Rebuild a product your own way, clean-room style
By Tubemk Studio · Sold by Tubemk
Takes an existing product apart, writes a clean spec, then builds your own version the clean-room way.
Free
Works with the AI agents you already use
ClaudeChatGPT
By Tubemk Studio · Sold by Tubemk
Takes an existing product apart, writes a clean spec, then builds your own version the clean-room way.
Free
Works with the AI agents you already use
Tubemk MCP lets your AI find and fetch skills. Running a local skill still needs the install step below.
Connect Tubemk to Claude
RecommendedFree, read-only, no sign-in. Then ask Claude: "Use the Tubemk skill clean-room-teardown"
Not working? See the setup guide

git clone https://github.com/wanghie/tubemarket-skills tubemarket-skills cd tubemarket-skills python3 -m venv .venv && source .venv/bin/activate pip install -r requirements.txt pip install mediapipe opencv-python ./setup
--- name: clean-room-teardown description: Take apart an existing product to understand how it works inside, write a clean specification, then build your own version in a fresh session that reads only the spec — the two-team clean-room method the tech industry uses to learn from competitors without infringing. Runs inside Claude or ChatGPT, no API key. Triggers on "teardown", "take this product apart", "reverse engineer", "build something like X", "how does product X work". --- # clean-room-teardown Learn from competitors by taking them apart, not by copying. Two teams, one wall: team A reads the original product and writes a spec, team B reads only the spec and builds. Team B has never seen the original, so what it builds is yours. This clean-room method has real case history in tech: a small company reverse-engineered the IBM PC BIOS with exactly this two-team model, was challenged, and prevailed because the build team had never seen the original code. ## When to use After market research has picked the slice to enter. Market research answers "which slice"; a teardown answers "how does the original work so I can build mine". In the wrong order you take apart a product before knowing whether it is worth building. ## Step 0 — Is it worth building? Answer before spending hours Three questions, written down: 1. Why this product, why now, and who exactly is YOUR user? 2. What is your wedge: copy the core plus improve exactly where? 3. Are you building for yourself, shipping publicly, or beyond? The further you go, the more strictly the clean-room rules below must hold from day one. If the answer is "not worth it yet", stop here. That is a good outcome, not a failure. ## Step 1 — Gather sources, public sources only Pull everything into an `input/` folder: demo videos, homepage, pricing page, docs, screenshots, the repo if it is open source. For videos, give the agent the transcript plus frames captured at timestamps. **Public material only.** No cracking, no data from behind someone else's login, no buying accounts to extract code. ## Step 2 — Decompose the core, do not describe the shell From `input/`, have the agent split it into analysis files: ``` features.md what the product can do, feature by feature ui-states.md screens, states, user flows api-endpoints.md which routes the product uses to talk to its server tech-stack.md what it runs on, what it builds itself, what it rents pricing.md price per plan, what is billed on ``` **Clean-room rule, non-negotiable:** in every analysis file describe in words; do NOT paste more than 3 lines of code, and never copy any passage longer than one sentence. Verbatim quotes are only for short factual strings (button labels, error messages). ## Step 3 — Write a clean spec From the analysis files write `SPEC.md`: for each feature, what the user gets, what triggers it, inputs and outputs, states, what happens on error, and **what it does NOT do**. Test: open a NEW agent session, paste the SPEC, ask it to build one feature. If it asks back about behaviour (rather than only about technical choices), the spec is not done. ## Step 4 — Name it yourself and find the wedge - **Give your product a name clearly different from the original.** No `X-clone`, `X-port`, `my-X`. Folder names live forever in git history; a folder called `stripe-clone` is a confession in any future audit. - **First-principles audit:** ask "if I did not know the original existed, what is the minimum spec that solves my user's pain?". Compare with SPEC: what is in SPEC but not in the answer, consider dropping; what is in the answer but missing from SPEC is your wedge. Without this step you get a 5% better copy, not a 10x different product. ## Step 5 — Build behind the wall Open a **new folder and a new agent session**. Copy in exactly two files: `SPEC.md` and the v0.1 task list. The first message of that session: ``` Read only SPEC.md in this folder. Do not read anything outside it. Do not ask me about behaviour, only about technical choices. Build part 1. ``` From now on, the old session writes no code and the new session never sees the original. If the spec is missing something, go back to the old session, fix the spec, copy the new version across. ## Step 6 — Compare Run the same user flow on the original and on yours and compare by eye: is the behaviour equivalent, and where it deliberately differs (the wedge), say so. Pass → ship; fail → back to step 5. ## Remember one line Tear down to understand why they win, then do it differently in the gap. A conclusion of "the same but cheaper" means you are not done — go back to the market matrix and find the slice nobody holds. ## LESSONS LOG Read this whole section BEFORE writing the first word of output. Each row is a mistake someone already pointed out; repeating it is worse than making it the first time. An empty table means this skill has never been corrected. **An empty table is honest — never invent sample rows.** - **Spot feedback:** "that's wrong", "not like that", "don't…", "drop…", "next time…", "from now on…", "I already told you", "still…", "how many times", or the user hand-edits the output you just gave and sends it back. - **What to log:** fix the output and sweep every place with the same problem — ALWAYS. Logging is CONDITIONAL: only when the mistake is RULE-WORTHY, i.e. it would come back with a different input. A ONE-OFF (wrong name, wrong number, wrong link, changed preference) gets fixed, not logged. When logging, add EXACTLY ONE row to the table below, in this very file, then say `Logged in clean-room-teardown LESSONS LOG: WRONG … → RIGHT …`. The **Source** column is required and takes one of three values: `user` (the user actually said it) · `review` (a review pass caught it) · `measured` (a tool run showed it). Never present the agent's own conclusion as if the user said it — a made-up source is worse than no row. The WRONG column describes observable behaviour. - **Cap: 10 rows** — not an arbitrary number. The parent rule is `SKILL.md < 300 lines`; this block takes 28 frame lines, plus one blank separator and 10 rows makes 39, so the skill body must stop under 261 lines. Log full and a new row arrives → promote the most repeated row to a named hard rule in the skill body and delete it from the log; never delete a row just because it is old. If the two rules clash, `< 300 lines` WINS. Check with `wc -l SKILL.md` (< 300) and `grep -c '^| 20' SKILL.md` (<= 10). | Date | Source | WRONG | RIGHT | |---|---|---|---|
Free