founder-content-agent

A Claude Code workspace that drafts founder-led content – LinkedIn posts, a newsletter, research round-ups – in one specific person’s voice, and refuses to publish any of it. Ten skills, three reference files, one orchestration file. No app, no API, no scheduler: you open the folder in Claude Code and ask for a draft.

It ships as a scaffold. The framework is real and complete; the voice, the audience and the customer vocabulary are yours to fill in, and the system is close to useless until you do.

Why it is built this way

Most “AI writes your LinkedIn” setups fail the same way: the model writes competent category-average prose that no reader can attribute to a person. Three things here fight that.

A voice profile with convictions in it. founder-voice is not a tone slider. It holds the founder’s positions, their emotional register, what they explicitly do not sound like, and a six-question test a draft has to pass. Positions are what let a draft close instead of hedge.

An observation log that outranks the profile. style-observations records every correction the founder actually made, with the reason and a generalised rule. Where the log and the profile disagree, the log wins – it came from a real edit rather than an intention. This is the part that compounds; a system without it makes the same mistake in a new costume every month.