JuriCloud Creative Copilot
JuriCloud Creative Copilot
A Codex skill for generating brand-safe legal-tech content
A Codex skill for generating brand-safe legal-tech content
Summary
Summary
JuriCloud Creative Copilot is a Codex skill I built to generate LinkedIn and campaign content for a legal AI brand without losing control of tone, claims, or visual direction. The goal was not to make a generic prompt that could produce posts quickly. The goal was to design a repeatable AI workflow that could create useful content for a high-trust category while staying inside strict brand and legal-safe boundaries.
JuriCloud Creative Copilot is a Codex skill I built to generate LinkedIn and campaign content for a legal AI brand without losing control of tone, claims, or visual direction. The goal was not to make a generic prompt that could produce posts quickly. The goal was to design a repeatable AI workflow that could create useful content for a high-trust category while staying inside strict brand and legal-safe boundaries.
Over one month, this system helped support a more consistent content engine and contributed to roughly 200 new LinkedIn followers through sharper, more repeatable brand communication.
Over one month, this system helped support a more consistent content engine and contributed to roughly 200 new LinkedIn followers through sharper, more repeatable brand communication.
How It Started
How It Started
I started with the simplest version of the problem: I wanted a way to generate LinkedIn posts, campaign ideas, and ad concepts for JuriCloud without rewriting everything from scratch each time.
At first, the workflow was mostly prompt-based. I could ask for a post or concept and get something usable, but the output was inconsistent. Some drafts sounded too generic. Some leaned too much toward startup hype. Some drifted into phrasing that was not right for a legal AI company. The ideas were there, but they were not controlled enough.
I started with the simplest version of the problem: I wanted a way to generate LinkedIn posts, campaign ideas, and ad concepts for JuriCloud without rewriting everything from scratch each time.
At first, the workflow was mostly prompt-based. I could ask for a post or concept and get something usable, but the output was inconsistent. Some drafts sounded too generic. Some leaned too much toward startup hype. Some drifted into phrasing that was not right for a legal AI company. The ideas were there, but they were not controlled enough.
The Problem
The Problem
That was the point where it shifted. I realized this was not really a copywriting problem. It was a workflow design problem.
That was the point where it shifted. I realized this was not really a copywriting problem. It was a workflow design problem.
The workflow needed to solve four problems:
preserve JuriCloud’s legal-safe positioning
generate repeatable content ideas faster
keep the tone premium, restrained, and professional
create a review loop before anything was published
preserve JuriCloud’s legal-safe positioning
generate repeatable content ideas faster
keep the tone premium, restrained, and professional
create a review loop before anything was published
What I Changed Over Time
I broke the system into separate layers so each part of the workflow had a job:
SKILL.mdto define the role, scope, and output formatjuricloud-brand-rules.mdto store positioning, approved language, banned claims, and message bucketsjuricloud-house-style.mdto control visual tone, layout families, typography, and color directionprompt-templates.mdto make generation repeatable across different campaign typesworkflow.mdto define the operating loop from intake to outputqa-checklist.mdto review outputs for brand, tone, and compliance riskformat-rules.mdto keep Meta ad formats consistent and adaptable
The biggest improvement was moving from one-off prompting to a structured skill.
What I Learned
The biggest lesson from building this was that AI content systems need constraint design more than they need raw generation.
If you let the model improvise too freely, it will often sound fine at first glance, but it will slowly drift away from the brand. In a category like legal AI, that drift matters. The difference between “smart” and “safe” is often just one sentence. The difference between “useful” and “off-brand” can be a phrase like “replace,” “guaranteed,” or “autonomous.”
I also learned that repeatability is more valuable than novelty for this kind of work. A legal-tech brand does not need a different creative personality every time it posts. It needs a system that can keep producing aligned content, week after week, without losing trust.
The QA checklist was another important part of the process. It made the workflow more reliable because every output had to be reviewed for:
positioning drift
banned phrases
overclaims
lawyer-replacement implications
generic SaaS tone
visual drift from the premium dark-and-gold system
That review layer is what makes the system feel usable in practice, not just clever in theory.
What I Changed Over Time
The biggest improvement was moving from one-off prompting to a structured skill.
I broke the system into separate layers so each part of the workflow had a job:
SKILL.mdto define the role, scope, and output formatjuricloud-brand-rules.mdto store positioning, approved language, banned claims, and message bucketsjuricloud-house-style.mdto control visual tone, layout families, typography, and color directionprompt-templates.mdto make generation repeatable across different campaign typesworkflow.mdto define the operating loop from intake to outputqa-checklist.mdto review outputs for brand, tone, and compliance riskformat-rules.mdto keep Meta ad formats consistent and adaptable
What I Learned
The biggest lesson from building this was that AI content systems need constraint design more than they need raw generation.
If you let the model improvise too freely, it will often sound fine at first glance, but it will slowly drift away from the brand. In a category like legal AI, that drift matters. The difference between “smart” and “safe” is often just one sentence. The difference between “useful” and “off-brand” can be a phrase like “replace,” “guaranteed,” or “autonomous.”
I also learned that repeatability is more valuable than novelty for this kind of work. A legal-tech brand does not need a different creative personality every time it posts. It needs a system that can keep producing aligned content, week after week, without losing trust.
The QA checklist was another important part of the process. It made the workflow more reliable because every output had to be reviewed for:
positioning drift
banned phrases
overclaims
lawyer-replacement implications
generic SaaS tone
visual drift from the premium dark-and-gold system
That review layer is what makes the system feel usable in practice, not just clever in theory.
What Helped Me Improve It
What helped me improve the skill was treating every new draft like part of a feedback loop.
When a post felt too broad, I narrowed the message bucket.
When the tone felt too promotional, I rewrote the approved language downward.
When the visuals felt too generic, I constrained the layout family and color behavior.
When the output felt too free, I added structure.
The system got better because I kept refining the boundaries around it.
That process also made the work easier to explain to other people. Instead of saying “I use AI to write posts,” I can now say: I built a controlled creative system that turns brand rules into repeatable content. That is a much more honest description of the work.
Outcomes
Outcomes
Over roughly one month, this workflow helped support a more consistent LinkedIn content engine and contributed to about 200 new followers.
Over roughly one month, this workflow helped support a more consistent LinkedIn content engine and contributed to about 200 new followers.
I would not frame the result as pure follower growth alone. The more meaningful outcome was that the content process became repeatable. I could move from idea to post more quickly, while staying inside a brand-safe message frame. That made it easier to publish consistently, and consistency is part of what helped the account grow.
Just as importantly, the posts started to feel more coherent. Instead of sounding like disconnected AI-generated drafts, they started to feel like part of the same point of view.
I would not frame the result as pure follower growth alone. The more meaningful outcome was that the content process became repeatable. I could move from idea to post more quickly, while staying inside a brand-safe message frame. That made it easier to publish consistently, and consistency is part of what helped the account grow.
Just as importantly, the posts started to feel more coherent. Instead of sounding like disconnected AI-generated drafts, they started to feel like part of the same point of view.
What this project shows
What this project shows
I do not treat AI as a way to bypass judgment. I treat it as a way to systemize judgment. In this case, that meant building a content workflow that could handle legal-tech messaging carefully, repeatedly, and with enough structure that it could be trusted.
I do not treat AI as a way to bypass judgment. I treat it as a way to systemize judgment. In this case, that meant building a content workflow that could handle legal-tech messaging carefully, repeatedly, and with enough structure that it could be trusted.
