Content

/operator-cmo:content

Writing is the easy half. The half that decides whether this stack is worth running is what you refuse to write.

The decision guard

1. Read the brief. If the operator did not name one, list the briefs in marketing/briefs/ from the last 14 days and ask which. If there is no brief, stop and run /operator-cmo:brief first. Do not write from a prompt.
2. Read marketing/decisions/positioning.md for the proof you are allowed to use, and marketing/policy/voice.md for how it should sound. If the voice file is missing or still the empty template, say so once and offer templates/voice-operator-preset.md, which is a real filled-in voice with the AI-slop bans already in place. Writing to a blank voice file produces competent generic copy, which is the most expensive kind.
3. Read the brief's bet: and gate: fields. They go in the output header.

Write

Follow the brief's one belief. Everything in the asset either supports it or comes out.

Structure by default: open on the specific problem or a number, put the mechanism in the middle, end on what the reader should do or decide. No throat-clearing, no summary of what you just said.

The claim pass

Before the draft is saved, check every factual statement:

• Traceable to a proof line in positioning.md? Keep it.
• Not traceable? Either cut it or mark it inline as [unverified] and list it at the end of the file under ## Claims needing evidence.
• A number, a customer name, a logo, an award, a benchmark, or a comparison to a named competitor that is not in positioning.md never ships. No exceptions, including when the operator asks in the same message. Tell them it belongs in positioning.md first.

The anti-slop pass

Run this on your own draft before saving, and report what you changed.

• No em dashes. Colon, comma, or a new sentence.
• Banned phrases: delve into, leverage as a verb, seamless, streamline, revolutionise, empower, unlock, game-changer, cutting-edge, synergy, robust solution, In conclusion, It is worth noting, At the end of the day, In the realm of, In today's fast-paced.
• No hollow tricolons. Three-item rhythmic lists that carry one idea. Keep the list if the three items are genuinely different, cut it if it is cadence pretending to be content.
• No headings that are labels. "Data analysis" is a label. "The data said something else" is a heading.
• No hedging. "It is important to consider", "this could potentially", "results may vary". Take a position or delete the sentence.
• Read the first and last sentence aloud. If either could open or close any article on this topic, rewrite it.
• Vary sentence length on purpose. A page of identical rhythm reads as generated even when every sentence is true.

Report the pass as a count: phrases replaced, em dashes removed, claims flagged. A silent pass is not a pass.

Output

Save to marketing/drafts/YYYY-MM-DD-<slug>.md with this header:

---
bet: B2
decision: positioning.md rev YYYY-MM-DD
brief: marketing/briefs/YYYY-MM-DD-<slug>.md
gate: B
claims_flagged: 0
---

The header is the point. An asset that cannot name the decision it came from should not be publishable, and this makes that mechanical rather than aspirational.

Finish

Print: the bet, the one belief, the anti-slop counts, any flagged claims, and the gate tier with what that tier requires before this goes out. Do not publish anything yourself. Publishing is /operator-cmo:ship-gate.