Prepare SourceBraid 1.0.1 Chrome and ChatGPT release

.agents/plugins/marketplace.json

@@ -0,0 +1,20 @@
+{
+  "name": "sourcebraid",
+  "interface": {
+    "displayName": "SourceBraid"
+  },
+  "plugins": [
+    {
+      "name": "sourcebraid",
+      "source": {
+        "source": "local",
+        "path": "./codex-plugin/sourcebraid"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity"
+    }
+  ]
+}

.github/workflows/release-checks.yml

@@ -0,0 +1,59 @@
+name: Release checks
+
+on:
+  pull_request:
+    branches: [main]
+  push:
+    branches: [main]
+    tags: ["v*"]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: release-checks-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  verify:
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v7
+      - uses: actions/setup-python@v7
+        with:
+          python-version: "3.12"
+      - uses: actions/setup-node@v7
+        with:
+          node-version: "22"
+          package-manager-cache: false
+      - name: Verify shared capture behavior
+        run: node --test tests/capture-utils.test.js
+      - name: Install locked ChatGPT MCP dependencies
+        working-directory: chatgpt-mcp
+        run: npm ci
+      - name: Test ChatGPT MCP with synthetic data and mocked GitHub responses
+        working-directory: chatgpt-mcp
+        run: npm test
+      - name: Validate the ChatGPT MCP deployment build without publishing
+        working-directory: chatgpt-mcp
+        env:
+          WRANGLER_SEND_METRICS: "false"
+        run: npm run build
+      - name: Verify Python behavior and package boundaries
+        run: python3 -m unittest discover -s tests -p "test_*.py"
+      - name: Build and verify reproducible submission bundle
+        run: |
+          if [ "$GITHUB_REF_TYPE" = "tag" ]; then
+            python3 scripts/build_release.py --verify-reproducible --require-clean --expected-version "${GITHUB_REF_NAME#v}"
+          else
+            python3 scripts/build_release.py --verify-reproducible --require-clean
+          fi
+      - name: Keep reviewed artifacts for download
+        uses: actions/upload-artifact@v7
+        with:
+          name: sourcebraid-submission-bundle-${{ github.sha }}
+          path: dist/sourcebraid-release-v*/
+          if-no-files-found: error
+          retention-days: 30

.gitignore

@@ -6,7 +6,22 @@ xcuserdata/
 ios/build/
 
 # Release archives are published separately.
-dist/*.zip
+dist/
+
+# Private website source is maintained outside public releases.
+website/
+
+# Hosted connector development and runtime secrets.
+chatgpt-mcp/node_modules/
+chatgpt-mcp/.wrangler/
+chatgpt-mcp/.dev.vars
+chatgpt-mcp/.env
+chatgpt-mcp/.env.*
+!chatgpt-mcp/.env.example
+
+# Generated setup coordinates are local configuration.
+sourcebraid-config.json
+sourcebraid-openai-review-config.json
 
 # Personal archives belong in a separate private repository.
 web-clips/

NOTICE

@@ -15,3 +15,9 @@ licenses continue to apply and are not replaced by the SourceBraid license.
 Captured articles, papers, PDFs, images, and other user-supplied source material
 are not part of SourceBraid and are not licensed under the project's MIT
 License.
+
+The separately deployed chatgpt-mcp service uses
+@cloudflare/workers-oauth-provider under its MIT License. Development tooling
+and transitive dependencies retain the licenses distributed with their npm
+packages. Consult chatgpt-mcp/package-lock.json for the pinned dependency tree;
+these dependencies are not relicensed by SourceBraid.

PRIVACY.md

@@ -1,23 +1,65 @@
 # SourceBraid Privacy
 
-Last updated: July 31, 2026
+Last updated: September 12, 2026
 
 SourceBraid is open-source software that stores captured sources in a GitHub
-repository selected and controlled by the user. SourceBraid does not operate a
-central content-storage or analytics service.
+repository selected and controlled by the user. The Chrome/iOS capture clients,
+local Codex plugin, and prepared hosted ChatGPT service have different data
+flows, described below. The hosted service is not yet deployed; this notice
+describes the implementation prepared for its release.
 
