One source to platform posts, in your voice: Multi-Step Agent Workflow for Claude Code, Codex or ChatGPT

Pulls the claims, numbers and quotes from one article, transcript or set of notes, drafts a native post per platform in your voice, and checks each draft against your samples and the source.

Prompt

Installs to .agents/skills/one-source-to-platform-posts-in-your-voice-multi-step-agent-work/

mkdir -p .agents/skills/one-source-to-platform-posts-in-your-voice-multi-step-agent-work && curl -sL "https://prompts-gpt.com/prompts/one-source-to-platform-posts-in-your-voice-multi-step-agent-workflow-for-claude-code-codex-or-chatgp/raw" -o .agents/skills/one-source-to-platform-posts-in-your-voice-multi-step-agent-work/SKILL.md
SKILL.md
Preview
---
name: one-source-to-platform-posts-in-your-voice-multi-step-agent-work
description: "Pulls the claims, numbers and quotes from one article, transcript or set of notes, drafts a native post per platform in your voice, and checks each draft against your samples and the source."
---

# One source to platform posts, in your voice: Multi-Step Agent Workflow for Claude Code, Codex or ChatGPT

Pulls the claims, numbers and quotes from one article, transcript or set of notes, drafts a native post per platform in your voice, and checks each draft against your samples and the source.

