content-pillars

---
name: content-pillars
description: Decides the four subjects you post about and turns them into a stocked idea bank. Use when you do not know what to post, when posting feels random, or when planning a month of content.
---

# Decide the four things you post about, and never stare at a blank page again

You end with four named pillars, a percentage split, and 32 titled post ideas.

## Claude does

1. Ask for six inputs, refusing to guess any: who you sell to, what you sell, the three questions clients ask most, three wins you own, three opinions you would defend publicly, your platform. Anything missing, ask once and stop.
2. Write the voice files first, `about-me.md` and `voice.md`, holding those answers. Both leading skill libraries converged on this persistent-context spine; every later skill reads it before drafting.
3. Propose 3 to 5 pillars, four being the target. Each gets a one-line definition, the audience pain it serves, and proof you can speak to it. Cut any you cannot give three real examples for.
4. Set a percentage split totalling 100, weighted toward the pillar that brings enquiries: dominant educational, smaller personal, smallest promotional.
5. Fix the eight formats: story, how-to, list, contrarian take, case study, question, observation, resource. Pillars times formats is the Justin Welsh content matrix. 4 x 8 gives 32 ideas.
6. Decide what Codex cannot: which pillar leads, which stories are safe to tell publicly, which claims need a source.

## Then Codex does

Codex crosses every pillar against every format, writes a working title and one-line angle for each of the 32 cells, tags each with its pillar bucket, and checks no two titles repeat a promise. Combination work, not judgement, so it batches.

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

Fill these in CODEX.md first: {PILLARS} (4 names plus definitions), {SPLIT} (percentages), {AUDIENCE}, {PROOF_BANK} (wins, questions and opinions from step 1), {PLATFORM}.

## Claude checks

1. Count the ideas. Under 32, or any empty cell, is a rerun.
2. Reject any title carrying a statistic, percentage or quote. Popular caption repos circulate precise-sounding numbers such as "+68% saves" with no traceable source. Do not seed folklore.
3. Reject hashtag suggestions. Metricool's 24.36m-post study measured hashtags at -31.70% views and -33.89% interactions.
4. Check every idea maps to {PROOF_BANK}. An idea you cannot evidence is a blank page with a title.
5. Check the percentages total 100 and the promotional slice is smallest.
6. Any failure: rerun, naming the failing cells. Never hand-patch.

## Rules

- 3 to 5 pillars, never more. A fifth is usually two versions of one.
- Score future posts against your own top 10%, not best practice.
- 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 builds the 32-cell matrix itself in the chat and says so.

## Built from

- charlie947/social-media-skills, https://github.com/charlie947/social-media-skills, 3,283 stars via api.github.com. Content-matrix skill behind a 415k-follower system.
- coreyhaines31/marketingskills, https://github.com/coreyhaines31/marketingskills, 46,945 stars via api.github.com. Pillar percentage allocations.
- Justin Welsh, The LinkedIn Operating System, https://learn.justinwelsh.me/linkedin. Origin of the pillars-by-formats matrix. 780k followers, 45,000 students.
- Metricool 2026 Instagram Study, https://metricool.com/press-release-instagram-study-2026/. 24,364,803 posts, 375,118 accounts. Hashtag effect.

# Content pillars idea bank - Codex task

You are expanding an approved set of content pillars into a full idea bank. You have no memory of the conversation that produced them. Work only from what is below.

## You are given

- {PILLARS} - four pillar names, each with a one-line definition and the audience pain it serves.
- {SPLIT} - the percentage of posts each pillar should take, totalling 100.
- {AUDIENCE} - who the content is for, in one line.
- {PROOF_BANK} - the owner's real material: the questions clients ask most, wins and stories they own, opinions they would defend publicly.
- {PLATFORM} - the single platform these posts are for.

## Produce

One file, `idea-bank.md`, in the working folder.

1. A table of the four pillars with their definition and percentage.
2. A matrix section of 32 rows minimum: every pillar crossed with all eight formats (story, how-to, list, contrarian take, case study, question, observation, resource). One row per cell, no cell empty.
3. Each row has exactly four columns: pillar, format, working title (under 12 words), one-line angle naming which item in {PROOF_BANK} it draws on.
4. A short "cannot write yet" list at the end: any cell where {PROOF_BANK} holds nothing to support it, named with what the owner would need to supply.

## Rules

- No statistics, percentages, prices or quotes anywhere in a title or angle. If a fact is needed, write `[needs source]` instead of a number.
- No hashtags anywhere in the output.
- No two titles may promise the same outcome. Rewrite the weaker one.
- British English throughout.
- Every angle must name a specific item from {PROOF_BANK}. If none fits, that cell goes on the "cannot write yet" list rather than being filled with something invented.
- Titles are specific to {AUDIENCE} and {PLATFORM}. Generic titles that would suit any business are a failure.
- Do not add pillars, remove pillars or change the percentages.

## Return

Print to stdout, in this order:

1. The absolute path of `idea-bank.md`.
2. `ROWS: <n>` - the number of matrix rows written.
3. `EMPTY_CELLS: <n>` - cells left unfilled.
4. `CANNOT_WRITE_YET: <n>` and the list of those cells.
5. `SPLIT_TOTAL: <n>` - the percentages added up.
6. `CHECKS: numbers=<none|found>, hashtags=<none|found>, duplicate_titles=<none|found>`.