Home / Ops / Housekeeping and Process

Affirmology Internal Resource Hub Build Brief v1

Updated Jun 26, 2026 · Affirmology_ResourceHub_BuildBrief_v1.md

Affirmology Internal Resource Hub Build Brief v1

For: a Claude Code session running on the Mac (has filesystem access to the AFFIRMOLOGY folder, the Cloudflare token, and can run shell + wrangler). How to use: paste this whole file as the kickoff. Read the reference file in section 9 before building.

0. What you are building (one paragraph)

A gated, internal HTML resource hub at resources.affirmology.ai for Jeff, Sol, and Colin. It renders the project's key documents as clean, navigable HTML (house style matching the reference playbook), organized by category, with the Project State front and center. A cron job on the Mac refreshes it every 1 to 2 hours: it notices new or changed docs in the folder, extracts a title, summary, category, and related docs from each, and regenerates and redeploys the site. Only the team (allowlisted by Cloudflare Access) can reach it.

1. Hard rules (do not break)

• Never touch the locked demo (demo.affirmology.ai) or its render settings.
• No em dashes anywhere, ever (Jeff's rule). Use periods, commas, or rewrite.
• Gated only. The site must never be publicly reachable. Cloudflare Access has to be in front before any content is live.
• Do not expose secrets. Skip and never render .env, .studio_token, any token or key file, and anything under /Volumes.
• Match the house style. Read the reference file (section 9) and reuse its CSS as the shared stylesheet. Same emerald/gold/cream language, cards, callouts, tables, table of contents.
• Incremental and idempotent. Re-running must not duplicate or thrash. Only changed docs get reprocessed (hash compare).
• Keep CLAUDE.md, MEMORY.md, and Affirmology_PROJECT_STATE.md at the repo root. Do not move them.
• Verify with evidence (open the built pages, test the Access gate in incognito) before claiming done.

2. Source content

• Root folder: /Users/jeffreyparker/CLAUDE/AFFIRMOLOGY
• Include: .md and curated .html documents in the root and in the doc taxonomy folders.
• Exclude: code directories (affirmology-agent/, affirmology-studio/, affirmology-app/, affirmology-atlas/, affirmology-demo-site/, affirmology-site/, affirmology-shared/), node_modules/, .git/, corpus/, baselines/, any binaries, build artifacts, and the secret files in section 1.

3. Categories (taxonomy)

Map every doc to exactly one. Mirror the CLAUDE.md taxonomy plus Gamification and a Start bucket:

Start here / Project State · Strategy · Product · Engine · Council · Video · Readings · Audio · Ops · Gamification · Archive

Classification: use filename and content heuristics first; for ambiguous docs, run a cheap LLM classify (Haiku) against this exact list.

4. Architecture

• Generate a static site into a new top-level folder: /Users/jeffreyparker/CLAUDE/AFFIRMOLOGY/resource-hub/ with a dist/ output, an assets/style.css (extracted from the reference), a manifest.json (state), and the generator script.
• Pages:
• index.html (home): categories grid, a prominent Project State link at the top, and a "Recently updated" list (last 10 by timestamp).
• One page per category: lists its docs with title, summary, and last-updated.
• One page per doc: rendered HTML body, breadcrumb, back-nav, last-updated, and a "Related" list.
• Navigation: persistent top nav with all categories on every page; breadcrumbs; visible "last updated" timestamps.
• Self-contained: inline the shared CSS or link assets/style.css. No external dependencies except a markdown library.

5. The generator script (the heart of it)

Write it in Python and reuse the agent venv and Anthropic key (/Users/jeffreyparker/CLAUDE/AFFIRMOLOGY/affirmology-agent/.venv, key in affirmology-agent/.env). On each run it:

1. Walk the source folder (section 2 rules), list candidate docs.
2. Load resource-hub/manifest.json. For each doc it stores: relative path, content hash, title, category, summary, topics, related (list of paths), last-built timestamp.
3. For each new or changed doc (hash differs from manifest): - Extract (one Haiku call): title, a 2 to 3 sentence summary, the category (from the section 3 list), and 3 to 6 key topics. - Relate: match this doc's topics and filename against every other manifest entry; store the top few overlaps as related. - Render: convert the markdown body to HTML (a markdown lib), wrap it in the shared template (top nav, breadcrumb, title, summary, body, related, timestamp). For docs that are already curated .html (like the reference playbook), link or embed them as-is rather than re-styling. - Update the manifest entry.
4. Regenerate index.html and every category page from the manifest (pure templating, so cross-links, "recently updated," and counts always reflect the current set).
5. Write everything to dist/.
6. Deploy dist/ to Cloudflare Pages (section 6).

