Tina CMS Integration Log

Session: 2026-02-28, ~14:20–15:00

What Was Done

1. Root package.json created

• Path: ~/Obsidian/baseworks-kb-shared-brain/package.json
• Added tinacms (^2.6.0) and @tinacms/cli (^1.6.0) as devDependencies
• Scripts: npm run dev wraps Tina + Astro together, npm run build runs Tina build then Astro build

2. Tina config created

• Path: tina/config.ts at repo root
• 5 collections mapping to vault directories:

Collection	Path	Extra Fields
Inbox	00-inbox/	—
Projects	01-projects/	status
Areas	02-areas/	format, location, program_dates, sessions
Resources	03-resources/	type
Archive	04-archive/	—

• All collections share: title, created, updated, tags, body
• Body uses raw string type (textarea) instead of rich-text — preserves wikilinks, tables, callouts as-is
• Areas collection has match.exclude: "**/*[(]*[)]*" to skip files with parentheses in paths (Tina limitation)

3. Issues encountered and resolved

Parentheses in file paths

• Files like 2026 (Spring) Study Group Montreal/ crash Tina’s indexer
• Renaming would break ~40 wikilinks across the vault
• Solution: match.exclude glob using character classes [(] and [)] (standard glob parentheses cause regex backtracking)

Numeric YAML tag values

• tags: [event, tokyo, 2026, intensive] — YAML parses bare 2026 as an integer
• Tina calls .replace() on tag values, which fails on numbers
• Fix: Quoted the numeric values in 2 files:
  • 02-areas/educational-programs/events/2026/2026-01-tokyo-intensive/2026-01-tokyo-intensive.md
  • 02-areas/educational-programs/events/2026/2026-01-tokyo-intensive/sessions/day-1-session-1.md

rich-text body parser limitations

• Tina’s rich-text type can’t parse complex Obsidian markdown (tables with wikilinks, pipe characters)
• Switched all body fields to string with ui: { component: "textarea" }
• Trade-off: Tina editor shows raw markdown instead of a WYSIWYG editor, but content is preserved perfectly

4. Tina Cloud setup

• Project created at app.tina.io under “Patrick Oancia’s Organization”
• Connected to p-oancia/baseworks-kb-shared-brain
• GitHub Authoring: “Act as self” (commits attributed to the actual editor)
• Co-authoring enabled (Patrick Oancia / pat@baseworks.com)
• Client ID: 0774d998-f43e-407b-8f2f-38d811a94b18
• Token type: Content (Read-only) — sufficient for build-time API fetching

5. GitHub Actions auto-deploy

• Updated .github/workflows/deploy.yml to include Tina build step
• Pipeline: Checkout → Node 20 → npm ci (root) → npm ci (site/) → tinacms build → astro build → wrangler pages deploy
• Secrets added to GitHub repo:
  • TINA_PUBLIC_CLIENT_ID
  • TINA_TOKEN
  • CLOUDFLARE_API_KEY (pre-existing)
  • CLOUDFLARE_EMAIL (pre-existing)
  • CLOUDFLARE_ACCOUNT_ID (pre-existing)
• First deploy succeeded (run #22527910907)

6. .gitignore updated

• Added: tina/__generated__/, node_modules/, .env
• Tina also generates its own .gitignore in site/public/admin/

7. Local .env created

• Path: ~/Obsidian/baseworks-kb-shared-brain/.env (not committed)
• Contains TINA_PUBLIC_CLIENT_ID and TINA_TOKEN

Current State

• Admin panel deployed at kb.baseworks.com/admin/index.html (behind Cloudflare Access)
• Local dev: cd ~/Obsidian/baseworks-kb-shared-brain && npm run dev starts both Tina + Astro
• Auto-deploy triggers on every push to master
• Backup tag: pre-tina-backup at commit 1cf6a7e (pre-Tina state)

Files Created/Modified

File	Action
package.json (root)	Created
package-lock.json (root)	Created (by npm install)
tina/config.ts	Created
tina/tina-lock.json	Created (by tinacms)
.env	Created (local only, not committed)
.gitignore	Modified (added Tina entries)
.github/workflows/deploy.yml	Modified (added Tina build step)
02-areas/.../2026-01-tokyo-intensive.md	Modified (quoted numeric tag)
02-areas/.../day-1-session-1.md	Modified (quoted numeric tag)
site/public/admin/index.html	Generated (by tinacms, git-ignored)

Post-deploy Fixes: Admin 404

Two issues prevented the admin panel from loading on the production domain:

Fix 1: Nested .gitignore (commit 47a40d6)

Tina generates a .gitignore inside site/public/admin/ that excludes index.html and assets/. Astro copies this into site/dist/admin/.gitignore. Wrangler respects .gitignore during deploy, silently skipping the admin files.

Fix: Added a step to .github/workflows/deploy.yml that removes site/dist/admin/.gitignore before the wrangler deploy step.

Fix 2: Wrong deploy branch (the actual root cause)

The deploy command used --branch=main but the Cloudflare Pages project has production_branch: "master". This meant all deployments went to the preview environment, not production. The custom domain kb.baseworks.com only serves production deployments.

Diagnosis: baseworks-kb.pages.dev/admin/index.html returned 404 but main.baseworks-kb.pages.dev/admin/index.html returned 200 — confirming content was deployed to preview only.

Fix: Changed --branch=main to --branch=master in .github/workflows/deploy.yml.

Verified working

• baseworks-kb.pages.dev/admin/index.html — HTTP 200
• kb.baseworks.com/admin/index.html — HTTP 200 (behind Cloudflare Access)
• Admin JS/CSS bundles load correctly (production build, not dev)

Remaining Work (Phase 2+)

• Test creating/editing a document through Tina admin
• Test creating/editing a document through Tina admin
• Consider adding Ksenia/Asia as a Tina Cloud collaborator
• Monitor for any frontmatter reordering issues after Tina edits
• Eventually: rename files with parentheses to remove the match.exclude workaround

Rollback

cd ~/Obsidian/baseworks-kb-shared-brain
git reset --hard pre-tina-backup
git push --force

This restores everything to commit 1cf6a7e (the known-good state before Tina).