kai-repurpose

Kai root note: knowledge/, harness/, and scripts/ paths in this skill live in the Kai install, not the user's project. Resolve them against the first ancestor directory of this SKILL.md that contains a knowledge/ folder (the Kai plugin root, ~/.claude/kai, or the kai-cmo-harness repo). MARKETING.md, memory/, and any output files live in the current project. If a referenced scripts/ command is not available in this install, say so, skip it, and continue with the file-based guidance — never fabricate its output.

Objective

One pillar piece turned into 15–25 channel-native assets, each standing on its own, staged into a posting schedule that keeps them from cannibalizing each other. A derivative that reads as a repost of the pillar has failed — the same insight gets a different hook, a different length, and the platform's own register every time it appears.

Done when

Work type campaign — floor E5/C3/O4 (harness/eco-floors.yaml), composite. Every derivative is a child work item under its own type: social-post (E5/C2/O3), email-lifecycle (E5/C3/O3), blog-post for the expanded LinkedIn article (E5/C3/O3). The campaign is CLOSED only when every child is CLOSED and the campaign-level threshold is met; one unshipped asset keeps it open.

• E5 — each published derivative returns a provider id and its public permalink reads back matching the approved text. This skill drafts; publishing is a separate approved action.
• C3 — every piece passes banned_word_check with zero violations, respects its platform's character and format limits, and a non-author read the set end to end for hook repetition.
• O4 — reach, engagement, and click-through against a threshold declared before the first post, read at the declared window from platform insights.

Constraints

• Read MARKETING.md from the project root first. If it does not exist, build it from the codebase — CLAUDE.md, README.md, PROJECT.md, package.json, landing pages, email/ad/analytics config — using the template from /kai-email-system, and confirm the draft. Do not ask the user what the product is.
• Three things must be known before production: the source content (pasted, linked, or a file path), which platforms get derivatives, and what matters most (social reach, email engagement, SEO, video views).
• Transcript, podcast, webinar, interview, or long-article sources require a quote mining pass first, and harness/references/transcript-video-research-rules.md loads before mining any third-party video, audio, podcast, webinar, or transcript material.
• Every quote keeps its exact source location — file path, URL, episode name, transcript timestamp, or paragraph locator — so any derivative can be traced back.
• Never invent quotes, credentials, or expert authority. Quote mining is source extraction; rewriting happens only in derivative files.
• Draft as a dry run. Publishing or scheduling requires explicit human approval.
• Each derivative stands alone — no "as I wrote in my blog post." Credit the pillar only where platform norms expect it.
• Different hook for every piece, even two pieces built on the same insight.
• Tone adapts to platform: LinkedIn professional, X punchy, TikTok conversational.
• Every piece runs the quality gate: zero banned words, zero AI slop, platform limits respected.
• Stagger the release over 1–2 weeks. Lead with the highest-reach platform, follow with niche channels.

Context

Need	Load
Repurposing mechanics and channel fit	knowledge/playbooks/content-repurposing.md
Mining third-party video, audio, or transcripts	harness/references/transcript-video-research-rules.md
Organic social format contract and gate minimums	harness/skill-contracts/social-post.yaml
Product, ICP, voice, current channels	MARKETING.md (project root)

Extraction map at workspace/repurposed/_extraction-map.md — one row per extract, with type, best channel, and format. Extract six kinds of material: key insights (3–5 standalone ideas), data points (stats, numbers, results), quotable lines (bold claims, memorable phrases), steps and frameworks, stories and examples, and contrarian takes.

Quote bank at workspace/repurposed/_quote-bank.md — per entry: source location; the quote or paraphrase, with direct quotes marked clearly and kept short; use case (hook, proof, objection, story, CTA, email subject, clip candidate); risk note (unsupported claim, private detail, medical/legal/financial claim, needs approval); and attribution requirement (guest name, customer approval, anonymous/internal, source citation).

Standard derivative set — adapt to the platforms chosen, but these are the format contracts:

#	Asset	Platform	Format
1–2	Text post per insight	LinkedIn	1200 chars
3	Carousel (framework)	LinkedIn	8–10 slides outline
4	Thread (full argument)	X	5–7 tweets
5–6	Single tweet per data point	X	280 chars
7	Single tweet (contrarian take)	X	280 chars
8	Carousel (key takeaways)	Instagram	5–7 slides outline
9	Caption (story/example)	Instagram	150–300 words
10	Video script (hook + insight)	TikTok	30–60 seconds
11	Video script (contrarian take)	TikTok	15–30 seconds
12	Newsletter section	Email	100–200 words
13	Standalone value-add email	Email	300–500 words
14	Expanded-angle article	LinkedIn	2,200–3,000 words
15	Shorts script	YouTube	30–60 seconds
16–25	Additional platform-specific variants	Various	Various

Output tree — same paths as v1:

workspace/repurposed/
├── _extraction-map.md
├── _quote-bank.md
├── _source-pillar.md          # copy of original for reference
├── linkedin/                  # post-insight-1.md, post-insight-2.md, carousel-framework.md, article-expanded.md
├── x-twitter/                 # thread-full.md, tweet-stat-1.md, tweet-stat-2.md, tweet-contrarian.md
├── instagram/                 # carousel-takeaways.md, caption-story.md
├── tiktok/                    # script-insight.md, script-contrarian.md
├── email/                     # newsletter-section.md, standalone-email.md
├── youtube/                   # shorts-script.md
└── _quality-report.md

Escalate when

• The pillar does not contain 15 distinct ideas — padding produces derivatives that hurt the brand.
• A quote's source location cannot be established, or its attribution requirement is unresolved.
• Source material contains medical, legal, or financial claims that need substantiation before any derivative repeats them.
• A third-party transcript's usage permission is unclear.
• The user wants the set scheduled or published without approval.