meeting-retro

name	meeting-retro
description	Retro a recorded meeting (transcript source = Notion page or local transcript file) to surface communication patterns, decision quality, follow-up clarity, and reusable lessons. Saves a single retro per meeting and supports a `patterns` cross-meeting synthesis. Invoke as /meeting-retro (latest), /meeting-retro <date or search query>, or /meeting-retro patterns.
category	execution
runtimes	["claude"]
pii_safe	true

meeting-retro

Turn each recorded meeting into a personal-leadership learning loop. Every retro captures what worked, what didn't, communication patterns, skeptic handling, decision quality, and follow-up clarity. Retros accumulate; the patterns subcommand surfaces recurring themes across all prior retros — that's the actual feedback loop.

Required setup

• TRANSCRIPT_SOURCE — notion or file (which input mode to use)
• RETRO_OUTPUT_DIR — absolute path where individual retro markdown files are written (e.g., <LOCAL_PATH>/retros/)
• RETRO_INDEX_FILE — absolute path to a markdown index that gets a one-liner per retro (e.g., <LOCAL_PATH>/retros/INDEX.md)

Optional providers

• Notion as transcript source — set NOTION_DATABASE_ID pointing at your meeting-notes data source. The skill uses mcporter over SSH to call notion-search and notion-fetch include_transcript=true. Set MCPORTER_HOST_ALIAS=<HOST_ALIAS> if mcporter is on another machine.
• Local transcript files — set TRANSCRIPT_FILE_GLOB (e.g., <LOCAL_PATH>/transcripts/*.md) if you record with Otter / Granola / a local recorder. The skill picks the most-recently-modified file matching the glob (or matches by date / substring).
• Obsidian mirror — set OBSIDIAN_VAULT_PATH to also write each retro into your vault (recommended if you use Obsidian as canonical memory).
• Semantic memory index — set MEMORY_INDEX_PROVIDER=haivemind (or another provider with a store_memory MCP tool) to index each retro for cross-meeting search. Without it, the patterns subcommand still works by filesystem-scanning RETRO_OUTPUT_DIR.
• Review dashboard — set RETRO_DASHBOARD_PAGE_ID if you want each retro also pushed as a sub-page under a Notion "Meeting Retros — Dashboard" parent for end-of-day scanning. Skip if you only want filesystem retros.

When to invoke

• /meeting-retro — retro the most recent meeting from TRANSCRIPT_SOURCE
• /meeting-retro <YYYY-MM-DD> — retro a meeting from a specific date
• /meeting-retro <substring> — search the transcript source for a title matching the substring
• /meeting-retro patterns — read all prior retros and surface cross-meeting patterns
• Auto-invoke on user phrases: "retro this meeting", "what should I learn from that meeting", "help me improve as a manager after that call"

Architecture

• Transcript source: Notion pages OR local files, controlled by TRANSCRIPT_SOURCE.
• Retro storage:
  • Filesystem (canonical): <RETRO_OUTPUT_DIR>/meeting_retro_<YYYY-MM-DD>_<slug>.md
  • Obsidian mirror (if OBSIDIAN_VAULT_PATH set): <OBSIDIAN_VAULT_PATH>/meeting-retros/meeting_retro_<YYYY-MM-DD>_<slug>.md
  • Semantic index (if MEMORY_INDEX_PROVIDER set): one entry per retro with tags meeting-retro, executive, leadership, <topic>
  • Index file: one-liner appended to RETRO_INDEX_FILE

Flow

1. Resolve the target meeting

No args / /meeting-retro:

• notion source: ssh <MCPORTER_HOST_ALIAS> 'mcporter call notion.notion-search "query=meeting" "query_type=internal"'. Pick the page with timestamp closest to today AND the longest body. List the top 5 with timestamps if ambiguous.
• file source: pick the most-recently-modified file in TRANSCRIPT_FILE_GLOB.

Date arg /meeting-retro 2026-05-14:

• notion: search that date string. file: glob files matching the date.

Substring arg /meeting-retro <text>:

• Search the source for matching titles. Pick the best match.

patterns arg:

• Skip to step 6 — synthesize from existing retros, no new fetch.

2. Fetch the transcript

Notion:

ssh <MCPORTER_HOST_ALIAS> 'mcporter call notion.notion-fetch "id=<page-id>" "include_transcript=true"'

Save to /tmp/meeting-transcript.txt. Strip Notion footnote URLs and angle-bracket markup:

import re
clean = re.sub(r'\[\^https://www\.notion\.so/[^\]]+\]', '', text)
clean = re.sub(r'<[^>]+>', '\n', clean)

File: Read the file directly. Same cleanup if the format wraps tokens in angle brackets.

