Communications OS

Three layers: your raw approved sources, a bilingual wiki Claude writes and keeps current, and the rules that govern how your function drafts, checks, and escalates. You read it. Claude maintains it.

• Set up the Claude Code desktop app and stand up your workspace in the first session, guided click by click
• A raw sources layer that holds approved announcements, releases, and legal-approved wording, read only, as your source of truth
• A bilingual messaging house and approved-position register, with every claim traced to the file that approved it
• Twelve operating rules (your CLAUDE.md) that keep the function traceable and keep Italian and English meaning-equivalent

No terminal, ever. You work in the desktop app and describe what you want in plain language.

Each executive gets a profile built only from what they have actually said and approved, so drafts start in the right voice.

• A voice profile for each executive you write for, built from their real approved communications, with corporate position kept separate from personal style
• A consistency check that flags whether a line is an approved position or a new one you have not cleared
• A bilingual equivalence check that compares meaning, tone, and certainty across Italian and English, not just fluency
• Every trait cited to its source, so you approve evidence, not impressions

Skills run on demand against your brain. The checks you run every day, captured once.

The brain starts reading where the function lives, and the day gets a rhythm.

• Scoped connectors for a press email label, one approved-documents folder, and your calendar, read only, each with its own governance card
• A daily rassegna stampa that turns coverage into a two-minute leadership brief and flags what needs a response today
• A memory gate that decides, with you, what the day teaches the brain and what stays temporary
• A confidentiality audit you run before you connect anything, so market-sensitive and employee-sensitive material stays off every connector

You connect where the work happens and keep the sensitive material out. Read only by default, and nothing is ever sent.

Define what good looks like once, with the failure modes and escalation points written down, then let the work run to it.

• An internal-announcement skill that drafts in the company voice, calibrated to the news, in both languages
• An executive-ghostwriting skill that writes in any loaded executive voice, and never invents a view
• A press-response skill that drafts a holding option and a fuller on-the-record option, every claim sourced
• A change-communication skill that maps a change to each audience, with employee-sensitive parts flagged for HR

Each skill runs the two checks and honors the stop conditions. You approve every output, and nothing is ever sent.

The preparation that makes the first thirty minutes calm instead of improvised.

• A crisis layer of your most plausible scenarios, with the stakeholders and questions each one brings
• Pre-approved holding language, in Italian and English, that says only what is confirmed and states what is not
• An escalation map: who approves what, in what order, with materiality always going to Investor Relations
• A crisis-drafting skill you drill under pressure, so a credible first draft exists in minutes

The OS drafts and checks. Legale and the AD own the release, and the materiality call is never yours alone.

One announcement, taken end to end, then the loop that makes every future one start sharper.

• One full cycle from source brief to approval pack to published record, both languages, every claim traced
• A decision record and a publication record you can reconstruct a year later, in minutes
• Post-publication monitoring that compares what the audience understood with what you intended
• A retrospective that proposes what the brain, the skills, the voices, and the rules should permanently learn, and you decide each one

The announcement is done when the function has learned from how it landed.