Cost control: only changed docs hit the LLM. Index and category regeneration is free templating. Log how many docs were reprocessed each run.

manifest.json entry shape:

{
  "path": "Affirmology_LaunchCRM_Playbook_v1.html",
  "hash": "sha256...",
  "title": "Launch Outreach + CRM Playbook",
  "category": "Strategy",
  "summary": "Operating plan for the June 28 warm outreach wave and the Airtable CRM...",
  "topics": ["launch", "crm", "airtable", "beta", "investors"],
  "related": ["Affirmology_LaunchTimeline_v1.md", "Affirmology_FundingStrategy_Memo_v1.md"],
  "last_built": "2026-06-26T18:00:00Z"
}

6. Hosting and access (Cloudflare)

• One-time setup (dashboard is fine, about 10 minutes): create a Cloudflare Pages project named affirmology-resources. Add the custom domain resources.affirmology.ai (DNS is in the same Cloudflare account as affirmology.ai). Create a Cloudflare Access application on resources.affirmology.ai with an allowlist policy for these emails only:
• jeff@jeffparker.love
• solballard@gmail.com (Sol)
• cjacobsandassoc@gmail.com (Colin)
• Token note: the existing CLOUDFLARE_API_TOKEN (held in affirmology-studio/.env) has Zone Settings, Cache Purge, DNS, Workers, and R2, but probably not Pages or Access scopes. Either add Pages + Access permissions to the token, or do the Pages-project and Access-policy creation once in the dashboard, then automate only the deploy with a Pages-scoped token.
• Deploy command (from the Mac): wrangler pages deploy /Users/jeffreyparker/CLAUDE/AFFIRMOLOGY/resource-hub/dist --project-name affirmology-resources

7. The cron job (every 2 hours, on the Mac)

Use launchd (preferred on macOS). Build three things:

1. A wrapper script resource-hub/rebuild.sh that: cd's to the folder, activates the venv, runs the generator, runs the wrangler deploy, and appends to resource-hub/rebuild.log.
2. A launchd plist ~/Library/LaunchAgents/ai.affirmology.resourcehub.plist with StartInterval of 7200 seconds (2 hours), pointing at rebuild.sh, with stdout/stderr to the log.
3. The load command: launchctl load ~/Library/LaunchAgents/ai.affirmology.resourcehub.plist

Note it only runs while the Mac is awake and online, which is acceptable (same constraint as the render heartbeat). Provide the exact plist contents and the rebuild.sh contents in your build.

8. Acceptance criteria (verify before claiming done)

• resources.affirmology.ai is reachable only behind Cloudflare Access. Incognito test: blocked until login; an allowlisted email gets in; a random email does not.
• Home shows the category grid, a prominent Project State entry, and a "recently updated" list.
• A known doc (the reference playbook) appears, rendered, in the Strategy category, with a summary and related links.
• Edit any doc, wait one cron cycle, and its page and timestamp update; unchanged docs are not reprocessed (confirm in the log).
• No em dashes anywhere in the output. No secret files rendered. Locked demo untouched.

9. Reference file (read this first)

/Users/jeffreyparker/CLAUDE/AFFIRMOLOGY/Affirmology_LaunchCRM_Playbook_v1.html

This is the visual and quality bar. Open it, read its <style> block, and extract it to resource-hub/assets/style.css. Match its layout language: emerald/gold/cream palette, rounded cards, left-border callouts (ok/warn/stop variants), header tables, the two-column table of contents, the pill tags. Every generated page should feel like it came from the same hand as this file.

10. Build order

1. Scaffold resource-hub/; extract the stylesheet from the reference file.
2. Generator working on a handful of docs locally; open dist/index.html in a browser to check the feel.
3. Add the manifest, incremental hashing, and the Haiku extraction and relate pass.
4. Create the Cloudflare Pages project, custom domain, and Access policy.
5. Wire wrangler pages deploy; confirm the gate in incognito.
6. Add the launchd cron; confirm a real 2-hour cycle picks up a changed doc.
7. Run the full acceptance pass in section 8.

Keep the first version lean. Hand-curated category mapping for the few docs that matter is fine for v1 if the LLM classify is flaky; the incremental cron and the Access gate are the parts that must be solid.