## Prompt

```
---
name: repurpose-to-platform-posts
description: Pulls the claims, numbers and quotes from one article, transcript or set of notes, drafts a native post per platform in your voice, and checks each draft against your samples and the source. Returns the drafts with a two-week posting calendar for approval. Use when asked to run the "One source to platform posts, in your voice" workflow.
---

# One source to platform posts, in your voice

Pulls the claims, numbers and quotes from one article, transcript or set of notes, drafts a native post per platform in your voice, and checks each draft against your samples and the source. Returns the drafts with a two-week posting calendar for approval.

## Inputs

Ask for any of these that the user hasn't given:

- `source` (e.g. "Paste the article, transcript or notes (4,000 characters max per run; paste the strongest sections if it's longer).")
- `voice_samples` (e.g. "Paste 2-5 posts or emails you wrote yourself, separated by ---")
- `platforms` (e.g. "LinkedIn, X thread, Instagram carousel, short video, newsletter")
- `cta` (e.g. "Read the full case study at acme.test/blog/churn")
- `audience` (e.g. "B2B SaaS founders between seed and Series A")
- `trend_angle` (e.g. "yes")
- `start_date` (e.g. "2026-11-23")

## Steps

Run the steps in order. Refer to saved results by name. Don't skip a step that says STOP.

1. **Manual start** — Start: Paste `source` and 2-5 things you wrote in `voice_samples`. List `platforms` (LinkedIn, X thread, Instagram carousel, short video, newsletter), the `cta`, `audience` and `start_date`. Set `trend_angle` to 'no' to skip the news search.
2. **Pull out what's worth posting** — From `source` only, list:
1. Core idea: one sentence.
2. Claims: up to 10, each with a verbatim quote.
3. Numbers: every statistic exactly as written, what it measures, and its stated origin (or 'origin not stated').
4. Quotable lines: 3-5 verbatim sentences that stand alone.
5. Story beats: any anecdote as setup, turn, result.
6. Off-limits: what the source hedges or only implies; posts must not state these as fact.
Add nothing that isn't in the source. If it's under 300 words, say at the top that the pack will be thin. Save the result as `facts`.
3. **Fingerprint the voice** — Describe how the writer of `voice_samples` writes, as exactly 10 rules another writer could follow, each with a short quote as evidence. Cover sentence length and variation, openings (quote 3 real first lines), words they repeat or never use, punctuation and formatting (emoji, line breaks, lists, capitals, exclamation marks), and stance (certainty, humour, how they refer to themselves).
Then list banned phrases that would sound wrong in this voice, starting from: delve, game-changer, unlock, in today's fast-paced world, it's not just X it's Y, let that sink in, here's the thing, buckle up. If the samples total under 200 words, call the fingerprint low-confidence. Save the result as `voice`.
4. **Find a timely angle (optional)** — If `trend_angle` is 'no', reply only 'Skipped: no trend angle requested.'
Otherwise search news and discussions from the last 30 days that connect directly to the core idea in `facts` and matter to `audience`. Return up to 3 angles: the event in one line, source URL and date, and one bridging sentence. If no link is direct, reply 'No timely angle found' rather than force one. Never suggest a news claim without its URL. Save the result as `angle`.
5. **Draft per platform** — Launch these as parallel subagents in one message, then wait for all of them. Write a native post for your platform from `facts`, in the voice in `voice`, for `audience`, ending with this call to action: `cta`. Use only claims and numbers from the facts, with exact figures, and nothing from Off-limits. Use an angle from `angle` only if one was found, with its URL in a note under the draft. If your platform isn't in `platforms`, reply only 'Not requested.' Under the draft add 'Claims used:' listing the claims or numbers you used. Save the result as `drafts`.
   - **LinkedIn post**: 1,200-1,900 characters. The first line (under 140 characters, roughly what shows before 'see more' on mobile) must make someone click. Short paragraphs, at most 3 hashtags at the end, no link in the body: put it in a 'First comment:' line.
   - **X thread**: 5-8 posts, each under 280 characters, numbered 1/ 2/ and so on. Post 1 must make sense on its own. One idea per post. A link only in the last post.
   - **Instagram carousel**: 7-10 slides, 25 words max per slide. Slide 1 is the hook and the last slide is the CTA. Then a caption under 150 words and 3-5 hashtags.
   - **Short video script**: 30-45 seconds, timestamped, the hook line in the first 2 seconds, on-screen text for each beat so it works with the sound off.
   - **Newsletter blurb**: 80-120 words, 3 subject line options under 50 characters each, and one link placement marked [link].
6. **Voice and fact review** — Check each draft in `drafts` against the voice rules and banned phrases in `voice` and against `facts`; leave out 'Not requested' sections. Look for (a) generic or AI-sounding phrasing; (b) drift from the source: changed or rounded numbers, claims not in the facts, more certainty than the source, anything Off-limits; (c) voice mismatches.
Per draft, under its platform heading: the corrected draft in full, a 'Fixed:' line (quote the old wording, name the rule), then its 'Claims used:' line.
End with a 'Reference' section copying word for word the 10 voice rules, banned phrases, claims and numbers, and Off-limits list, for the next check. Save the result as `drafts`.
7. **Check until it passes** — Review loop (max 3 passes): Use the Reference section as the source of truth; ignore 'Fixed:' and 'Claims used:' lines. Pass only if every draft meets all of these:
1. Every number matches the Reference exactly.
2. No claim goes beyond the Reference; nothing Off-limits is stated as fact.
3. No banned phrases, and none of: 'not X, but Y' contrasts, a rhetorical-question opener, one-word closers, a three-item list in every paragraph.
4. At least 8 of the 10 voice rules followed.
5. Limits: LinkedIn 1,900 characters max with a first line under 140; each X post under 280; carousel slides 25 words max; newsletter 80-120 words; video 45s max.
6. The CTA appears once.
Feedback quotes the failing line and names the rule; the reviser keeps the Reference section unchanged. Save the result as `drafts`.
8. **Posting calendar** — Make a 2-week posting calendar from `start_date` for the drafts in `drafts` (ignore Reference). Columns: date, day, platform, post (first line), asset needed (image, slides, video shoot, none), CTA, notes. Space the same idea at least 2 days apart and put the most detailed format first. Times of day are suggestions: note to check your own analytics. Save the result as `calendar`.
9. **Edit and pick** — STOP and ask the user to approve before continuing: Read the reviewed drafts and calendar. Edit your copy and note which drafts you'll post in the approval note. Approve to export; reject to stop.
10. **Save the post pack** — Finish: Return `drafts` and `calendar` as one document.

## Done when

Every step has run, each approval was given, and you've reported what was produced and where it was saved.
```

## Usage Notes

Run it in Claude Code or Codex: save as .claude/skills/repurpose-to-platform-posts/SKILL.md (or .agents/skills/ for Codex) and say "run the One source to platform posts, in your voice workflow"; parallel steps run as subagents. Or paste it into ChatGPT or Claude and answer its questions. It asks for: source, voice_samples, platforms, cta, audience, trend_angle, start_date. Build or run it step by step at prompts-gpt.com/workflows/repurpose-to-platform-posts.

