Content Repurposer

Content Repurposer — Transcript to Multi-Platform Content

Goal

Take a transcript (podcast, talk, video, meeting) and produce three pieces of content in parallel:

1. Tweet thread (5-10 tweets)
2. LinkedIn post (single post, 150-300 words)
3. Newsletter draft (500-800 words)

Each format has its own voice, structure, and constraints. They share the same source material but are NOT copy-pasted versions of each other.

Inputs

• Transcript: A file path ($ARGUMENTS) or pasted text. Any format — raw transcript, SRT, VTT, or plain text.
• Author name (optional): Defaults to "Nick" if not specified.
• Topic focus (optional): If the transcript covers multiple topics, specify which to focus on.

Process

Step 1: Read and Extract

Read the transcript. Identify:

• Core thesis — the single biggest idea (1 sentence)
• 3-5 key insights — supporting points, stats, quotes, or stories worth highlighting
• Best quotes — punchy, shareable lines (exact words from transcript)
• Narrative arc — the logical flow from setup to payoff

Do NOT summarize the transcript. Extract the raw material that makes each format work.

Step 2: Generate All Three Formats in Parallel

Use the Task tool to spawn 3 parallel agents (use model: "sonnet" for each). Each agent gets:

• The core thesis, key insights, best quotes, and narrative arc from Step 1
• The format-specific template from templates.md
• The format-specific example from examples/

Spawn them like this:

Agent 1: Tweet thread — Task tool, subagent_type: "general-purpose", model: "sonnet"
Agent 2: LinkedIn post — Task tool, subagent_type: "general-purpose", model: "sonnet"
Agent 3: Newsletter draft — Task tool, subagent_type: "general-purpose", model: "sonnet"

Each agent writes its output. You collect and present all three.

Step 3: Present Output

Show all three pieces of content with clear headers. Save to data/repurposed_content/ with timestamp:

data/repurposed_content/
├── {date}_tweet_thread.md
├── {date}_linkedin_post.md
└── {date}_newsletter_draft.md

Voice & Tone

• Register: Smart-casual. Like explaining something interesting to a friend who's also sharp.
• No corporate speak: No "leveraging synergies", no "thought leadership", no "in today's fast-paced world".
• No exclamation marks in tweets or LinkedIn. One max in newsletter (if genuinely exciting).
• Contractions are fine: "don't", "isn't", "we're" — write like you talk.
• Specific > generic: "We cut onboarding from 3 weeks to 4 days" beats "We improved onboarding significantly".

Format-Specific Rules

See templates.md for detailed templates and examples/ for worked examples. Summary:

Tweet Thread

• 5-10 tweets, each standalone-readable
• Tweet 1 is the hook — bold claim or surprising stat, no preamble
• Last tweet is a summary + soft CTA
• No hashtags. No "1/" numbering. No "Thread:" label.
• Each tweet under 280 chars

LinkedIn Post

• Single post, 150-300 words
• Hook line (first line visible before "see more") must be arresting
• Short paragraphs (1-2 sentences each)
• End with a question to drive comments
• No hashtags in body. 3-5 hashtags only at the very end, separated by a line break.

Newsletter Draft

• 500-800 words
• Subject line + preview text included
• Starts with a story or scenario, not "In this issue..."
• Sections with headers
• Ends with a single clear takeaway or action item
• Conversational but slightly more polished than tweets

Edge Cases

• Transcript too short (< 500 words): Produce tweet thread + LinkedIn only. Skip newsletter, note why.
• Multiple distinct topics: Ask user which to focus on, or pick the most compelling one.
• Transcript is an interview: Attribute quotes properly. Use "According to [Guest]..." in newsletter.
• No clear thesis: Flag this to the user. Still produce content but note it may need a stronger angle.
• Non-English transcript: Produce content in the same language as the transcript.

What Good Output Looks Like

Read examples/podcast_example.md for a complete worked example showing a transcript excerpt transformed into all three formats. Use it as a quality bar — your output should match that level.

First-Run Setup

Before executing, check if the workspace has a .gitignore file. If it doesn't, assume the user is new to this skill. In that case:

1. Ask the user if this is their first time running this skill
2. If yes, walk them through how it works and what they need to configure/set up (API keys, env vars, dependencies, etc.)
3. Let them know that Nick wishes them the best!