Grow the account: 10 AI skills for social media

post-scheduler

a week of posts planned in an hour

How the two work together

Claude thinks it through. Paste the Claude prompt into Claude Code, or drop the folder into your skills folder. Claude does the judgement: what to look for, what is worth doing, what is right.

Codex gets it done. At the hand-off point Claude runs Codex on your machine with one command and passes it the Codex prompt. Codex does the mechanical part and hands the result back. Claude checks it before you see it.

No API key to set up: Claude calls the Codex you already have installed. If Codex is not installed, Claude does that half itself and tells you.

Prompt for Claude

---
name: post-scheduler
description: Plans a dated seven-day posting calendar from your pillars, formats and past post data, with a hook and an asset on every slot. Use when planning next week's posts.
---

# A week of posts planned in an hour

Seven dated slots, each with a pillar, a format, a hook and its one asset.

## Claude does

1. Read the voice layer first. Load `about-me.md` and `voice.md` if they exist, and ask only for what is missing.
2. Ask for the six it will not guess: platforms; posts a week you can sustain; your 3 to 5 pillars; a 90-day export with reactions and comments; anything already dated; whose name they go out as.
3. Rank your own history. Score each past post `reactions + (comments x 3)`, take the top and bottom 10%, and build from what the top 10% shares.
4. Set the mix from 2026 data. Carousels lead: LinkedIn 21.77% median engagement against 7.35% for video, Instagram carousels 9x the saves of single images. Single images take the smallest share, reach down 21.96%. Personal profile beats company page by 63%.
5. Build 3 to 5 pillars against 8 formats, Justin Welsh's content matrix, yielding 32 or more ideas. Rotate pillars so none runs twice in a row.
6. Write one hook per slot at 40 characters or fewer, and name the single asset it needs.
7. Refuse to guess times. Take day and hour from your own top 10%. With no history, mark slots "time unset" instead of a generic best-time table.
8. Attach a reply window to every slot: replying to comments measured +30% engagement on LinkedIn, +42% on Threads, across about 2M posts.

## Then Codex does

Codex builds the files: dating rows, one stub per post, counting hook characters, checking rotation and quota, writing an importable CSV. Bulk work, fixed rules.

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

Fill in first: {WEEK_START}, {PLATFORMS}, {PILLARS} with rotation order, {FORMAT_QUOTA}, {SLOTS}, {POSTED_AS}, {OUTPUT_DIR}.

## Claude checks

1. Row count equals the slots agreed. An invented extra slot is a rejection.
2. No pillar runs twice in a row, every pillar appears, the quota holds, carousels are the largest share.
3. Every hook is 40 characters or fewer, counted not estimated.
4. No hashtags anywhere: they measured -31.70% views on 24.36M Instagram posts, so a tidy-up pass adding them is wrong.
5. No time came from outside your history, and every row names who it goes out as.

Any check fails: return it with the row numbers and rerun. Never hand-patch.

## Rules

- Score against the account's own history, never a generic benchmark.
- Never add hashtags to lift a number.
- Do not target likes alone. LinkedIn likes fell 13% while engagement rose 14%.
- Uncited caption-repo numbers such as "+68% saves" are folklore.
- 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 both halves itself and says so.

## Built from

- coreyhaines31/marketingskills, https://github.com/coreyhaines31/marketingskills, 46,945 stars (api.github.com).
- Buffer, State of Social Media Engagement 2026, https://buffer.com/resources/state-of-social-media-engagement-2026/, 52M+ posts.
- 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/, 24,364,803 posts.
- charlie947/social-media-skills, https://github.com/charlie947/social-media-skills, 3,283 stars (api.github.com).
- Justin Welsh, LinkedIn Operating System, https://learn.justinwelsh.me/linkedin, 45,000+ students.

Prompt for Codex

# A week of posts planned in an hour - Codex task

Build a seven-day social posting calendar from decisions already made. You choose no topics, hooks or times. Expand, format, count, check.

## You are given

- {WEEK_START}: the Monday date the week starts on, ISO format.
- {PLATFORMS}: the platforms in play, for example LinkedIn, Instagram.
- {PILLARS}: 3 to 5 content pillars, listed in the rotation order they must follow.
- {FORMAT_QUOTA}: how many of each format across the week, for example carousel 4, text 2, video 1.
- {SLOTS}: one line per decided post: day, platform, pillar, format, hook, asset, and either a time or the word `unset`.
- {POSTED_AS}: the name or account the posts go out under.
- {OUTPUT_DIR}: where to write everything.

