--- name: lead-meeting-prep description: | Prepare one of Sascha's recurring meeting rounds: compare against the last Sally summary of the same round, pull the round's Basecamp state from the cockpit, and write a German agenda with "Seit letztem Mal", decisions, blockers and talking points. Use when the user wants to prepare a recurring meeting or build its agenda. Triggers: "Meeting vorbereiten", "Weekly vorbereiten", "Agenda für …". compatibility: opencode, claude-code --- # lead-meeting-prep Prepares one recurring round (Weekly GF, Regeltermin …) in the `lead-*` family. One run = one round → one German prep in the chat. A 1:1 with a team member is out of scope: point the user to `lead-radar person ` and stop. Follow-up after the meeting belongs to `lead-meeting-followup`. ## Sources of truth Read before acting; these hold the facts, this skill holds only behaviour: - `/home/sascha.koenig/p/PM/.okf/processes/meeting-runden.md`: round profiles (participants, Bezug, filing location, Sally search term, Meins lists) - `/home/sascha.koenig/p/PM/.okf/systems/basecamp.md`: project, list and folder IDs - `/home/sascha.koenig/p/PM/.okf/processes/it-basecamp-regeln.md`: approval protocol for any Basecamp write An ID that fails: resolve by name, carry on, and mention the mismatch in the final report. The run is read-only. Load the `basecamp` skill before any Basecamp CLI call; check flags with `basecamp --agent --help`. **Comments** (Basecamp comments, Zammad internal notes): load `sascha-internal-communication-style` and write the German text in that style. End every comment with one empty line and the AI marker in italics, exactly: ```md *✨unterstützt durch KI✨* ``` Show the comment in the proposal exactly as it will be posted, marker included. **Mentions**: name every colleague the comment addresses or affects (owner, new owner, whoever is asked something) as a real mention, so they get a notification. A plain name notifies no one. Do not mention Sascha: Basecamp and Zammad writes run under his account. A comment with no addressee gets no mention. Show the mention in the proposal as `@Vorname Nachname`. - Basecamp: `[@Vorname Nachname](person:)`, ID from `.okf/organization/`, fallback `basecamp people pingable --jq '.data[] | select(.name == "…")'`. Not a bare `@Name`: the CLI resolves it fuzzily and can hit the wrong person. Only people with access to the project are pingable; for anyone else, say so in the proposal. - Zammad (internal note, `text/html`): `Vorname Nachname`, uid via `GET /api/v1/users/search?query=`. A typed `@Name` is plain text there. Zammad subscribes the mentioned user to the ticket and notifies them. Mentions count only when the note is created, and only for agents with access to the ticket's group. ## Steps ### 1. Identify the round Match the user's wording (round name, participant, or an appointment from Sally/calendar) against the profiles in `meeting-runden.md`. On a unique match, name the round and go on; otherwise ask, listing the profiles. Collect every field of that profile marked `offen` that this run needs (Sally search term, participants, Turnus) and ask about all of them in one question. Keep the answers for this run and for step 5. **Done when** you hold the round, its Bezug projects (Basecamp IDs), its Meins lists, its filing location and a Sally search term (or the user's explicit "keiner"). ### 2. Last time: Sally Use the Sally MCP tools (`search_appointments`, `get_recordings`, `get_summary`, `get_transcription`; read each tool's schema for its arguments). Search with the round's search term, take the most recent past appointment that has a recording, and fetch its summary. Pull the transcription only where the summary is too thin to tell what was promised or decided. Extract every promise and decision from that meeting as one line each: what, who, by when. When Sally finds nothing (no term, no hit, not authenticated), say so in the prep and continue without "Seit letztem Mal". **Done when** you hold the date and recording ID of the last meeting and