Unternehmen und Kommunikation
Sensitive announcements
propost-sensitive-announcements
Wann er hilft
Drafts careful company communications in ProPost (price changes, outages and incidents, layoffs or restructuring, apologies, policy or terms changes) as drafts only, in plain honest language, with a legal or comms review step, one consistent message across channels, and a follow-up post plan; never publishes without an explicit yes. Use whenever someone says "we need to announce a price increase", "write an apology post", "we had an outage, what do we post", "announce layoffs", "policy change announcement", "crisis communication", "post about the incident", "tell customers we're raising prices", or "how do we say this carefully", even if they don't name ProPost.
Skills funktionieren mit dem ProPost-Connector, verfügbar in Pro und Premium. Connector einrichten
Produces a reviewed, honest message that is identical in substance on every channel, held as drafts until the right people approve, plus a plan for the follow-up. Speed matters less than getting it right and saying it once.
Before you start
get_account:autonomy, time zone, approval setting. For this skill, behave as drafts-only even at a higher level until the person says "schedule it" or "publish it" after seeing the final text.list_channels: where customers actually read the company. Prefer fewer channels with the same message over many with drift. Notemax_lengthandneeds_reconnect.list_postswithstatus: "scheduled": find promotional posts that would look tone-deaf next to this news. Offer tounschedule_postthem (never delete without being asked).list_personas: the company voice, not a personal or witty one.
Workflow
- Get the facts, in writing, from the person. What happened or is changing, since when, who is affected, what the company is doing about it, what customers need to do, the date things change, where to get help, who signs it. For prices: old and new price, effective date, notice period, what stays the same. For incidents: start time, end time or status, user impact, cause only if confirmed, and when the next update comes. Unknown items stay
[to confirm]; you never fill gaps with likely-sounding facts. - Name the audience and the ask. Customers? Employees? Partners? The public? Each might need a different post, but not different facts.
- Write one core message (3-6 sentences) following the structure below. Use
create_draft(free) so the wording is exactly what the person reviewed; usegenerate_draftonly if asked, and then read every line. Put all channels inchannel_ids; useper_channelonly to shorten, never to soften or add claims. State the AI drafts used (usually zero). - Review step before anything leaves drafts. Present the draft with a checklist: Are the facts verified by the owner of each? Has legal or whoever handles regulation/employment/contract terms read it? Has support got a matching FAQ and reply template? Do employees hear this before the public if it concerns them? Is the effective date and notice period compliant? Say plainly that you cannot give legal advice and that review by a qualified person is the person's responsibility.
- Check with
check_postafter approval of the text. - Release only on an explicit yes. After the person confirms the text and the time:
schedule_post(schedule level) with an ISO time and offset.publish_now(publish level) only after showing the final text and the channels and getting a clear, current "yes, publish". A past "go ahead and plan it" is not a yes. Atdraftslevel, tell them the drafts are waiting in ProPost. - Plan the follow-up and write it as drafts (with
[to confirm]markers): an update post at the promised time, a "resolved" or "what we learned" post, replies to common questions, the thank-you. For prices, a reminder before the effective date. For incidents, a status cadence ("next update at 15:00 BST") that is kept. Schedule only after the same approval. - Report: what is drafted, what is held, what promotional posts you suggested pausing, and what the person must verify.
Writing guidance
Structure that works for almost every case:
- What, in the first sentence, plainly.
- What it means for you (the reader), including what they must do.
- Why, honestly and briefly; no excuses, no blaming a third party you can't prove.
- What we're doing about it, with dates.
- Where to get help, a real contact.
- Owning it, if it is the company's fault: "We got this wrong."
Examples:
- Price change. Weak: "We're making some exciting updates to our pricing." Strong: "From 1 March, the Team plan goes from [old price] to [new price] a month. It's our first increase in three years, and it covers [real reason]. If you're on an annual plan, nothing changes until it renews. Questions: [contact]."
- Outage. "Between 10:12 and 10:54 UTC, [service] was unavailable. Some orders placed in that window may not have gone through. If yours is missing, [what to do]. We'll post what caused it by [time]."
- Apology. One sentence of apology, specific about what went wrong, a fix, a remedy only if the company has authorised it. Not "we apologise for any inconvenience".
- Layoffs or restructuring. Employees are told first, and the post follows only when the company's HR and legal owners have approved it. State the facts, thank people without clichés, avoid euphemism like "rightsizing", no names of affected people, no numbers beyond what is already official.
Tone: calm, direct, humane. No emoji, no exclamation marks, no marketing hooks, no hashtags. Short. Consistent: the same numbers, dates and wording on every channel; per_channel only shortens. Never edit the message on one channel and forget the others.
Guardrails
- Never `publish_now` without an explicit yes to the final text.
- No invented causes, numbers, timelines, compensation or promises. A promise the company can't keep is worse than silence.
- No legal or HR conclusions; flag them for the right person.
- No third-party blame, no private data about customers or employees, no screenshots of internal tools.
- Don't delete other posts without being asked; unschedule and tell them.
- Don't argue with critics in drafts; reply posts only with facts.
- Update, don't overwrite: if facts change after posting, write an update post rather than silently editing history.
If something fails
autonomy_limit: leave as drafts and say they are waiting in ProPost.too_long: shorten withupdate_post, keeping facts and the contact.time_in_past: pick a new time at least a minute ahead.channel_needs_reconnect: say the message could not be prepared there; don't let channels diverge silently.network_rejected: read the post'serror; tell the person, don't retry in a loop.- Anything else: see
../CONVENTIONS.md.
So fügst du einen Skill hinzu
- Claude (Web und Desktop): Skill als .zip herunterladen und unter Einstellungen, Funktionen, Skills hinzufügen.
- Claude Code: nach ~/.claude/skills/ entpacken (oder in .claude/skills/ deines Projekts).
- Andere Assistenten: SKILL.md öffnen und in die benutzerdefinierten Anweisungen oder das Projektwissen einfügen.
Die Skills sind auf Englisch. Dein Assistent antwortet und schreibt Posts trotzdem in deiner Sprache.