Example output

At a glancePost pack rollout: Nov 23 – Dec 6, 2026Ledgerly churn case study turned into an X thread and newsletter
  1. Newsletter blurbNov 24Send 8:00 with case study link
  2. X thread (7 posts)Nov 27Mid-morning, before/after churn chart, link in 7/
  3. X quote-tweet of 3/Dec 2Resurface the 71% stat, link in reply

Example for: a churn case study from fictional SaaS "Ledgerly," 3 short voice samples, platforms X thread and newsletter, CTA "Read the full case study at acme.test/blog/churn," audience B2B SaaS founders (seed to Series A), trend angle "no," start date 2026-11-23.

Post Pack: "We stopped onboarding everyone the same way"

Status: Approved by user with edits to X post 3. Trend angle skipped (none requested). Passed review on pass 2.

X thread

1/ Ledgerly's monthly churn went from 6.1% to 3.8% in five months.

We didn't add features. We stopped onboarding everyone the same way.

2/ The old flow: one 9-step setup for every account. Solo bookkeepers and 40-person finance teams got the same checklist.

3/ What the data showed: accounts that never connected a bank feed in week one made up 71% of churned customers.

4/ So we cut setup to one question first: "How many people will use this?" The answer picks the path.

5/ Solo users now see 3 steps. Teams see 6, plus an invite prompt on day 2.

6/ Our support lead Priya put it best: "Nobody quits software they've actually started using."

7/ Full breakdown, including what didn't work: acme.test/blog/churn

Claims used: 6.1% to 3.8% in five months; 9-step flow; 71% of churned accounts; 3 vs 6 steps; Priya quote.

Newsletter blurb

Subject options:

  1. Our churn fix wasn't a feature (31)
  2. 6.1% to 3.8% in five months (27)
  3. One question, two onboarding paths (34)

For two years every Ledgerly customer got the same 9-step setup. Then we looked at who left: 71% of churned accounts never connected a bank feed in their first week. Now onboarding starts with one question about team size, and the answer sets the path. Solo users get 3 steps. Teams get 6 and an invite prompt on day 2. Monthly churn dropped from 6.1% to 3.8% over five months. The write-up covers the two experiments that failed first. [link]

Claims used: 9 steps; 71%; 3 vs 6 steps; 6.1% to 3.8%.

Calendar (Nov 23 – Dec 6, 2026)

DateDayPlatformPost (first line)AssetCTANotes
Nov 24TueNewsletterFor two years every Ledgerly customer...noneCase study linkSend 8:00, check your open-rate history
Nov 27FriX threadLedgerly's monthly churn went from 6.1%...Before/after churn chartLink in 7/Post mid-morning; adjust to your analytics
Dec 2WedX (repost)Quote-tweet 3/ with the 71% statnoneLink in reply5 days after thread

Reference

Voice rules (excerpt): short declarative openers; first person plural "we"; no emoji; numbers written as digits; admits failures plainly.
Banned phrases: delve, game-changer, unlock, let that sink in, here's the thing, buckle up, "it's not just X, it's Y."
Off-limits: The source says the new flow "likely" improved NPS, so posts make no NPS claim. No attribution of revenue impact.

Written by AI from this prompt. Your results will vary.

Run it in Claude Code or Codex: save as .claude/skills/repurpose-to-platform-posts/SKILL.md (or .agents/skills/ for Codex) and say "run the One source to platform posts, in your voice workflow"; parallel steps run as subagents. Or paste it into ChatGPT or Claude and answer its questions. It asks for: source, voice_samples, platforms, cta, audience, trend_angle, start_date. Build or run it step by step at prompts-gpt.com/workflows/repurpose-to-platform-posts.

Make it repeatable

Keep using this prompt without pasting it in every time.

Prompt details

By
Prompts-GPT editors
Published
How prompts are written and checked

Did this prompt work?

Keep exploring

More in Multi-Agent Pipelines