ProPost
Get started
All skills

Company and communication

The AI team

propost-ai-team

When it helps

Sets up, runs and supervises the ProPost AI team (the Crew: Maestro the lead, Scout for research, and specialists Lina, Kai, Iris, Theo, Mila and Remy per network) through the connector: checks its status, sets how far it may go on each channel with the person's yes, runs it for one brand or one specialist, follows the run, answers the "needs you" items and pauses it. Use whenever someone says "use my AI team", "run the team", "what is the team doing", "let the team draft my week", "what's waiting for me", "approve what the team wrote", "stop the team", "pause the bots", "let the team schedule for me", or asks about Maestro, Scout, Lina, Kai, Iris, Theo, Mila or Remy, or wants to talk to one of them ("ask Maestro how the week went", "ask Iris why this was held", "team stand-up").

Skills work with the ProPost connector, available on Pro and Premium. Set up the connector

Helps the person put ProPost's AI team to work and keep control of it: who is on the team, what each channel lets it do, a run for the brands they choose, and a clear list of what is waiting for their decision.

Before you start

  1. get_account: plan (the team is part of Pro and Premium), AI drafts left (every team post uses one monthly draft), time zone, and this connection's autonomy.
  2. get_team_status: enabled, paused, manual_runs_left_today, drafts_left_this_month, each agent with its channels and state, each channel's mode (drafts, schedule or publish), brand_id, specialist and can_reply_to_people, the open needs_you count and the latest_run. If available is false the plan does not include the team.
  3. list_brands (and get_brand_brief where a brand looks thin): the team writes with each channel's brand brief, so a brand with an empty offer or audience produces generic posts. Offer to fix it first (see propost-multi-brand-channels).
  4. list_posts with status: "scheduled": what is already going out, so a run does not stack on a full week.

The team: Maestro leads and plans each run within the allowance, Scout researches trends per brand, and one specialist per network family: Lina (LinkedIn), Kai (X, Threads, Bluesky, Mastodon), Iris (Instagram, Pinterest), Theo (TikTok, YouTube), Mila (Facebook, Google Business), Remy (Telegram, Reddit). Only specialists with a connected channel are on a person's team. Posts they write show made_by in list_posts and get_post.

Workflow

  1. Report status in one short block: on or paused, which agents work on which channels and brands, each channel's mode in plain words (drafts = asks first, schedule = fills the queue and leaves a heads-up the person can stop, publish = posts on its own, Premium), how many items wait, the last run's summary, drafts left.
  2. Team is off (`team_off` or `enabled: false`). Starting it is done by the person in ProPost, Your team. Say so; nothing in the connector turns it on. You can still prepare brands and channel modes for when it is on.
  3. Set modes with a yes. Propose a mode per channel and why (a new client's page: drafts; a steady personal LinkedIn: schedule). Only after the person agrees, set_channel_mode per channel. It can never be above this connection's autonomy (autonomy_limit), and publish needs Premium (autopilot_required). Never raise a mode on your own to make a run "work"; default advice is drafts until they trust the output.
  4. Run. run_team needs autonomy schedule or higher. Before calling, state the scope and the cost: "Lina and Kai for Harbor Cafe, up to N posts, one monthly draft each, X left". Scope with brand_id and/or agent (lina, kai, iris, theo, mila, remy) when the person names one; without them the whole team runs. Each channel follows its own mode. Keep the returned run_id.
  5. Follow the run. get_team_run with the run_id: status, per agent state, step, done / total and post_ids, plus the latest events. Poll a few times at sensible intervals, not in a tight loop; if it is still going, tell the person it is working and give them the run_id. When it finishes, read the summary and get_post the post_ids you will report on.
  6. Handle "needs you". list_needs_you: review items are drafts the team held back (the reason: the channel's mode, a never-say word, a missing image or video, no free slot) and heads_up items are posts it queued that the person may stop. For each, show the post text (get_post with the item's post_id), the channel and brand, the reason, and propose an answer. Then resolve_needs_you with the person's choice:
    queue: put the held draft in the next free slot (schedule level). Run check_post first and fix what it reports.
    keep_draft: leave it as a draft.
    dismiss: close a heads-up, the post stays queued.
    stop: take a queued post back out as a draft. Improve text with update_post before queueing when the person asks; tell them you changed a post the team wrote.
  7. Talk to a teammate. When the person wants to hear from one of them ("ask Maestro what happened today", "Iris, why was this held?", "what would Theo post next?"), call ask_teammate with agent (maestro, scout, lina, kai, iris, theo, mila, remy), their question in their own words and, when it is about one post, its post_id (the exchange is then added to that post's chat in ProPost). Relay the answer as theirs ("Iris: ..."), unedited in substance. The answer comes from the account's real records and changes nothing: when the teammate suggests an action, ask the person, then use the matching tool. For a round of updates, use the team_standup prompt or ask Maestro, then each specialist who has channels.
  8. Pause and resume. pause_team with paused: true stops new work (queued posts stay queued; use unschedule_post for a single one); false resumes. Do this when asked, or offer it when a brief is wrong.
  9. Report: what the team made (by channel and brand), what is queued and when, what still needs a decision, drafts left, runs left today.

Writing guidance

  • Say what is happening in plain words: "Lina is drafting 2 LinkedIn posts for Nine Studio, Kai is on X" beats pasting event codes.
  • Review the team's posts against the brand brief, not against your taste: right language, no never_say words, no invented facts. Flag, do not silently rewrite.
  • For a review list use a table: brand, channel, reason held, first line, proposed action.
  • When the output is generic, the fix is usually the brief (save_brand: offer, audience, topics, always, never_say, faq), not another run.

Guardrails

  • Changing a mode, running the team and queueing a held draft each need the person's yes in this conversation, naming the channel or brand.
  • Never use run_team to get around a drafts-only connection or an empty allowance; a run spends one monthly draft per post. Say the count first.
  • Never publish_now on the team's posts without an explicit yes after showing the text and channel.
  • Do not run twice in a row hoping for better text; one run, review, adjust the brief, then decide.
  • A team in a shared space follows that space's roles: only owners, admins and editors may run it. A contributor or read-only member's connection cannot (see ../CONVENTIONS.md, Team spaces and roles).
  • The team does not reply to comments or messages through the connector; can_reply_to_people only says whether a channel supports it later.

If something fails

  • plan_required: the team is part of Pro and Premium; tell the person.
  • team_off: the person starts the team in ProPost, Your team.
  • paused: ask before resuming with pause_team (paused: false).
  • busy: a run is already going; follow it with get_team_run.
  • daily_limit: no manual runs left today; the daily run still happens.
  • no_channels: connect a channel first (the person does this in ProPost).
  • blocked: the plan needs renewing before the team can write.
  • autonomy_limit: this connection is too low for run_team, resolve_needs_you with queue, or that mode. Leave items as they are and tell the person where to raise it (Settings, Integrations).
  • autopilot_required: publish is Premium; offer schedule.
  • brand_not_found / agent_not_found / run_not_found / not_found: re-read list_brands, get_team_status, list_needs_you.
  • quota_exceeded / weekly_limit / no_slots: the allowance or queue is full; say which and what is left as drafts.
  • Anything else: see ../CONVENTIONS.md.

How to add a skill

  1. Claude (web and desktop): download a skill as a .zip, then add it in Settings, under Capabilities, Skills.
  2. Claude Code: unzip it into ~/.claude/skills/ (or your project’s .claude/skills/).
  3. Other assistants: open the SKILL.md and paste it into the assistant’s custom instructions or project knowledge.

Skills are written in English. Your assistant will still answer and write posts in your language.