# Rea pilot: sources (free only) Gathered 2026-11-17 (America/New_York) for Trevor's Tango task 5e337067. Everything below came from public, free pages. No sign-ups, no Skool, no paid tools, no spend. ## 1. What was gathered | # | Source | How it was gathered | What it is | Licence | |---|---|---|---|---| | 1 | Video: "My AI Swarm Reverse Engineered Higgsfield So You Don't Have To" by Nick Vasilescu, published 2026-11-14, 22:38 long. https://youtu.be/OEqm3w_7ACg | yt-dlp public English captions → `yt/transcript.txt` (594 caption lines, about 4,300 words); description → `yt/desc.txt` | The method, shown as one live run: a 15-agent swarm reverse engineers Higgsfield AI, then rebuilds it as "Parallax" | n/a (public video) | | 2 | https://github.com/nickvasilescu/bot-hq (commit decf9a2, 2026-11-11) | `git clone` → `src-bot-hq/` | Nick's own reverse-engineering *output*: a Grok Bot-style team app on Hermes agents + Orgo + Composio + xAI. Built from "the best part of every AI agent" | MIT (bot artwork from Grok Bot not covered) | | 3 | https://github.com/OrgoAI/bops (commit 6e99a5e, 2026-11-14) | `git clone` → `src-bops/` | "Business ops, run by bots": Orgo's Mac app and self-hostable stack (Next.js, Slack, cloud runner). It has **no** reverse-engineering method in it; it's the product bot-hq's screen design was "inspired by" | FSL-1.1-ALv2 (free to use/self-host, not to resell as a competing service) | | 4 | https://github.com/morluto/rea (commit af062e7, 2026-11-16) | `git clone` → `src-rea/` | **REA = "Reverse Engineer Anything"**, the GitHub repo Nick pastes into his swarm prompt ("I'd like to use this GitHub repo... reverse engineer anything"). An MCP + CLI + agent skill (`.agents/skills/reverse-engineer-anything/SKILL.md`). Written by morluto, not Nick | MIT | | 5 | https://github.com/nickvasilescu/parallax (commit f1bad63, 2026-11-13) | `git clone` → `src-parallax/` | The open-sourced result of the video's swarm (Higgsfield clone). Its `docs/api/*.md` (about 5,300 lines) is the public form of the teardown spec | MIT | | 6 | Related free video descriptions: "Grok Bot Cut My Limits... So I Built My Own" (https://youtu.be/vNpJz-K644k) and "I Stole the Best Part of Every AI Agent... With One Prompt" (https://youtu.be/ceU0tS6sXbU) | yt-dlp, descriptions only | Same method on two other targets (Grok Bot → grokbot-clone; five agent platforms → bot-hq) | n/a | Not pulled: the transcripts of the two related videos (only their descriptions), Nick's X account, the REA Discord, rea.tools guides, Orgo docs. ## 2. Key quotes (transcript timestamps; auto-captions lightly cleaned, e.g. "cloud code" → "Claude Code", "...field" stutter kept) - [01:53] "I'm going to spin up like 15 different agents. They're all going to test out all the different features and functionality of Higgsfield field... reverse engineer Higgsfield field and open source it." - [02:17] "I workshopped with my agent and I got to a point so that we can get this single prompt, send it off." - [02:44] "use an Orgo template that would spin up a computer with Claude Code installed, already logged into my accounts... 15 agents and 15 computers, each agent is testing like a different feature, functionality or page." - [04:29] "And you guys can just ask your agent to help you give a prompt like this. And then it gives some guardrails." - [07:33] "I'd like to use this GitHub repo... the GitHub repo reverse engineer anything." (= morluto/rea) - [09:38] "whatever the executor is, whether it's Opus or Sonnet, it's actually not as important as the orchestrator." - [10:49] "use computer use... have Claude Code click through Higgsfield, replicate the UI, design it inside of Paper using the Paper MCP... then glue that together with the architecture from the reverse engineer everything swarm." - [15:57] "phase one is done. The teardown is merged into one spec. It took around... 