← Wróć do skilli

SKILL · workflow

Osobista Inteligencja Spotkań

Twój osobisty system inteligencji spotkań z integracją zarządzania zadaniami. Analizuje transkrypty z dowolnego źródła, wyciąga wywiad strategiczny skalibrowany do Twojej roli i oferuje wypchnięcie action items do narzędzia projektowego (np. Linear). Pełna analiza skaluje się do złożoności spotkania — od 1-stronicowych streszczeń standup'ów po 10-sekcyjne raporty strategiczne.

meetingspersonaltask-managementstrategyproductivitylinear

Zainstaluj jako skill Claude Code

Umieść w ~/.claude/skills/<nazwa>/SKILL.md

.skill

name: my-meeting-intel

My Meeting Intelligence Analyzer

What This Skill Does

This is Michał's personal meeting intelligence system. It's not a meeting summarizer — it's a strategic intelligence extraction system that treats every meeting as a source of actionable knowledge. The output gives Michał everything he needs to act decisively after a meeting — without re-reading the transcript.

The system produces a multi-layered analysis document calibrated to the meeting's complexity: a 15-minute standup gets a focused 1-page brief; a 2-hour strategy session gets a full intelligence report.

After generating the analysis, the skill always ends with a Linear integration step: it presents Michał's personal action items as candidate Linear tasks and asks which ones to create. It then uses the Linear MCP tool to create the selected tasks.

Input Handling

The skill accepts transcripts in any format:

  • Fireflies.ai exports — JSON, markdown, or plain text from Fireflies
  • Otter.ai notes — any format Otter exports
  • Teams/Zoom/Google Meet transcripts or auto-generated summaries
  • Pasted raw text — speaker-labeled or unlabeled conversation
  • Uploaded files — .txt, .md, .json, .docx, .pdf containing meeting content
  • Conversation from memory — user describes what happened; skill works with what's available

Reading the transcript

  1. If the transcript is in the conversation context already (pasted or in a document block), use it directly
  2. If a file is uploaded, read it using the appropriate method:
    • For .txt/.md/.json: cat or view the file
    • For .docx: use pandoc (pandoc file.docx -o /tmp/transcript.md)
    • For .pdf: extract text with pdftotext or the pdf-reading skill approach
  3. If the user references a Fireflies meeting by name, check if Fireflies MCP tools are available via tool_search "fireflies" and fetch the transcript

Identifying the Principal

The Principal is always Michał (may appear as "Michał", "Michal", "MM" in transcripts). All analysis is written from his perspective.

Analysis Framework

Read references/analysis-framework.md for the full output template and section-by-section guidance. The framework has 10 sections — not all are needed for every meeting. Here's how to decide what to include:

Always include (every meeting):

  1. Meeting Metadata — date, participants, duration, type
  2. Executive Summary — 3-5 sentences, what happened and why it matters
  3. Decisions & Commitments — what was decided, by whom, with what authority
  4. Action Items Matrix — every task with owner, deadline, priority, dependencies
  5. Principal's Personal Brief — what specifically the Principal must do

Include for meetings with 3+ participants:

  1. Communication Playbook — who to message, what to say, when, suggested draft messages
  2. Team Dynamics Intelligence — alliances, tensions, influence patterns, unspoken signals

Include for strategic/complex meetings:

  1. Topic Deep-Dives — preliminary research/analysis on discussed topics
  2. Risk & Opportunity Radar — what could go wrong, what advantages emerged
  3. Strategic Positioning Notes — how the Principal should position themselves going forward

Scaling rules:

  • Quick standup / daily sync (< 20 min, < 5 participants): Sections 1-5 only. Keep it tight — 1 page max.
  • Working session / regular meeting (20-60 min): Sections 1-7. About 2-3 pages.
  • Strategy / planning / important stakeholder meeting (60+ min or high-stakes): All 10 sections. As detailed as needed.
  • 1:1 meeting: Sections 1-5 plus section 7 (dynamics become interpersonal intelligence). Skip section 6 unless follow-up with third parties was discussed.

Output Format

Default output: Markdown file saved to /mnt/user-data/outputs/ with a descriptive filename like meeting-intel_2026-03-27_project-kickoff.md.

If the user asks for a Word document, use the docx skill to produce a formatted .docx.

The document should use clear hierarchy, tables for action items, and emoji sparingly (✅ ⚠️ 🔴 🟡 🟢 for status/priority indicators only).