 ## Data handled by SourceBraid
 
 - The Chrome extension and iOS app process the page, document, selected text,
   notes, tags, and images that the user chooses to save.
+- The Chrome extension reads the active page only after the user starts a Save
+  or Download capture and accepts the first-use disclosure. A fallback download
+  remains on the user's device.
 - Captured Markdown, metadata, original PDFs, and related assets are sent
   directly to the GitHub repository configured by the user.
-- GitHub credentials and SourceBraid configuration are stored locally by the
-  relevant client. Users should use a fine-grained GitHub token restricted to
-  the intended archive repository.
-- The ChatGPT and Codex plugin can read the configured archive, build a local
-  SQLite search cache, and perform an explicitly confirmed deletion through
-  GitHub.
+- Chrome/iOS credentials and configuration are stored by the relevant client
+  using browser storage or iOS Keychain. Users should restrict each fine-grained
+  GitHub token to the intended archive repository.
+- The local Codex plugin reads the configured archive, stores cached Markdown
+  and a disposable SQLite index locally, and can perform explicitly confirmed
+  deletion or metadata migration through GitHub. Its credentials come from the
+  user's local environment, configuration, or GitHub CLI session.
+- When a user invokes the plugin, requested search results, fetched archive
+  content, and operation results are made available to the active ChatGPT or
+  Codex environment and are processed under that service's terms and data
+  settings. Installing or publishing the plugin does not by itself transfer an
+  archive or make its repository public.
+
+## Prepared hosted ChatGPT service
+
+The separate service in `chatgpt-mcp/` is prepared for
+`https://mcp.sourcebraid.com/mcp`. Its deployment, live authentication tests,
+and public OpenAI review/publication remain release steps.
+
+- Connection uses OAuth 2.1 authorization codes with PKCE S256. The user enters
+  a fine-grained GitHub PAT on the SourceBraid HTTPS consent page, never in a
+  ChatGPT prompt. It must have Contents read-only access for exactly one private
+  repository. The service receives the PAT, repository, branch, and root folder
+  and verifies the GitHub identity and private repository before archive access.
+- The implementation uses `@cloudflare/workers-oauth-provider` version `0.10.3`.
+  The provider stores OAuth grant/client metadata in Cloudflare KV and encrypts
+  the GitHub PAT in its grant properties. The PAT is used server-side to call
+  GitHub and is not returned to ChatGPT tools or placed in public logs, listings,
+  screenshots, or release packages.
+- Requested search terms, repository metadata, Markdown, and results pass
+  through the SourceBraid Worker on Cloudflare and are processed transiently.
+  The implementation creates no permanent archive-content database or local
+  SQLite index in the service. OAuth connection records are separate from
+  archive content.
+- ChatGPT receives the requested search results and fetched sources and
+  processes them under OpenAI's terms and the user's applicable data settings.
+  Hosting on Cloudflare therefore adds a data-processing step compared with
+  the local Codex workflow; archive access is not wholly local.
+- Hosted tools are read-only. They do not save captures, modify or delete
+  GitHub content, perform metadata migration, or run GitHub workflows.
+- SourceBraid includes no advertising or archive-content analytics. Do not
+  enable request-body, token, search-query, or document logging when deploying
+  or operating the Worker.
 
 ## External services
 
@@ -26,12 +68,66 @@ the material selected by the user. GitHub Actions may run Docling to convert
 uploaded PDFs. Those services process data under their own terms and privacy
 policies.
 