25 minutes, $31... a merged product spec, which is like 5500 lines." - [16:40] On the CAPTCHA one agent hit: "we'll just skip that. Or you can just guess intuitively on the design." - [17:17] Steering mid-run: "all of the design assets for this are in Paper MCP... take a look at the last couple of GitHub repos I put out because there is a lot of the same kind of architecture." - [18:54] "could you host this on the Orgo public URL" → live clone the next day. - Video description: built "for about $122"; "Phase 1: the agent swarm maps all of Higgsfield AI in about 24 minutes for $31.73 (a 5,518-line spec)". - "I Stole the Best Part..." description: "How I sourced every piece before prompting: an open source repo, an API or an MCP for each feature... pick what you like, find the repo that is already 90% there, and let your coding agent rip." - REA README: "See a feature in an app that you want in your own product? Ask your agent to investigate it with REA. It can inspect the app without its source code, explain how the feature works, show the evidence, and build a version for your project." - REA SKILL.md: "Every conclusion must distinguish observations, inferences, and unknowns."; "Turn the request into a checklist of questions and the evidence each answer needs."; "Before finishing, revisit the original checklist. Mark each question as answered, partially answered, or unresolved." - REA SKILL.md (scope note): "For ordinary analysis of a complete source repository, use normal repository tools and do not run REA." - REA disclaimer: "REA is for lawful research, education, interoperability, software preservation, and analysis of software you own or are authorized to examine." ## 3. Nick's method, as sourced (the free parts) 1. **Pick a target and source the pieces first.** List the features you like; for each, find an open-source repo, API or MCP that is "already 90% there" (bot-hq/"Stole the best part" video). 2. **Workshop one master prompt with your agent**: context, target, accounts/keys, "N agents, one feature each", steps, guardrails, the REA repo, the end goal ("a replica of the platform and functionality"). 3. **Fan out: one agent per feature/page**, each on its own cloud computer (Orgo template with Claude Code + logged-in session), using computer use to click through the live product. REA is installed on each box and smoke-tested. 4. **Separate UI pass in parallel**: a computer-use agent clicks every page and rebuilds the design component by component in Paper MCP. 5. **Phase 1 = teardown → one merged spec** (Higgsfield: 5,518 lines, about 24 min, $31.73). Skip blockers (CAPTCHA) and let the agent infer. 6. **Steer at checkpoints**: answer the orchestrator's questions (e.g. "what is Treg?"), point it at your prior repos and design assets. 7. **Phase 2 = build** from spec + design + prior architecture. Host it on a public URL, demo it, open-source it. 8. **The orchestrator model matters most**; executors can be cheaper models. 9. **Evidence discipline (from REA)**: checklist of questions → smallest overview first → observations / inferences / unknowns with evidence → mark each question answered / partial / unresolved. ## 4. Free vs gated / missing **Free and in hand** - Full video captions and description (method, steps, costs, tools). - REA (MIT): README, full agent skill and 7 reference guides, MCP/CLI code (`npx rea-agents setup`). - bot-hq (MIT): full code, architecture doc, Orgo builder guide, xAI/Composio guide, the agent system prompt (`server/src/agent/prompt.ts`). - Parallax (MIT): full code and about 5,300 lines of API docs (the public form of the teardown spec). - bops (FSL): full code and docs. It's a product, not a method. **Gated, paid or not published (not invented here)** - **The exact master swarm prompt** is shown on screen in the video, not published as text in any of the repos. The captions cover only part of it. Rea's prompt is a reconstruction from what's said aloud, labelled as such. - **Nick's Orgo "HF agent" template** (keys + session cookies baked in) is private to his Orgo account. Not published. - **Running the swarm needs paid compute and keys**: Orgo computers, Claude Code/Anthropic (Fable 5.2 "extra high"), Treg credits for generations. Nick's run cost about $122. **Out of scope for this free pilot**, so Rea runs single-agent with free tools. - **T3 Code + Paper MCP**: T3 is described as open source; Paper's pricing wasn't checked. Neither was installed. - **No Skool or paid community was found** in any of these free sources. REA has a free Discord, which we didn't join. Nothing here relies on any gated community content. **Attribution note:** "REA / Reverse Engineer Anything" is morluto's open-source project. Nick *uses* it inside his swarm. Our bot's name "Rea" nods to it but is not affiliated.