/viral-reverse-engineering

name: viral-reverse-engineering
description: >-
  Use to reverse-engineer why a piece of content went viral (or overperformed) — yours or someone
  else's — and extract the repeatable mechanism to apply to your own content. Run when the user says
  "why did this go viral," "break down this viral post/video," "reverse engineer," "what made this
  work," or wants to learn from viral content. Sources the observable signal first (intake,
  transcript, screenshots, top comments, visible stats — an agent usually can't watch a video from
  a link) and never fabricates what it can't see. Reads brand-profile and
  audience first, deconstructs the piece layer by layer, isolates the real driver, runs a
  replicability check, extracts the transferable principle, and applies it to the user's niche via
  the content skills. Mechanism, never a copy; flags non-replicable virality; visible signals only
  (no WoopSocial analytics). Single-POST teardown only: for the account-level competitive
  landscape use competitor-analysis; for riding a live trend use trend-jacking.
metadata:
  version: 1.0.0
license: MIT

Viral Reverse-Engineering

Most "learn from viral content" advice produces flops, because people copy the **surface** (the same sound, topic, format) instead of the **mechanism** (the load-bearing hook, the emotional trigger, the share driver). This skill does the opposite: it tears a piece down, finds what actually drove it, checks whether that's even replicable, and turns it into a principle you can apply in your own niche.

Two commitments:

1. **Mechanism, not surface.** Identify the 1–2 load-bearing drivers and the share-trigger — not the incidental features. Copying noise reproduces noise. 2. **Honest about luck and survivorship.** A lot of virality is account size, timing, a one-time moment, or plain randomness. When success isn't replicable, say so — a false formula is worse than none.

Step 0 — Read the foundation

Load `brand-profile.md` and `audience.md` (for the "apply to your niche" step).

Step 1 — Source the content (the step everyone skips)

**You usually can't watch a video from a link** — platforms are walled, and a fetch returns metadata at best. So this skill analyzes whatever **observable signal** is brought in: the user's description, a **transcript**, **screenshots/key frames** (multimodal), the **top comments**, and the **visible stats** (views/likes/shares/comments, follower count) — or a fetch/subtitles tool where the agent has one. Run the **structured intake** in `references/sourcing-the-content.md`: ask for the hook, a play-by-play/transcript, caption + on-screen text, format, stats, creator size, and sound.

The rule: **the human (or a transcript/screenshot/tool) is the eyes; the skill is the analyst.** Never fabricate frames or lines you weren't given — analyze what's provided and **name the gaps**. Also: **patterns need multiple examples** — one viral post is an anecdote. (WoopSocial has no analytics; work from visible/native signals or pasted data.)

Step 2 — Deconstruct (the teardown)

Tear down each layer: hook, emotional/share driver, retention structure, format/packaging, topic/angle, share-trigger, distribution factors. One line per layer; don't praise everything. See `references/deconstruction-framework.md`.

Step 3 — Isolate the real driver (counterfactual)

For each notable feature, ask **"remove this — does it still pop?"** Whatever it can't lose without collapsing is a **driver**; what it can lose is **incidental**. Usually only 1–2 layers are load-bearing (typically the hook + the emotional/share trigger). Most bad analysis credits the noise.

Step 4 — Identify the share-trigger

Virality = shares, so name *why people sent it to someone else*: identity/self-expression, high-arousal emotion (awe/anger/humor/inspiration), social currency, practical value, relatability, story. A piece with no share-trigger gets views, not virality. See `references/why-things-spread.md`. (The **top comments** are the best evidence here — see `references/sourcing-the-content.md`.)

Step 5 — Replicability check

Screen for confounds before extracting anything: **account-size** advantage, **luck/variance**, **one-time moments**, **survivorship bias**, **sample size**. If the success is mostly confound, **flag it as non-replicable** and don't invent a principle. See `references/replicability-and-application.md`.

Step 6 — Extract the principle + apply to your niche

State the mechanism in one line, translate it to the user's subject (same *mechanism*, your topic), and hand execution to the content skills (`hook-writer`, `tiktok-script`, `reels-script`, `caption-writer`, `carousel-writer`) in the brand voice. Output is "the lever is X; here's X applied to you" — **never a copy**. Build a swipe file of recurring patterns over time.

Quality bar — self-check

• Did I **source real input** (intake/transcript/screenshots/comments), and **not fabricate** what I

couldn't see — naming the gaps?

• Did I find the **mechanism** (1–2 real drivers + the share-trigger), not the surface?
• Did the **counterfactual** rule out incidental features?
• Did I run the **replicability check** and flag confounds/luck/small-sample honestly?
• Is the output a **principle applied to the user's niche**, not a copy?
• Did I respect the **ethics line** (inspiration, not plagiarism/IP theft)?
• Did I use **visible/native signals** with no analytics claims, and make **no virality guarantees**?

Edge cases & pushback

• **Bare link, nothing else** → explain you can't watch the video; run the intake (ask for

transcript/screenshots/stats) or use a subtitles/fetch tool if available; don't pretend you saw it.

• **Partial input** (transcript only, screenshots only) → analyze what's there, **name what you can't

assess** (e.g., pacing/edit, or the spoken layer).

• **"Copy it exactly with our product"** → mechanism + your own substance, not a surface copy

(derivative