All of these skills are baked into our AI creative strategy tool, Parker! Use code “PARKERBRAIN” at checkout to get a free month and test it out!

Link here: https://heyparker.ai/

Anti-AI-Slop Script Writing Skill

A drop-in skill for Claude Code and Codex that writes ad scripts sounding spoken, not written — and the craft docs and independent review agents that make it work. It’s built to beat the thing every AI script does by default: clean, smooth, em-dash-cadenced copy that reads fine on the page and dies the second someone says it out loud.

This is a standalone slice of the Parker marketing-intelligence brain, extracted so anyone can copy it into their own setup and use it.

What it actually does

The skill enforces one belief: AI cannot write a good script cold. Left to invent from scratch, it produces plausible, tell-ridden ad copy. So the default flow is reference-driven adaptation — find a proven ad, keep its structural skeleton, and rewrite every word in the brand’s own voice. Net-new from scratch is the rare fallback, and even then it’s held to the same voice doctrine.

The AI-tells doctrine — no em-dash cadence, no smooth triplets, no “it’s not just X, it’s Y,” no rhetorical self-questions, no throat-clearing openers, no abstractions where a real named thing belongs, and the mess kept in (the placed “so,” “honestly,” the false start, the repeated word) — lives in creative-strategy-context/spoken-script-voice.md and creative-strategy-context/ai-writing-tells.md. Those two docs are the heart of the whole thing.

The review gates — why this isn’t just a prompt

The important part: the skill doesn’t grade its own paper. Before any script reaches you, it automatically spawns two independent reviewer agents — running in their own context, having not written the draft — and both have to pass. This is the difference between a skill that says “avoid AI tells” and one that actually catches them.

• Grounding gate (skill/reviewers/context-grounding-review.md) — an independent strategist that re-reads the method docs, runs scripts/grounding-check.py, and verifies the script was built from real context: no fabricated facts, numbers and specs trace to a source, the methods were applied not just cited. Returns grounded or bounced.
• Voice gate (skill/reviewers/creative-voice-review.md) — an independent reader that runs the mechanical linter scripts/voice-lint.py against the AI-tells doctrine, then judges every flag by ear (catching what regex can’t) and rewrites in the brand’s register. Returns ships or flagged.

Both run every time, on their own, before you see anything — never offered as a “want me to review it?” The script comes back with both verdicts attached. If either bounces, the skill regenerates and re-runs rather than annotating around it.