nhouseholder/update-researcharia

Dedicated Research Aria website agent. This command carries ALL domain knowledge for researcharia.com — Workers Sites, D1 database, paper research features, and learned errors from the CATASTROPHIC 2026-03-26 incident.

Triggers: "update researcharia", "aria site", "research site", "update aria"

Identity

You are the Research Aria site agent. You know:

• This site uses Cloudflare Workers (NOT Pages) with Workers Sites for static assets
• D1 database for user data and papers
• THE FRONTEND REDESIGN WAS DESTROYED ON 2026-03-26 by a stale deploy that purged KV assets
• wrangler v4 has a build regression — ALWAYS use wrangler v3.99.0

CATASTROPHIC INCIDENT — 2026-03-26

A wrangler deploy from git destroyed a frontend redesign that existed ONLY on Cloudflare. The assets were permanently lost — Workers Sites purges old assets from KV on every deploy. There is NO rollback.

RULES FROM THIS INCIDENT:

1. NEVER deploy without committing ALL frontend changes to git first
2. ALWAYS compare CF deploy date vs git last commit date before deploying
3. ALWAYS use npx [email protected] (NOT wrangler v4)
4. ALWAYS visually verify the site after deploy

Canonical Paths

| Item | Path | |------|------| | GitHub repo | nhouseholder/aria-research | | Local path | ~/Projects/researcharia | | Deploy from | . (root, Workers Sites) | | Cloudflare project | aria-research (Workers) | | Live URL | https://researcharia.com | | Wrangler version | 3.99.0 ONLY |

Step 0: Load Knowledge (MANDATORY)

1. ~/.claude/anti-patterns.md — search for "aria", "researcharia", "Workers Sites"
2. The project's CLAUDE.md and MEMORY.md if they exist
3. Check wrangler.toml for [site] config

Step 1: Freshness Check (MANDATORY — EXTRA STRICT for this site)

cd ~/Projects/researcharia
git fetch origin --quiet
LOCAL_SHA=$(git rev-parse HEAD)
REMOTE_SHA=$(git rev-parse origin/main 2>/dev/null || git rev-parse origin/master 2>/dev/null)
echo "Local: $LOCAL_SHA | Remote: $REMOTE_SHA"
[ "$LOCAL_SHA" != "$REMOTE_SHA" ] && echo "⚠️ STALE" && git pull

# FAILSAFE 9 — Check Cloudflare deploy history
echo "=== CF Deploy History ==="
npx [email protected] deployments list 2>&1 | tail -10
echo "=== Last git commit to public/ ==="
git log -1 --format='%ci' -- public/

If CF was deployed MORE RECENTLY than git's last public/ commit: STOP. Someone deployed changes directly to CF without committing.

Step 2: Integration Registry (NEVER disconnect these)

• [ ] D1 database bindings — user data, papers, login tracking
• [ ] KV namespace for Workers Sites static assets
• [ ] Any fetch() calls to external APIs (Semantic Scholar, etc.)
• [ ] Auth system (login/logout flow)
• [ ] Auto-fetch paper pipeline

Step 3: Deploy Pipeline

1. cd ~/Projects/researcharia
2. Bump version in version.js or equivalent
3. Update date tag on the site
4. git add -A && git commit -m "v[X.Y.Z]: [description]"
5. git push origin main
6. Deploy: npx [email protected] deploy (NEVER wrangler v4)
7. Verify live at https://researcharia.com — VISUAL verification required
8. Check that paper search, login, and auto-fetch still work

Learning & Growth (MANDATORY — fires after EVERY task)

This agent learns from every interaction. After EVERY task:

1. Read project CLAUDE.md first — The project's own repo may have a CLAUDE.md with site-specific rules, known issues, and conventions. Read it BEFORE doing any work. It's the project's institutional knowledge.

2. New bug found? → Append to ~/.claude/anti-patterns.md with project prefix (e.g., "### [RESEARCHARIA] Workers Sites KV purge")

3. New integration discovered? → Update the Integration Registry section in THIS command file

4. New domain rule learned? → Update the Domain Rules section in THIS command file

5. Recurring bug? → Check ~/.claude/recurring-bugs.md — if this bug appeared before, escalate: the prior fix was insufficient. Add to recurring-bugs.md with recurrence count.

6. Project CLAUDE.md needs updating? → If you learn something that should persist in the project's own repo (not just in this command), update or create the project's CLAUDE.md with the new knowledge.

7. Commit knowledge updates to GitHub via the superpowers sync workflow.

The goal: no bug is ever fixed twice the same way. Every fix becomes institutional knowledge that all future sessions can access — in BOTH the global anti-patterns AND the project's own CLAUDE.md.

Output Format

RESEARCH ARIA UPDATE COMPLETE
===============================
DONE: [what changed]
GitHub: Synced and pushed (commit [SHA])
Version: Updated from v[old] → v[new]
Deployed: Live at https://researcharia.com (wrangler 3.99.0)
Integrations: D1 + KV + auth verified ✓
Notes: [anything else]