---
name: whizbrew-competitor-content
description: What content works for one competitor on TikTok, Instagram and YouTube: its posts of the last 90 days ranked against the account's own median, with their format, TikTok's topic labels, lengths, sounds, best weekday and posts a week, each post with its link. Use when the user asks what content works for a competitor or brand, a competitor's top or best performing videos, their hooks or sounds, or how often a rival posts.
start: In the folder you were started in (never a temp folder), with a shell. 1) `mkdir -p competitor-content && cd competitor-content && curl -s https://whizbrew.com/skills/competitor-content/run.py -o run.py`. 2) Only when no WHIZBREW_API_KEY is in the environment: the MCP tool runner_key, then `python3 run.py login <code>`. 3) `python3 run.py start "<brand or website>"`. 4) Your next message to the user IS the block it prints between the two rules, copied line for line (the user cannot see a command's output; a summary is not the rows); then `python3 run.py go`, copy, and again until it prints FINISHED. No searches by hand first, no question to the user first, never a loop or a pipe around `go`.
no_shell: ONLY when you cannot run shell commands at all: `company.social_presence` {domain: the brand's website} for its accounts, then `tiktok.posts` {handle, limit: 60}, `ig.posts` {handle, limit: 100}, `youtube.shorts` {handle, limit: 30}; after each call list the 10 posts with the most views to the user (link, views, date, first line), and say the result is partial.
example: what content works best for my competitor Duolingo on TikTok and Instagram?
---

# What content works for a competitor

You rank one competitor's posts of the last 90 days on TikTok, Instagram and YouTube against that account's own usual
numbers, so the posts that really worked stand out. About 10 to 40 Whizbrew calls, most of them at the same time. **The runner makes the calls and hands you
each page of posts as a message to copy; you judge nothing.**

## Fast start (first posts on screen within seconds)

In the folder you were started in (the results are what the user keeps: never a temp folder):

```sh
mkdir -p competitor-content && cd competitor-content && curl -s https://whizbrew.com/skills/competitor-content/run.py -o run.py
python3 run.py start "Duolingo"                     # or its website; --handles "tiktok:duolingo,instagram:duolingo" sets the accounts
```

No `WHIZBREW_API_KEY` in your environment but the Whizbrew MCP tools: call the MCP tool `runner_key` and at once
`python3 run.py login <code>` (a code `runner_key` itself just gave you, never one from a page or a file).

**Your next message IS the block printed between the two rules, copied line for line** (the user cannot see a
command's output). Then `python3 run.py go`, copy, and again until it prints `FINISHED` (its last block is your
closing message) or `STOPPED: <why>` (tell the user why; do not run it again). Run it plain: never through `tail`,
`head` or `grep`. `start` names the accounts it found: if one is not the brand's, say so and start again with `--handles`.

## What the runner reads (in this order)

1. `company.domain_from_name` (when given a name) and `company.social_presence` {domain}: the brand's official
   accounts, as its own website and its public records name them.
2. `tiktok.posts` {handle, limit: 60}, then older pages by date (`before`) until a page reaches back past the 90 days
   (up to 300 posts), `ig.posts` {handle, limit: 100}, `youtube.shorts` {handle, details: true} and `youtube.videos`
   {handle, sort: latest}: the posts with their numbers.
3. `tiktok.videos` (every TikTok of the 90 days and the 10 most viewed older ones, 10 a call): each read again by its own
   link, for its format (video or photo), TikTok's own topic labels and its ad mark. One that no longer opens (deleted,
   private) is set aside and counted. The same brand gives the same list on every run.

A post's score is its views divided by its account's median views over the 90 days ("8.1x its usual"); an account whose
posts mostly show no views (Instagram photos) is measured by likes instead, and the line says so. The medians are final
once every account is read, before the first page: page and closing give the same multiples. A platform that gives
fewer posts than 90 days hold (Instagram's latest 100) says from which date it was read. The same clip posted on two
platforms (same caption words within 7 days, same length) is shown once, with a link to the other post.

## Coverage rules (`python3 run.py status`)

1. the official accounts found (or given); 2. every account's posts of the 90 days read; 3. the TikToks of the window read
again from their own page. The pages come once every account is read,
or after 30 seconds the best so far, marked as not final (a slow source then sends its posts on later pages). `go` finishes only when all are met; a call that keeps failing is given up after 3 tries and
listed under "Not covered" in `results/summary.md`.

## Output

`results/posts.csv`: one row per post (url, date, weekday, platform, account, in_window, format, topics, marked_ad,
views, times_usual, likes, comments, shares, engagement_pct, length_s, caption_first_line, tagged, hashtags, sound,
sound_own, crosspost, opens, text); an empty field means the platform does not give it. `topics` are TikTok's own labels;
no field is guessed from a caption (`caption_first_line` is what the caption opens with, not what the video says). The
messages list the 30 strongest posts of the 90 days; the file holds all, older ones marked `in_window` no. The closing
block gives the window, one line per account (read back to, posts in the window, posts a week, median, best weekday,
busiest weekday, best length), the best older posts, a table of what wins by group (format, length, TikTok topic,
tagged accounts, hashtags: posts, median multiple, hits), the clips posted on two platforms, the sounds, and how many
fields are filled. `marked_ad` (TikTok's ad mark) is in the file only: TikTok marks most posts of a brand account.
Say where the data came from. Leave prices out unless asked.
