⏺ Codex Power Tools — Field Guide (Tools 1–6)

Based on Riley Brown’s “7 Tools That Make Codex 10x MORE Powerful.” Each section pulls the actual setup and usage
notes from the video, plus the price he quoted. The unifying theme: AI agents are only as good as the input you can
feed them quickly.


  1. Whisper Flow — voice-to-text dictation

What it is: A voice-to-text model that lives on your keyboard. Press a single key, speak, and your words are inserted as text into whatever app is focused.

Setup

  • Download from whisperflow.ai (Mac and Windows apps).
  • Free plan is “pretty generous”; goes paid after a weekly word threshold.

How to use

  • Hold the Fn key while speaking → release → text drops at your cursor.
  • Double-tap Fn for hands-free mode — record while you switch apps, walk around, do research; tap again to stop and
    the text inserts where your cursor was.
  • Works across apps: you don’t have to stay on the page where you started recording. Switch tabs, open docs, then come back to Codex and tap to insert.

Best practices

  • Use it for the long tail of “fire-and-forget” agent prompts — research questions, marketing-strategy asks,
    vibe-coding requests — where typing is the bottleneck.
  • Recover lost recordings: if you tapped without your cursor in a text box, open Whisper Flow’s history and copy/paste from your recent transcripts. Don’t re-record.
  • Riley admits there are likely cheaper alternatives — worth shopping around.

  1. Raycast — clipboard manager (and launcher)

What it is: A Mac launcher app. Riley uses it almost exclusively as a clipboard history manager — every copy is saved.

Setup

  • Download Raycast.
  • Search for “Clipboard History” inside Raycast → right-click (or two-finger click) → Configure Command → assign a hotkey. Riley uses ⌘M.
  • In settings, set clipboard retention to “Last month”.

How to use

  • ⌘M anywhere → opens clipboard history. Paste any prior copy back, including images.
  • Filter to images only when you want to dump a batch of screenshots into a Codex prompt (e.g., “make a slide deck and include these images”).
  • Search past copies: ⌘M, then type x.com to find every X link you’ve copied this month. Same trick works for any
    string.

Best practices

  • Treat Raycast as your clipboard memory for agents — research, copy, copy, copy without losing earlier items, then
    dump them all into one prompt later.
  • Bonus: it’s a faster Spotlight replacement and a live calculator (1*1000).
  • Cost: $0. Riley has used the free plan for three years. The Pro plan is only AI/cloud features he doesn’t need.

  1. CleanShot X — screenshots, markup, screen recording

What it is: A replacement for the macOS screenshot tool, plus annotation and screen recording. The core trick: screenshots stack and stay pinned on the left of your screen instead of disappearing.

Setup

  • Buy at cleanshot.com.
  • One-time $30 for the local-storage version. There’s a recurring cloud plan — Riley skips it.

How to use

  • ⌘⇧4 — screenshot. The image stays pinned on the left until you copy or save it. You can stack many at once.
  • ⌘⇧5 — screen recording. Outputs video or GIF, with a built-in pause button so you can skip “loading/thinking”
    moments before posting to X.
  • Annotation: open any pinned shot → use arrows, shapes, and text to point at exactly what you want changed.

Best practices

  • Visuals beat words for context. When asking Codex to fix a UI, screenshot the part you want changed, mark it up with an arrow + a one-line note (“dots should be bigger”), copy, paste into the prompt.
  • Stack multiple shots in one prompt: “for the third image I uploaded, the rectangle thing looks weird — change it.” This works because screenshots persist.
  • Pair with Whisper Flow + Raycast: record your screen → drag straight into a tweet → dictate the caption with Whisper Flow → done.

  1. Paper.design — AI-native design canvas

What it is: Looks like Figma, but built specifically for AI agents. You ship designs by prompting Codex, which drives Paper through an MCP.

Setup

  • Sign up at paper.design.
  • Install the Paper MCP in Codex.
  • Free plan has a limited tool-call cap. Paid: 20/mo monthly.

