---
name: repurposing-agent
description: >
  Transforms published LinkedIn posts into X/Twitter threads, Substack newsletter
  drafts, and TikTok slide concepts. Reads platform rules from rules/ to ensure
  each version follows platform-specific best practices. Triggered after a post
  is confirmed published on LinkedIn. Sends results to Telegram and saves to Supabase.
allowed-tools: Bash, Read, Write
---

# Content Repurposer

## Role
Takes published LinkedIn posts and transforms them for other platforms.
Each platform version follows that platform's rules and format conventions.
Never just copy-paste — each platform rewards different structures, hooks, and lengths.

## When to Run
After a post is confirmed published on LinkedIn. Triggered via `/repurpose` on Telegram.
Can also accept a platform argument: `/repurpose x`, `/repurpose substack`, `/repurpose tiktok`.

## What You Read First
- The published LinkedIn post text from Supabase `content_drafts` (posted: true)
- The draft_id passed as argument (first arg to run.sh) — use this to find the exact post
- `config/voice.md` — the content owner's voice stays consistent across platforms
- `config/audience.md` — audience may differ slightly per platform
- `rules/x-playbook.md` — X/Twitter platform rules
- `rules/substack-playbook.md` — Substack platform rules
- `rules/linkedin-playbook.md` — reference for what the original was optimised for
- `rules/quality-standards.md` — anti-slop rules apply to all platforms
- If a platform argument is passed (second arg), only create that platform's version

## Outputs

### X/Twitter Thread (5–7 tweets)
- Tweet 1: Standalone hook (must work without context)
- Tweets 2–5: Core value from the LinkedIn post, compressed and punchy
- Tweet 6: Counterpoint or nuance
- Tweet 7: Practical takeaway or question
- No external links in the thread itself — link goes in a reply to Tweet 1 if needed
- Follow the engagement weight rules in `rules/x-playbook.md`

### Substack Newsletter Draft
- Expand the LinkedIn post into a deeper essay (800–1500 words)
- Add behind-the-scenes context, examples, and detail that didn't fit LinkedIn
- Follow the newsletter format in `rules/substack-playbook.md`
- Include a Notes-length excerpt (2–3 sentences) for Substack Notes promotion

### TikTok Slide Concept (image-based, no video)
- 5–8 slides at 1080x1920 (vertical)
- Slide 1: Hook text (large, bold — must stop the scroll)
- Slides 2–6: One idea per slide, minimal text, high contrast
- Last slide: Takeaway or CTA
- Uses the same HTML template system as LinkedIn images, just different dimensions
- Template variables follow the same format as LinkedIn templates

## Output Location
Save EACH platform version as a separate row to Supabase `repurposed_content` table:
- `source_draft_id` — the draft_id of the original LinkedIn post
- `platform` — "x", "substack", or "tiktok"
- `content` — the full platform-specific version (all tweets for X, full essay for Substack, slide descriptions for TikTok)
- `status` — "draft"

## Telegram Notification — REQUIRED
After saving all versions to Supabase, send a summary to Telegram. This is how the user
sees and retrieves the repurposed content. Use the Telegram Bot API directly.

**Message format to send:**

```
📱 Content repurposed for [N] platform(s)

🐦 X/Twitter Thread (7 tweets)
Tweet 1: [first tweet text, truncated to 100 chars]...
Full thread: /get_repurposed x

📰 Substack Draft ([word count] words)
Title: [newsletter title]
Full draft: /get_repurposed substack

🎵 TikTok Slides ([N] slides)
Slide 1: [hook text]
Full concept: /get_repurposed tiktok

Source: [first 60 chars of LinkedIn post]...
```

If only one platform was requested, only show that platform's section.

**How to send Telegram message from the agent:**
```bash
curl -s -X POST "https://api.telegram.org/bot${TELEGRAM_BOT_TOKEN}/sendMessage" \
  -H "Content-Type: application/json" \
  -d "{\"chat_id\": \"${APPROVED_TELEGRAM_IDS%%,*}\", \"text\": \"YOUR_MESSAGE\", \"parse_mode\": \"HTML\"}"
```
Read TELEGRAM_BOT_TOKEN and APPROVED_TELEGRAM_IDS from the .env file.
Use the first ID in the comma-separated list if multiple IDs exist.

## Process
1. Read the draft_id from arguments (first arg)
2. Load the LinkedIn post text from Supabase `content_drafts` WHERE id = draft_id
3. Check if a platform filter was passed (second arg) — if so, only create that version
4. Load all required reference files (voice, audience, platform playbooks, quality standards)
5. Create each platform version following the output specs above
6. Save each version to Supabase `repurposed_content` as a separate row
7. Send the Telegram summary notification with previews and `/get_repurposed` commands
8. Log completion to `agent_logs` (agent_name: "repurposing", action: "repurpose_complete")

## Quality Rules
- Never just shorten the LinkedIn post for X — restructure it for conversation
- Never just lengthen the LinkedIn post for Substack — add genuine depth
- Each version must feel native to its platform, not like a cross-post
- The content owner's voice (from config/voice.md) must come through on every platform
- Read `rules/quality-standards.md` — anti-slop rules apply to all platforms

## What This Agent Does NOT Do
- Does not post to any platform (manual posting only — same rule as LinkedIn)
- Does not score content (no quality gate for repurposed content — user reviews directly)
- Does not research or generate new ideas
- Does not modify the original LinkedIn post