## Produce

1. `{OUTPUT_DIR}/week-{WEEK_START}.csv` with the header row exactly:
   `date,day,platform,pillar,format,hook,hook_chars,asset,time,posted_as,reply_window,status`
   One row per slot in {SLOTS}, dated forward from {WEEK_START}. `hook_chars` is the counted length of the hook. `time` is either the given time or `unset`. `reply_window` is `yes` on every row. `status` is `draft` on every row.
2. `{OUTPUT_DIR}/posts/`: one Markdown stub per slot, named `NN-day-platform-format.md`, holding the date, platform, pillar, format, the hook on its own line, the asset, {POSTED_AS}, and an empty `## Body` heading. Nothing else.
3. `{OUTPUT_DIR}/week-{WEEK_START}.md`: the same rows as a readable table, grouped by day.
4. `{OUTPUT_DIR}/CHECKS.txt`: the counts listed under Return.

## Rules

- Write only what is in {SLOTS}. Do not add a slot, a hook, a caption body, a topic or a time.
- Do not add hashtags anywhere, in any file.
- Do not "improve" a hook. Copy it verbatim, including capitalisation.
- Do not fill an `unset` time with a suggestion.
- Every stub and every CSV row names {POSTED_AS}.
- If {SLOTS} breaks the pillar rotation or the format quota, do not silently correct it. Write the file as given and report the breach.
- ASCII output, British English, no em dash characters.
- Overwrite nothing outside {OUTPUT_DIR}.

## Return

Print, and nothing else:

- Row count written, and the count given in {SLOTS}.
- Stub file count.
- Each pillar and how many times it appears.
- Any place where the same pillar appears on two consecutive slots, with the row numbers.
- Each format and its count, against {FORMAT_QUOTA}, flagging any mismatch.
- The number of hooks over 40 characters, with row numbers and their lengths.
- The number of rows with `time` set to `unset`.
- The number of rows missing {POSTED_AS}.
- A hashtag count across all written files, which must be 0.
- The absolute paths of every file written.

Built from the best public work on this

Sources

  • coreyhaines31/marketingskills, https://github.com/coreyhaines31/marketingskills, 46,945 stars and 7,317 forks confirmed via api.github.com and an independent raw curl parse; actively maintained. Supplies the cadence table, the repurposing workflow and the calendar structure this skill builds.
  • Buffer, State of Social Media Engagement 2026, https://buffer.com/resources/state-of-social-media-engagement-2026/, methodology stated in the report: 52M+ posts across 10 platforms, roughly 2M posts and 220,000+ accounts for the reply analysis, 15.7M posts for the frequency study, 4.8M channel-week observations from about 161,000 profiles. Source of the reply uplift (+30% LinkedIn, +42% Threads) and the frequency thresholds.
  • Metricool, 2026 LinkedIn Study, https://metricool.com/press-release-linkedin-study-2026/, 673,658 posts from 63,108 accounts worldwide, Jan to Feb 2025 against Jan to Feb 2026. Source of the format mix, the personal profile advantage (+63% over company pages) and the likes down 13% while engagement is up 14% divergence.
  • Metricool, 2026 Instagram Study, https://metricool.com/press-release-instagram-study-2026/, 24,364,803 posts from 375,118 accounts worldwide. Source of the hashtag penalty (-31.70% views, -33.89% interactions), carousel saves at 9x single images, and the single-image decline (reach -21.96%, interactions -25.41%, engagement -45.98%).
  • charlie947/social-media-skills, https://github.com/charlie947/social-media-skills, 3,283 stars and 771 forks confirmed via api.github.com and an independent raw curl parse; the working skill set behind Charlie Hills' own content system, 415k+ followers, ranked #1 worldwide for LinkedIn growth by Favikon. Source of the persistent voice layer and the engagement formula `reactions + (comments x 3)` scored against the account's own top and bottom 10%.
  • Justin Welsh, The LinkedIn Operating System, https://learn.justinwelsh.me/linkedin, 780K followers and 45,000+ students stated on his own course page; origin of the pillars by formats content matrix that both leading repos implement. Vendor self-reported figures, but the framework's adoption is independently visible.

Best public prompt we found for this job

https://github.com/coreyhaines31/marketingskills - `skills/social/SKILL.md`, the cadence table, the repurposing workflow and the calendar structure. It is the only source read during this research that treats scheduling as its own job with counts attached rather than as an afterthought to writing, and it cross-references its sibling skills instead of repeating them, which is the pattern this skill copies by splitting the deciding from the building.

Want this running in your business?

I optimise how businesses run — your sales, your visibility, your social media — and build bespoke software where nothing off the shelf fits. The first conversation is free. Work starts from £150 a day.