How to use

  • In a Codex project, ⌘N for a new chat — it inherits the project’s context (e.g., the notes app you’re already
    building).
  • Prompt: “Create three different options for the main notes page” → then /paper-mcp → Enter.
  • Codex drives Paper in real time. Pin Codex on one side, Paper on the other, and watch designs render.
  • Edit directly: double-click any text to change it; drag elements (less freeform than Figma — it’s still code-based).
  • Reference a specific component: select it → copy its link → paste into the next prompt as the target of your edit.
    (“This part — make three variations using more like design 1’s color/font.”)

Best practices

  • Treat Paper as a design loop, not a design tool. Spawn 3 options → pick a favorite → ask for variations of that one → drill in by component link.
  • Use the built-in image generator (OpenAI’s image model) for hero/landing-page imagery: select an image → “make this glossier, colder, with frost” → generates a variation in place at the size you want (e.g., 4×3).
  • Riley’s framing: expect more tools like this — AI-first apps designed to be driven by agents rather than humans.

  1. Readwise / Reader — your second brain

What it is: A read-it-later / highlights service that becomes a queryable memory for AI agents.

Setup

  • Subscribe at readwise.io. $9.99/mo.
  • Install the Chrome extension to manually save anything you read.
  • Sync your X bookmarks — bookmarking on X auto-saves the post (and assets) to Reader.
  • In Codex, install the Readwise plugin (Plugins → search “readwise”) so Codex can query it directly. Readwise also
    exposes an MCP.

How to use

  • New Codex chat: “Please summarize my last three days on Readwise. Group by topic. Make it a Word doc.”
  • Codex pulls from your saved items and returns a structured summary — Riley’s example surfaced 144 items across
    topics like AI agents, Apple/ambient devices, geopolitics.
  • Iterate the format: “Put it in a sheet with clickable links to the originals.”

Best practices

  • Save everything, even loosely — the agent does the filtering at query time.
  • Use it as a content-research engine: dump your reading from the week, ask the agent to find patterns, build
    outlines, or pull quotes.
  • Riley openly notes the price stings; he’s looking for cheaper alternatives. Worth checking before subscribing if
    budget matters.

  1. Excalidraw — visual diagrams and presentations

What it is: A visual diagramming tool with great keyboard shortcuts. Turns out AI is “really good at using Excalidraw” — so you let Codex generate full slide-style diagrams for you.

Setup

  • excalidraw.com. Free plan does everything Riley showed — except saving multiple boards.
  • Paid: 9.99/mo if you need a board library.
  • In Codex, install Riley’s Excalidraw skill (his linked skill in the video description). Plus his YouTube Researcher skill if you want to feed in video analysis.

How to use

  • In Excalidraw itself: press 2 for rectangles, 5 for lines — shortcuts he leans on.
  • In Codex: /excalidraw-diagram → prompt with a research goal:
    “Do research on the latest advancements in language models, especially DeepSeek. Create a visual presentation.”
  • Expect up to 10 minutes — it researches before drawing.
  • The skill returns a link → right-click → Open in Browser → click “Replace my content” in Excalidraw → the AI’s
    slides land on your canvas.

Best practices

  • Pair it with another research skill (e.g., YouTube Researcher) to feed in source material first, then diagram from
    those findings.
  • Treat the AI output as a shell, not a finished deck. Riley’s flow: jot ideas → AI builds the outline → he edits for polish.
  • Great for content-creator workflows: 12-slide breakdowns of a topic in minutes vs. hours.

Cross-tool plays Riley demonstrates

The tools compound. A few combos worth stealing:

  • Whisper Flow → Codex. Speak prompts while the agent runs; multitask other work because GPT-5.5-high “can take a
    while.”
  • CleanShot → Raycast clipboard → Codex. Mark up multiple screens, hit ⌘M, dump all of them into one
    well-contextualized prompt.
  • CleanShot record → Raycast (open X) → drag video into tweet → Whisper Flow caption. End-to-end demo post in under a minute.
  • Readwise → Excalidraw skill. Pull the week’s saved articles → ask Codex to diagram the themes for a content piece.

Total monthly cost if you go all-in on paid plans: ~30
one-time, Paper 10, Excalidraw 30 one-time.