The selected recommendations are now tracked as system rules so research improves the machine, not just the reading list.
2. Official OpenAI Docs MCP in Codex
Codex
Use official OpenAI docs for model, API, Agents SDK, Responses API, and product guidance before relying on memory.
Ask Codex to consult OpenAI docs when the task touches OpenAI APIs or current model behavior.
3. Context7 MCP in Codex
Codex
Use current library docs for React, Next, Vercel, Prisma, Tailwind, shadcn, and other fast-moving frameworks.
Prefer Context7 over generic web search for coding docs.
4. Playwright MCP in Codex
Codex
Use browser QA and screenshots for important app screens before claiming a UI is finished.
Attach desktop and mobile screenshots to every serious Cortex or SharedSpace UI pass.
10. Multimodal setup scoring
Cortex
Score YouTube, podcasts, TikTok, X, docs, and articles against builder leverage, product upside, research value, and noise.
Promote only Test and strong Watch cards into the build backlog.
11. Cortex Research Queue UI
Cortex
Make captured research visible as scored cards with actions, not hidden markdown files.
Use this page as the daily research cockpit.
19. Claude CLI auth registration in OpenClaw
OpenClaw
Keep Claude Code auth connected and registered so OpenClaw can route Claude work when needed.
Recheck only if a future doctor run reports Claude CLI auth drift.
22. Trusted Homebrew formulae only
Operating System
Trust only the exact formulae needed for our toolchain, never entire taps casually.
Use formula-level trust and review new taps before installing.
23. Stable Node track for product apps
Operating System
Pin product apps to the stable Node track so builds do not drift across shells, CI, and deployments.
Use the app-level Node version file when working inside Cortex.
24. Vercel AI Gateway default
Cortex
New Vercel apps should route model calls through AI Gateway unless there is a clear reason not to.
Use the model routing helper as the default pattern for new app APIs.
25. Model routing rule
Operating System
Fast model for routine work, deep model for synthesis and builds, docs-backed model work for current API decisions.
Keep the rule visible in Cortex and use it in new product specs.
26. TaskFlow for long-running work
OpenClaw
Use TaskFlow when research or builds need durable state, child tasks, waits, or overnight continuation.
Run deep research jobs as TaskFlow-style queues, not loose chat threads.
27. Conservative ClawHub installs
OpenClaw
Install skills deliberately; do not force overwrite local changes unless Ricky explicitly approves the specific update.
Treat ClawHub updates as reviewable proposals when local edits exist.
29. Weekly belief-changed card
Cortex
Every week, the research queue should produce one card: what changed, why, and what build decision follows.
Make this the top recurring Cortex Premium artifact.
30. Applied AI learning loop
Operating System
Teach AI through one workflow, one product bet, one tool test, and one design critique each week.
Convert research into a weekly applied curriculum instead of abstract lessons.