A press release is not a press department
Here is the number that shaped everything about this build.
79% of journalists now use AI in their own workflows. More than half of them oppose pitches written by one. That is Cision's 2026 State of the Media, and the sample is over 1,800 working journalists.
Read it twice. The people you want to reach use the same tools you do. They just don't want to be on the receiving end of yours.
So when I built a PR role for the OPVS marketplace, I built it against the thing everyone else is building. @di-atomic/pr-manager shipped on 23 July at tier beta. Its sharpest feature is the list of things it will not do.
Who this is for
You have something worth telling people about. A funding round. A customer result. A product that finally does the thing.
You also have no press department.
So you write a release. It reads fine. Then the real work starts, and none of it is writing. Where does the release live? Who receives it? Under what embargo terms? Attached to which media kit? With which boilerplate — and when was that boilerplate last touched?
Then a journalist replies with a question nobody has decided the answer to.
That gap is the job. A press release is not a press department. Installing a press skill gives you a well-formed document. It does not give you the person who knows what to do with it.
What you hire
A profession is not a bundle. It links a toolkit of skills by version range and copies none of them. Each one stays independently owned and independently versioned. What the profession adds is the judgment about which to reach for, in what order, and when to stop.
pr-manager links 17 skills and carries 7 workflows. Those workflows are 84 steps and 42 rules-of-engagement lines. Behind them sit 12 checklists and 21 lines of negative scope — the things it is not allowed to do.
It starts at probationary trust. It has to earn its way up.
The toolkit
Eight of the linked skills it calls directly. Nine it reads as methodology, then executes with its own writing. That split matters more than it sounds: calling a skill that has no methods just fails.
The eight it calls. @spideriq/publish-skills is the press backbone — 28 press methods across releases, kits, boilerplates, contacts and embargoes. @spideriq/mail-skills sends the one-to-one journalist email. @opvs-ai/agentboard runs the Newsroom board and holds the human approval gate. @opvs-ai/agentdocs stores the press posture: boilerplate, spokesperson bios, forbidden claims, approval chain. @opvs-ai/agentmemory remembers which angle landed with which outlet. @opvs-ai/opvs-protocol carries the guild traffic. @di-atomic/smartlead covers the volume lane and the deliverability gate. @di-atomic/media-generator renders kit art and headshots.
The nine it reads. @di-atomic/press-release-writer is the format authority. @di-atomic/outbound-engine supplies the cold-agent framework — and gets overruled, which I'll come back to. @di-atomic/copy-engine shapes the prose. @di-atomic/voice-builder holds the spokesperson's register, so a quote sounds like the human it is attributed to. @di-atomic/case-studies supplies a customer story as a pitch angle; it reads one, it never writes one. @di-atomic/changelog-writer is the source of record for a product release. @di-atomic/explainer-illustrations plans the kit visuals. @di-atomic/section-designer lays out the newsroom. @di-atomic/distribution-launcher sequences syndication.
The two @opvs-ai and @spideriq families are not mine. They are linked, not absorbed. That is the point of a profession.
The seven workflows
client-intake runs on the day you equip it, and gates everything else. Approved boilerplate per language. A named spokesperson. Whoever approves their quotes. The forbidden-claims list. Nothing drafts before this exists.
newsroom-page-build stands up the public newsroom after intake. Publishing releases does not build a newsroom. The page is its own deliverable.
press-release-drafting is the core operation. Trigger it on a news event, or let it fire when changelog-writer publishes a product release.
press-kit-maintenance runs quarterly, and also fires the moment anything finds a boilerplate older than 90 days.
journalist-pitching runs only after the release exists. You cannot pitch a thing that isn't there.
embargoed-release-coordination runs when the news warrants a real exclusive.
inbound-press-response fires on an event: a journalist query lands in the mailbox. That is the highest-value inbound a brand gets and the easiest one to answer badly.
What it refuses
This is the part I care about.
No unreviewed machine-written pitch. Drafting with AI is fine. Sending unreviewed optimises directly against what the recipient told a survey they want.
One follow-up. Not a ladder. Two independent datasets covering more than 6,300 emails agree: 300 words or fewer, and exactly one follow-up. This deliberately overrules outbound-engine, which advises two to three touches with a fresh angle each time. That advice is correct — for cold sales. Journalist pitching is not cold sales. The profession takes outbound-engine's structure and throws away its cadence.
No blasting. No identical body to a segment. It works the contact list; it does not source one.
Publish is never a preview. publishPressRelease fires a journalist notification that cannot be recalled. Taking the page down afterwards does not unsend the email.
An embargo is a promise, not a timer. embargoPressRelease mints one preview token per bound contact and hands them over exactly once. No later read returns them. So embargo-and-send is a single atomic step, audited by a timestamp field — mint the tokens without mailing them and they are gone, with journalists already notified.
The media kit is never behind a form. Per-asset downloads, file sizes shown before the click, a download-all option. I checked 30 real newsrooms. A gated kit is the fastest way to look amateur to press.
And on the Newsroom board, Live is not a done column. Publishing is the middle of the job. Coverage is the only done column, with Closed — No Pickup as the honest second ending, so the conversion number stays true.
The guild
Every profession opens an address. This one is pr-manager@profession.md.
Every brand running the role writes the same craft back to it — which angle earned the pickup, which outlet actually answers, which spokesperson framing survived the edit. It compounds there, anonymised, across all of them.
A skill ships knowledge. A profession ships knowledge plus a loop that gets sharper.
What it can't do yet
Coverage discovery has no backend in this toolkit. So v0.1 only learns from coverage a human reports back. The loop is real but half-open until a mentions backend exists.
No linked skill sends a LinkedIn DM, and 62% of journalists are there. It will advise that lane. It will not claim to have worked it.
media-monitoring and crisis-comms are both deferred to v0.2, for the same honest reason: I would rather ship seven workflows that work than nine where two are promises.
Equip it
opvs profession equip @di-atomic/pr-managerIn the first 30 days, run client-intake and let it build the newsroom. Then put one real story through it. Watch where it stops and asks — that is the design working, not the design failing.
You will get a press department that is ready before the news breaks, and a conversion number that means something because nothing was counted as coverage until it ran.