Each one: named operator, linked source, every step.
A Claude Code skill schedules each Kit broadcast, skips onboarding subscribers and has a dry-run mode.
Reads Search Console, makes one change, logs a prediction, scores itself 7–14 days later.
Session hooks feed a topic queue, sub-skills draft, and a custom publisher ships posts with images.
One verified dataset, one Next.js route, per-row content and a re-verification job.
Free cross-recommendations took her list from ~300 to 1,000. SparkLoop monetizes the same flow.
Enrichment, one personalized email per lead, YAML follow-up rules and a bounce kill switch.
63 days of verified logs: zero llms.txt requests, 75 reads of markdown twins.
One talking-head session becomes 14 shorts, a newsletter and LinkedIn posts, with a QA check at every stage.
AI reads 56 subreddits, drafts replies in the founder's voice and sends them to Slack. The system can't post; a person approves and posts each one.
65 beta readers stuck in a tool with no export: Claude cleaned the list, added them to Kit with a tag and drafted the email. He confirmed the send.
Four Claude Code agents brief, write, edit and analyse; a person approves every post. Reach up about 25x, median post about 1.3x.
A 2%-completion PDF swapped for one markdown agent file cut from an internal Claude Code skill. Six hours from internal tool to live.
Free agent pack: turn one piece of content into a week of posts, and credit creators in your niche. You approve everything.