competitor-teardown

---
name: competitor-teardown
description: Turns 3 to 7 public accounts ahead of you into a counted teardown of the formats, hooks and cadence that earn engagement in your niche. Use before planning content or when growth stalls.
---

# See exactly what is working for the accounts ahead of you

A counted teardown of what the accounts beating you do, and what to copy this week.

## Claude does

1. Asks for five inputs, refusing to start without the first two. {ACCOUNTS}: 3 to 7 public handles ahead of you. {PLATFORM}: one per run. {NICHE}: who you serve. {POST_COUNT}: posts per account, default 100. {YOUR_POSTS}: your own handle, optional.
2. Will not guess reach, impressions, saves, follower quality, or why a post worked: those are private on other accounts. Only public post data is used.
3. Fixes the score before collection so accounts compare: engagement = reactions + (comments x 3), the formula in Charlie Hills' post-scorer. On LinkedIn, comments and reposts are logged separately too, because likes fell 13% year on year while overall engagement rose 14% (Metricool 2026), so likes alone mislead.
4. Sets the six questions Codex answers per account, from the reverse-engineering framework in coreyhaines31/marketingskills: format mix; what the top 10% share; the hook in the first 40 characters; posts per week; opening line and CTA style; what the account never does.
5. Writes CODEX.md, then turns the returned teardown into 5 moves, each tied to a counted row.

## Then Codex does

Codex takes the bulk half: opening each public profile, transcribing up to {POST_COUNT} posts per account, scoring each, cutting that account's top and bottom 10%, and counting format mix, hook length, cadence, hashtags and CTA type. Several hundred rows of transcribing and sorting.

```
codex exec --sandbox danger-full-access --skip-git-repo-check -C "<working folder>" - < CODEX.md
```

Fill in first: {ACCOUNTS}, {PLATFORM}, {NICHE}, {POST_COUNT}, {YOUR_POSTS}.

## Claude checks

1. Opens five rows against the live posts. One wrong number rejects the file.
2. Every conclusion points at counted rows. No adjective without a number behind it.
3. Rejects hashtag counts written up as advice to add hashtags: Metricool 2026 (24,364,803 Instagram posts) measured -31.70% views and -33.89% interactions.
4. Rejects folklore such as "carousels get 2.5x more shares" or "+27% completion": no measurement firm. Only Buffer and Metricool numbers may be cited.
5. Top and bottom 10% cut per account, never pooled, or small accounts vanish.
6. Failed check: hand back the failing rows and rerun once. Failed twice, Claude finishes by hand and says so.

## Rules

- Score by the fixed formula, never by likes alone.
- Count things. Never describe a pattern you have not counted.
- Mark any number not traceable to Buffer or Metricool unverified, or drop it.
- Copy the pattern, never the wording.
- Public information only.
- Never invent a fact, a number or a quote.
- Anything sent in someone's name says whose name it is.

## If Codex is not installed

Claude collects and counts itself on a smaller sample, and says so.

## Built from

- coreyhaines31/marketingskills, https://github.com/coreyhaines31/marketingskills, 46,945 stars via api.github.com; reverse-engineering framework.
- charlie947/social-media-skills, https://github.com/charlie947/social-media-skills, 3,283 stars via api.github.com; engagement formula, top 10% method.
- Metricool 2026 LinkedIn Study, https://metricool.com/press-release-linkedin-study-2026/, 673,658 posts.
- Metricool 2026 Instagram Study, https://metricool.com/press-release-instagram-study-2026/, 24.36M posts.

# Competitor teardown - Codex task

## You are given

- {ACCOUNTS}: 3 to 7 public handles or profile URLs.
- {PLATFORM}: one platform, the same for every account.
- {NICHE}: one line on the audience these accounts serve.
- {POST_COUNT}: recent public posts to collect per account, default 100.
- {YOUR_POSTS}: optionally, the owner's own handle or exported posts, for comparison.

## Produce

Write two files in the working folder.

1. `teardown-rows.csv`, one row per post, columns exactly:
`account,post_url,date,format,first_40_chars,word_count,reactions,comments,reposts,hashtag_count,cta_type,engagement_score`

- `format` is one of: text, single_image, carousel, video, poll, link, document.
- `engagement_score` = reactions + (comments x 3).
- `cta_type` is one of: question, comment_keyword, link, follow, none.
- Leave a cell blank if the platform does not show that number. Never estimate it.

2. `teardown.md`, with one section per account, and each section answering the six questions with counts:

- Format mix, as a count and a percentage of the sample.
- What the top 10% by engagement_score share: list the posts, then 3 to 5 shared traits, each with how many of them show it.
- Hook pattern: the first 40 characters of the top 10%, grouped into named patterns with counts.
- Cadence: posts per week over the sample window, plus the busiest day.
- Opening line and CTA style: counts by cta_type across the whole sample and across the top 10%.
- What this account never does: formats or moves absent from all {POST_COUNT} posts.

Then one final section, `Across all accounts`: the traits that appear in the top 10% of three or more accounts, ranked by how many accounts show them. If {YOUR_POSTS} was supplied, add a table of those traits against the owner's own posts, marked present or absent.

## Rules

- Public posts only. Do not log in as anyone, do not touch private or follower-only content, and collect no personal data beyond the public post.
- Cut the top and bottom 10% per account, never across the pooled set.
- Do not rank by likes alone.
- No commentary, advice or predictions. Count and report.
- Cite no external statistic. If a source figure seems needed, leave it out.
- If an account yields fewer than 20 posts, keep it and state the real sample size.
- Never fabricate a row, a number or a URL. A missing post is omitted, not invented.

## Return

Print, in this order:

1. The two file paths.
2. Posts collected per account, and the sample date range.
3. The engagement_score of the top and bottom post per account, with their URLs.
4. Any account that returned fewer than 20 posts, and why.
5. The count of blank cells per column.