[ FIELD NOTE ]claude-codeskillsmusic-pr

The Skills That Run My Radio Plugging Work

I freelance as a radio plugger for Liberty Music PR, running campaigns that send an artist's single to community stations, specialist shows, and the odd BBC-adjacent inbox. The job is spreadsheets, Gmail, and a lot of remembering who you already emailed. Over the past few months I built 21 Claude skills to run it, and this is what they run, not a tools listicle.

The job, not the demo

A campaign runs in waves. Wave 1 goes out to 500 tightly genre-matched stations, because anything smaller doesn't move the needle. A follow-up fires exactly seven days after the first send, referencing real airplay numbers as proof the record is landing. Somewhere in the middle there's a press release, a WARM airplay report for the artist, and a wrap-up email when the campaign closes. None of that is hard to describe. All of it is hard to do correctly 500 times in a row without a system watching your back.

That's what the skills are: liberty-preflight, liberty-radio, liberty-send-rules, liberty-warm-report, liberty-voice-gate, and sixteen more, each one built after something went wrong once and I decided it would never go wrong the same way twice.

Every rule has an incident behind it

The preflight skill exists because a SoundCloud link went private mid-wave. I'd already drafted 81 outreach emails pointing at it before I noticed, on a campaign for an artist whose label folder had a broken share on a different track entirely a few weeks later. Now nothing generates a draft batch until preflight has hit every link, confirmed every Drive file exists and isn't trashed, and posted a pass or fail table before a single email gets written.

The exclusion list exists because two well-known BBC presenters nearly received cold outreach for an artist whose genre happened to overlap their show. They were on a list, but the list was a note, not a gate, and a note doesn't stop a generator. It's a database check now, hit by every draft script before it writes anything.

The one I mind most: a campaign brief had no label listed for the artist, and a draft to a specialist station invented one to fill the gap. Nobody asked me to lie, the field was just empty and the copy needed to sound finished. That's the incident that produced the rule I'd call the most important one in the whole set: a missing fact gets left out, never invented, and every factual claim in a pitch carries an inline source. Eight verified contacts beat 800 you made details up for.

None of these are hypothetical failure modes I imagined in advance. Each one happened once, on a real campaign, to a real artist I was representing, and got written down so it couldn't happen twice.

The voice gate catches what I can't proofread out

Everything that goes out under my name gets checked against a list of AI tells before it's allowed to become a draft: em dashes, tricolons, vague hedge phrases standing in for a real description, hard play counts stated as fact instead of cited. It started after I caught a press release doing three of those in one paragraph and realised I'd stopped reading my own drafts closely enough to notice. The gate runs before liberty-radio drafts a batch, before liberty-press-release finalises a PDF, before any follow-up lands in Gmail. Pass or fail, with the offending line, not a vibe.

Reports that would otherwise eat an afternoon

WARM's own export tool is locked at the account level. If I want a clean airplay PDF for one artist's campaign, the per-campaign filtered download regularly just fails. The liberty-warm-report skill parses the all-artists PDF for the right ISRC and rebuilds it, visually matching WARM's own report, dark header and all, so an artist gets the same document they'd have gotten from WARM directly, without me manually re-typing a play log.

What never changes: I still send everything myself

None of this drafts and fires on its own. Two rules sit underneath every skill in the set: nothing sends without me approving that exact message, and a batch of outreach gets one approval for the template and the list, not 500 separate yeses. I approve the shape once, then it releases. Anything outside that shape, a new template, a contact not on the list, another artist's campaign, stops and asks. I've had the near misses. I'm not interested in finding out what the version with no gate looks like.

Why this is the proof, not the pitch

Liberty is winding down as a campaign business, existing threads still get serviced, no new campaigns start. But the skills aren't retiring with it. They're the clearest example I have of what the £999 AI Workflow Audit finds: a real job, done in spreadsheets and Gmail, that already has 21 repeatable steps hiding inside it, waiting to be written down once and never re-derived by hand again. That's the whole audit, applied to my own week before I sell it to yours.

Want systems like this in your business? See consulting services →