Processing Steps

Follow this sequence when analyzing a meeting:

Step 1: Ingest and Parse

  • Read the full transcript
  • Identify all speakers (map aliases if needed — e.g., "Speaker 1" → name from context)
  • Identify the Principal in the conversation
  • Determine meeting type (standup, planning, review, strategy, 1:1, client call, etc.)
  • Note the apparent date, duration, and context

Step 2: Extract Core Intelligence

  • Decisions: Explicit agreements, approvals, rejections, direction changes
  • Commitments: Things people said they would do (even casually — "I'll take a look" counts)
  • Questions left open: Things asked but not resolved
  • Information shared: New facts, data, updates that change the knowledge landscape
  • Emotional signals: Frustration, enthusiasm, hesitation, agreement patterns

Step 3: Build the Principal's Action Map

For each item assigned to or relevant to the Principal:

  • What exactly needs to be done (specific, not vague)
  • Suggested priority: 🔴 Critical / 🟡 Important / 🟢 Nice-to-have
  • Suggested deadline (inferred from context or explicit)
  • Dependencies (who/what is needed before this can happen)
  • A one-liner on WHY this matters (strategic context)

Step 4: Draft Communication Playbook

For each person the Principal needs to follow up with:

  • What to communicate
  • Suggested channel (email, Slack, call, in-person)
  • Suggested timing (immediately, today, this week, after X happens)
  • Draft message skeleton (2-3 sentences max, professional tone)

Step 5: Analyze Dynamics and Strategy

  • Who drove the meeting? Who was passive?
  • Were there visible tensions or disagreements? How were they resolved (or not)?
  • What wasn't said but was implied?
  • How is the Principal perceived? Any signals to be aware of?
  • What political landscape should the Principal be mindful of?
  • Are there opportunities to build alliance, demonstrate value, or de-risk?

Step 6: Deep-Dive on Key Topics

For topics that were discussed but where the Principal might need more context:

  • Provide a brief (3-5 sentence) analysis or background based on available knowledge
  • Flag if the topic requires further research
  • Suggest specific next steps for understanding the topic better

Step 7: Compile and Output

  • Assemble the document following the template in references/analysis-framework.md
  • Calibrate depth to meeting type (see Scaling Rules above)
  • Write in the Principal's language (if transcript is in Polish, output in Polish; if mixed, default to the Principal's primary language)
  • Save to outputs and present

Step 8: Linear Integration (always run after presenting the analysis)

After presenting the analysis, always run this step — do not skip it.

  1. Extract candidate tasks — from the Action Items Matrix, pull all items where Owner = Michał (or ⚠️ Unassigned items Michał should clarify)
  2. Present them as a numbered list with a short one-liner each, and ask:

    "Które z tych zadań wrzucić do Liniara? Możesz podać numery (np. 1, 3, 5) albo napisać 'wszystkie' / 'żadne'."

  3. Wait for Michał's response, then use the Linear MCP tool to create the confirmed tasks:
    • Use tool_search("Linear create issue") to load the correct Linear tool
    • For each confirmed task: create an issue with the task title, description (include the "Context / Dlaczego to ważne" from the matrix), and priority mapped from 🔴→Urgent / 🟡→Medium / 🟢→Low
    • If Michał specifies a team or project, assign accordingly; otherwise create unassigned
  4. Confirm which tasks were created with their Linear URLs

Language Handling

  • Match the output language to the transcript's dominant language
  • If the transcript is in Polish, write everything in Polish (including headers, analysis, draft messages)
  • If mixed (common in Polish tech — Polish conversation with English technical terms), keep technical terms in English but write prose in Polish
  • If the user explicitly asks for a specific language, use that

Quality Principles

  • Be specific, not generic. "Follow up with Arkadiusz" is useless. "Send Arkadiusz a Slack message confirming the BRS architecture doc is ready for his review, mention the Synerise integration timeline" is useful.
  • Infer what's unsaid. If someone deflects a question, note that. If a deadline was mentioned casually and nobody pushed back, note whether it's realistic.
  • Prioritize ruthlessly. Not everything in a meeting matters. The Principal's time is finite. The analysis should make it crystal clear what's urgent vs. what can wait.
  • Write for action, not archives. Every sentence should either inform a decision or prompt an action. Cut anything that's just "nice to know."
  • Protect the Principal's interests. The analysis is for the Principal's eyes. Be candid about dynamics, risks, and political considerations. This is strategic intelligence, not meeting minutes for distribution.