+The hosted ChatGPT connection additionally uses Cloudflare Workers/KV and
+OpenAI. These providers process connection or requested content data according
+to their respective service terms and privacy policies. The GitHub repository
+and Git history remain the authoritative archive.
+
+If the user configures Ghost or Blogger source APIs, the Chrome extension sends
+the necessary request and configured API credential directly to that service.
+SourceBraid does not include GitHub tokens or optional API credentials in its
+exported plugin configuration.
+
 ## Control and deletion
 
 Users control the archive repository and its Git history. They can delete
-content through GitHub or through SourceBraid's preview-and-confirm deletion
-flow, revoke tokens at GitHub, clear local extension or app data, and delete
-the local search cache.
+content through GitHub or the local Codex plugin's preview-and-confirm deletion
+flow, clear local extension or app data, and delete the local search cache.
+Deleting a clip does not erase its earlier Git history.
+
+For the hosted connection, disconnect SourceBraid in ChatGPT and revoke its
+fine-grained PAT at GitHub to prevent further repository access. This does not
+delete the archive, local caches, or previously returned content in a ChatGPT
+conversation. Use the corresponding service controls for those copies. Contact
+the maintainer through the support page to request assistance with hosted
+connection records; never include a token or private archive content in a
+public issue.
+
+## Deutsch
+
+SourceBraid speichert Captures in einem vom Nutzer kontrollierten
+GitHub-Repository. Chrome und iOS schreiben direkt dorthin; ihre Zugangsdaten
+liegen im Browser-Speicher beziehungsweise iOS-Schlüsselbund. Das lokale
+Codex-Plugin nutzt lokale Zugangsdaten, Markdown-Kopien und einen neu
+aufbaubaren SQLite-Index. Es kann nach Vorschau und neuer ausdrücklicher
+Bestätigung Clips löschen oder Metadaten migrieren.
+
+Der gehostete ChatGPT-Dienst unter `https://mcp.sourcebraid.com/mcp` ist
+vorbereitet, aber noch nicht bereitgestellt oder öffentlich freigegeben. Nach
+Bereitstellung verbindet OAuth 2.1 mit PKCE S256 ChatGPT mit SourceBraid. Ein
+Fine-grained-GitHub-PAT mit Contents-Leserechten für genau ein privates
+Repository wird ausschließlich auf der SourceBraid-HTTPS-Zustimmungsseite
+eingegeben, niemals in ChatGPT. OAuth-Grant- und Client-Metadaten liegen in
+Cloudflare KV; der OAuth-Provider verschlüsselt den PAT in den Grant-Daten.
+Der PAT wird serverseitig für GitHub verwendet und nicht an ChatGPT-Werkzeuge
+ausgegeben.
+
+Angeforderte Suchbegriffe, Repository-Metadaten, Markdown und Ergebnisse werden
+vorübergehend durch SourceBraid/Cloudflare verarbeitet und als angeforderte
+Ergebnisse an ChatGPT übermittelt. Es gibt keine dauerhafte
+SourceBraid-Inhaltsdatenbank und keinen gehosteten SQLite-Index. Die Bedingungen
+und Dateneinstellungen von GitHub, Cloudflare und OpenAI gelten für die
+jeweilige Verarbeitung. Gehostete Werkzeuge haben ausschließlich Lesezugriff;
+sie speichern oder löschen keine GitHub-Dateien. SourceBraid enthält keine
+Werbe- oder Inhaltsanalysefunktionen. Tokens, Suchanfragen und Dokumentinhalte
+dürfen beim Betrieb nicht protokolliert werden.
+
+Zum Beenden des gehosteten Zugriffs SourceBraid in ChatGPT trennen und den PAT
+bei GitHub widerrufen. Das löscht weder das GitHub-Archiv und seine Historie
+noch lokale Caches oder bereits an ChatGPT übermittelte Inhalte. Dafür gelten
+die jeweiligen Löschfunktionen. Bei Fragen zu Verbindungsdaten hilft der
+Maintainer über die Supportseite; Tokens und private Inhalte gehören niemals
+in öffentliche Issues.
 
 Questions can be opened at
 https://github.com/patrickschiller/sourcebraid/issues.