3. Extract meeting structure

Parse:

• Title + date + attendees (from page properties or filename + first lines)
• Auto-generated summary (if present)
• Action items (look for ### Action Items or similar)
• Decisions (lines containing "agreed", "decided", "consensus", "will be")
• Conflict points ("skeptical", "but", "I disagree", "I don't", "pushed back")
• Speaking patterns (who pushed back, who endorsed, who was quiet)

4. Generate the retro

Each dimension gets a short observation grounded in transcript quotes:

1. Communication quality — clarity, analogies, over/under-explaining
2. Skeptic handling — pre-empt, redirect, get defensive?
3. Stage-gating discipline — on-roadmap vs concept; unintentional commitments?
4. Audience read — match audience level; know the decision-maker?
5. Decision quality — clear decisions; anything left dangling?
6. Follow-up clarity — owned + dated action items, or vague "team to figure out"?
7. Time management — drift; high-value sections rushed or skipped?
8. What worked — 1-3 quoted moments that landed well
9. What to do differently next time — 1-3 concrete patterns

Be honest. This is for improvement, not flattery. If something went poorly, say so plainly with evidence.

5. Save the retro

Slug: 2-4 word kebab-case from the meeting topic (e.g., demo-prep, eng-weekly-feedback).

Filename: meeting_retro_<YYYY-MM-DD>_<slug>.md.

Body:

---
name: meeting-retro-<YYYY-MM-DD>-<slug>
description: One-line summary + top lesson
type: feedback
---

# Meeting Retro: <Title> — <YYYY-MM-DD>

**Attendees:** ...
**Source:** <notion URL or file path>

## What worked
- ...

## What to do differently
- ...

## Executive-leadership observations
1. Communication: ...
2. Skeptic handling: ...
3. Stage-gating: ...
4. Audience read: ...
5. Decision quality: ...
6. Follow-up clarity: ...
7. Time management: ...

## Quoted moments
- "<exact quote>" — <speaker>, <context>

## Action items I own
- [ ] ...

## Related retros
- [[meeting-retro-<prior-date>-<slug>]]

## Top lesson for next similar meeting
<ONE-LINE TAKEAWAY>

Write paths:

1. <RETRO_OUTPUT_DIR>/meeting_retro_<YYYY-MM-DD>_<slug>.md (canonical)
2. If OBSIDIAN_VAULT_PATH: also write to <OBSIDIAN_VAULT_PATH>/meeting-retros/...
3. If MEMORY_INDEX_PROVIDER: store via mcp__<provider>__store_memory with category=global, tags=[meeting-retro, executive, leadership, ]
4. Append one-liner to RETRO_INDEX_FILE: - [Meeting Retro <date> — <slug>](<path>) — <top lesson>
5. If RETRO_DASHBOARD_PAGE_ID: create sub-page via notion-create-pages with the full retro markdown as body. Title: <YYYY-MM-DD> — <Meeting title> — retro.

6. Report back

Compact summary:

# Retro: <Meeting Title> — <date>

**Top 3 lessons:**
1. <strongest pattern>
2. <second>
3. <third>

**What worked:** <one quoted moment>

**What to fix next time:** <one specific action>

Full retro: <path>

End with: "Want cross-meeting patterns? Run /meeting-retro patterns."

patterns subcommand

1. List all meeting_retro_*.md in RETRO_OUTPUT_DIR (and OBSIDIAN_VAULT_PATH/meeting-retros/ if set).
2. Pull the "Executive-leadership observations" + "Top lesson" lines from each.
3. Cluster by dimension.
4. Surface repeats: same pattern in 3+ retros → real growth edge.
5. Output:
   • Strengths (consistently good across N retros)
   • Growth edges (repeats)
   • One-off issues (low signal)
   • Recommended next focus — pick ONE growth edge with a way to track progress
6. Save as meeting_retro_patterns_<YYYY-MM-DD>.md in RETRO_OUTPUT_DIR (so the meta-loop is itself retro-able).

Hard rules

• Be honest, not flattering. Flattery breaks the feedback loop.
• Quote the transcript for claims. No paraphrase-passing-as-evidence.
• Single dimension per observation. Don't blend "communication AND audience read".
• No vague advice. "Be clearer" is useless. "Acknowledge AI limits in first 2 minutes when skeptics are present" is actionable.
• Cross-link related prior retros explicitly.
• One retro per meeting. Update in place if the slug already exists.

Verification before claiming done

1. ls <RETRO_OUTPUT_DIR>/meeting_retro_<...>.md confirms the file exists.
2. If Obsidian mirror configured: confirm the write returned success.
3. If memory index configured: confirm the store call returned an ID.
4. Confirm RETRO_INDEX_FILE has the new line.

Report measured numbers (file size, memory id) — never quote unmeasured stats.

Adapt to your toolchain

• Don't have Notion? Set TRANSCRIPT_SOURCE=file and point at local recorder output.
• Don't have a semantic index? The patterns subcommand reads the filesystem directly.
• Don't want a dashboard? Skip RETRO_DASHBOARD_PAGE_ID — filesystem retros are fully self-contained.