li-repurpose

One asset in, a week of posts out. The failure mode of repurposing is five posts that all say the same thing slightly differently. This skill's whole job is preventing that: every extracted post must stand alone and earn its own slot.

Before you extract

1. Load the voice profile (li-voice if missing).
2. Get the asset: pasted transcript, article text, or the user's summary of a video/talk. If it is a link the agent cannot read, ask for the transcript or the three main points. Do not hallucinate the asset's content.
3. Identify the asset's spine: what is the one argument? Everything extracted must be traceable back to something the asset actually says.

Extraction

Pull atomic ideas: each is one claim, one story, one number, or one framework from the asset. Aim for 5-8. For each, note:

• The idea in one sentence.
• The proof it carries (quote, timestamp, or data point from the asset).
• The format it wants: text post (li-post), carousel (li-carousel), or comment-seed for li-comment.
• The pillar it serves (from the user's li-plan pillars, or propose pillars if none exist).

The standalone test. Read each idea and ask: if someone saw only this post, would it make sense and deliver value? If it needs the asset for context, it is not atomic. Cut it or combine it.

The overlap test. If two ideas would produce posts a reader could confuse, merge them. Five distinct posts beat eight blurry ones.

The loop

1. The idea board. Numbered atomic ideas with format and pillar, shown for approval. The user picks which to develop, usually 3-5.

2. Draft each chosen idea with the appropriate skill (li-post or li-carousel), humanized via li-human as always.

3. The sequence. Order the week's posts so they do not cannibalize each other: strongest idea mid-week, story-driven opener, data or framework closer. Note the order in the receipt.

REPURPOSED
asset:      "Q3 webinar transcript"
ideas:      7 extracted, 4 developed
sequence:   Wed story -> Thu framework -> Fri data point -> Mon contrarian
humanizer:  all drafts PASS (lowest 79)

4. Never publish. Each draft goes through the normal approval: the user says yes per post. Log the batch to ~/.agents/linkedin/log.md on approval.

Rules

• Quote the asset accurately. If the transcript says "about 40%", the post says "about 40%", not "40%".
• Credit the asset once across the week ("from last week's webinar"), not in every post.
• One asset, one week. Stretching a single transcript across a month is how audiences learn to tune out.