets disabled rather than obeyed. These three are clean
+      # that fails on arrival gets disabled rather than obeyed. These four are clean
       # on main today and can be held that way; the step prints how many findings it
       # is NOT reporting, so a filtered pass never reads as "the docs are clean".
       # Widen the list as a rule's backlog reaches zero — never past it.
+      # tree/research-cites-repo joined on 2026-09-28, when Crew's own design and records left
+      # docs/research/ (which holds studies of systems outside this repo) at zero findings.
       - name: Documents live where the rule says (placement only)
         if: ${{ !cancelled() }}
-        run: python3 scripts/docs-lint.py --only tree/loose-at-root,tree/no-index,status/folder-disagreement
+        run: python3 scripts/docs-lint.py --only tree/loose-at-root,tree/no-index,status/folder-disagreement,tree/research-cites-repo
+
+      # The Crew manual and its landing page quote dialog titles, name what a
+      # `biorouter serve` browser can do and say who may add people. Each rule
+      # reads the code it describes, and the suite first shows every rule
+      # failing on the wording a 2026-09-27 audit found, then runs it on the tree.
+      - name: The Crew manual agrees with the code it describes
+        if: ${{ !cancelled() }}
+        run: node --test scripts/check-crew-manual.test.mjs
 
       # The published registry, the desktop fallback snapshot and the
       # compiled-in private set are one catalog in three forms. Being written by

.github/workflows/rust.yml

@@ -608,12 +608,20 @@ jobs:
 
           # The token is exchanged for a cookie and leaves the address bar. It is
           # not consumed (SD-9 in docs/deployment/serve-decisions.md): the
-          # readiness loop above has already redeemed it.
-          curl -s -o /dev/null -D "$RUNNER_TEMP/h" "http://127.0.0.1:18777/?t=citoken"
-          grep -qi '^HTTP/1.1 303' "$RUNNER_TEMP/h" || fail 'no redirect after the token exchange'
+          # readiness loop above has already redeemed it. The answer is a page of
+          # the daemon's own origin that moves on to /, not a 303: a redirect
+          # drops a SameSite=Strict cookie on a navigation that `serve --open`'s
+          # file: page started (routes::web_ui::exchange_bounce).
+          curl -s -o "$RUNNER_TEMP/bounce.html" -D "$RUNNER_TEMP/h" "http://127.0.0.1:18777/?t=citoken"
+          grep -qi '^HTTP/1.1 200' "$RUNNER_TEMP/h" || fail 'the token exchange did not answer its page'
           grep -qi 'set-cookie: biorouter_session=' "$RUNNER_TEMP/h" || fail 'no session cookie'
           grep -qi 'HttpOnly' "$RUNNER_TEMP/h" || fail 'the session cookie is readable from script'
           grep -qi 'SameSite=Strict' "$RUNNER_TEMP/h" || fail 'the session cookie is sent cross-site'
+          grep -qi 'cache-control: no-store' "$RUNNER_TEMP/h" || fail 'the exchange page is cacheable'
+          grep -qi 'referrer-policy: no-referrer' "$RUNNER_TEMP/h" || fail 'the exchange page leaks its address'
+          grep -q 'http-equiv="refresh" content="0;url=/"' "$RUNNER_TEMP/bounce.html" \
+            || fail 'the exchange page does not move on to the interface'
+          ! grep -q citoken "$RUNNER_TEMP/bounce.html" || fail 'the exchange page carries the token'
 
           # With the cookie, the shell and its runtime configuration.
           curl -fsS -b 'biorouter_session=citoken' http://127.0.0.1:18777/ > "$RUNNER_TEMP/shell.html"

README.md

@@ -67,13 +67,16 @@ Work with papers, genes, diseases, drugs, cohorts, and experimental data in the
 | **Analyze and visualize** | Write and run R or Python, fit machine-learning models, and create figures and interactive dashboards. |
 | **Work with desktop software** | Biorouter Copilot can inspect and operate applications on the backend computer, with your approval for each request. |
 | **Build and reuse tools** | Create small research apps with Agent Drafter, run parallel chats, and save or schedule recurring tasks as workflows. |
+| **Work with your lab** | In Crew, chat in channels, share files and give AI agents tasks the whole channel can see, in a workspace one lab member hosts on a shared Linux server. |
 
 Copilot keeps observations and approvals separate for each chat. The applications themselves share a desktop, so choose the model before opening sensitive records or research data. See [Copilot setup and limits](docs/extensions/built-in/computer-controller.md).
 
+Crew runs in the desktop app and in the `biorouter crew` commands, not in a `biorouter serve` browser. Each member connects with their own server account over SSH. Crew's Private and Public settings decide which AI models may read the workspace, and the stricter of your connection's and the workspace's setting applies. See the [Crew user manual](docs/crew/README.md).
+
 ## Get started
 
 1. **[Download Biorouter](https://biorouter.ucsf.edu/download)** for macOS, Windows, or Linux.
-2. **Choose a model.** UCSF users can select **Versa API Azure** or **Versa API Bedrock** under Institutional Models. For local inference, choose **Llama Server** and download a model. You can also connect a commercial provider.
+2. **Choose a model.** UCSF users can select **Versa API Azure** or **Versa API Bedrock** on the **Institutional** tab. For local inference, choose **Llama Server** and download a model. You can also connect a commercial provider.
 3. **Start a chat.** Add a paper or dataset, or install an agent from BAAM.
 
 For example, attach a few papers and ask:
@@ -96,6 +99,7 @@ Biorouter uses a shared Rust agent core, an Electron and React desktop app, and
 - [Documentation](docs/README.md): guides and references by topic.
 - [Architecture](docs/architecture/system-overview.md): how the components fit together.
 - [Knowledge bases and BioOKF](docs/knowledge-base/README.md), [workflows](docs/workflows/README.md), and [Apps SDK](docs/apps-sdk/sdk-reference.md).
+- [Crew user manual](docs/crew/README.md): joining or hosting a lab workspace, channels, files, agent tasks and server setup.
 - [Contributing](CONTRIBUTING.md): build from source, test, and submit changes.
 - [Report an issue](https://github.com/BaranziniLab/biorouter/issues). Report security concerns privately using [SECURITY.md](SECURITY.md).