its full promise/decision list, or a stated reason why there is none. ### 3. Current state: Basecamp Run from `/home/sascha.koenig/p/PM`: ```bash python3 tools/it-cockpit/cockpit.py --json --scope runde --projects # Bezug projects of the round python3 tools/it-cockpit/cockpit.py --json --scope meins # 🔒 Meins by list → the round's Meins lists ``` `runde` returns per project `offen` (count), `ueberfaellig`, `diese_woche`, `still_30d`, `wartend_nachfassen`, each item with title, list, assignees, due, `still_tage`, description and URL. For a round whose Bezug is the whole portfolio (Weekly GF), use `--scope portfolio` instead of `runde`. Report anything under `errors` or `drift` in the prep footer. The cockpit takes 1–2 minutes; run both calls in parallel if the harness allows. Then trace each promise from step 2 to its current state: find the matching to-do (by title, owner, or via `basecamp` search/show for detail), and note done / in progress / overdue / not found. **Done when** every promise from step 2 has a state, and every item in the round's Meins lists and in `ueberfaellig` is classified (step 4 categories). ### 4. Write the prep Classify each topic as `Entscheidung nötig`, `Blockiert`, `Überfällig`, `Update reicht` or `Nicht besprechen`. Order the agenda: blockers, decisions, short updates, backlog/strategy last. Airtime is limited: put a topic on the agenda only when it needs the other participants; the rest goes under "Updates" or "Nicht diese Woche". Surface uncomfortable topics plainly; mark anything uncertain as `[unklar: …]`. Write in Sascha's voice: direct, pragmatic, plain internal German (see `sascha-internal-communication-style` for wording). Fill this template in the chat; leave out a section only when it is empty, and say so in one line: ```md # – Vorbereitung YYYY-MM-DD ## Kurzfazit [2–4 Sätze: Lage, Fokus, größte Risiken] ## Seit letztem Mal () - [Zusage/Entscheidung] — [wer] → **[erledigt / läuft / überfällig / nicht auffindbar]**, [Stand in einem Satz, Link] ## Agenda-Vorschlag 1. [Thema] — Ziel: Entscheidung / Klärung / Update 2. ... ## Entscheidungen nötig - [Thema] - Kontext: - Optionen: - Empfehlung: - Konkrete Frage im Meeting: ## Blocker / Eskalationen - [Thema] — blockiert durch [Person/Sache], gewünschte Entscheidung oder Hilfe ## Überfällige / kritische To-dos - [Todo](URL) — Empfehlung: erledigen / verschieben / delegieren / streichen ## Updates, die kurz reichen - [Thema] — 1–2 Sätze Status ## Nicht diese Woche besprechen - [Thema] — Grund ## Gesprächsnotizen / Formulierungen - „[präzise Formulierung für schwierige Themen]“ ``` **Done when** every promise from step 2 appears in "Seit letztem Mal", every overdue item appears in the prep, and each agenda item names its goal. ### 5. Close out End with a short footer: sources used (Sally recording ID, cockpit scopes), ID mismatches, drift/errors. Then offer, as one numbered list, only what applies: 1. **Profile fields:** the `offen` values the user answered in step 1, as exact replacement lines for `meeting-runden.md`. After an explicit yes, edit the profile (only those lines) and validate the bundle with the `validate` skill. 2. **Basecamp prep document:** on request only. Location from the round profile, default 🔒 Meins → folder `Meetings`; title `YYYY-MM-DD – Vorbereitung` (meeting date); `--no-subscribe`; pass the Markdown from a temp file in the scratchpad; read it back and report the URL. 3. **To-do changes** the prep suggests (owner, due date, move, park, complete): propose them per the approval protocol in `it-basecamp-regeln.md` and execute only approved items, each with a short reason comment (style and AI marker as above). **Done when** the prep is delivered and every offered write is either executed after an explicit yes or left untouched. ## Guardrails - Every "Seit letztem Mal" line traces back to the Sally summary/transcript; every state traces to a Basecamp item or reads "nicht auffindbar". - Basecamp, OKF and Sally stay unchanged unless the user explicitly approves a write from step 5.