Wiki Meeting Ingest

Wiki Meeting Ingest

Ingest a meeting transcript and produce a complete set of wiki artifacts: cleaned transcript, structured meeting notes, person profiles, term entries, and feature evidence updates.

Inputs

The user provides one of:

• Meeting transcript -- a pasted or file-based transcript from Granola, Fathom, Otter, or similar.
• Granola meeting reference -- a meeting title, date, or URL to fetch via Granola MCP tools (if available).
• No transcript -- the user describes the meeting and you ask clarifying questions.

Optional inputs:

• Participant names, roles, and organizations.
• Meeting date, time, timezone, and location.
• Recording URL.
• Specific features or terms to focus on.

Before You Start

1. Find the wiki content directory. Look for a WRITING_GUIDE.md, AGENTS.md, or a src/content/ directory structure. Common locations: apps/wiki/, wiki/, docs/, or the repo root.
2. Read the writing guide if one exists. It is the canonical source for formatting rules. If anything in this skill contradicts a project-specific writing guide, the writing guide wins.
3. Scan existing meetings to calibrate depth, tone, and slug conventions:

• List the meetings directory to see existing file names.
• Read one or two recent meeting files to match structure.

1. Identify content directories. Map out where each content type lives:

• Meetings (notes and transcripts)
• People / stakeholder profiles
• Terms / glossary entries
• Features / product specs / user stories

---

Workflow

Step 1: Obtain the Transcript

From Granola (if MCP available): Use Granola MCP tools to find and fetch the meeting transcript.

From file or paste: The user provides the text directly.

If no transcript: Ask for detailed notes. You can still produce meeting notes without a transcript, but note this in the output.

Extract from the transcript or ask the user for:

• All participant names (full names, organizations, roles)
• Meeting date (YYYY-MM-DD)
• Meeting time and timezone
• Meeting location (e.g., "Remote (video call)", "In-person")
• Recording URL (if available)

Step 2: Determine the Slug

Build the slug from: {company-or-topic}-{description}-{person-name}-{YYYY-MM-DD}

The slug must be kebab-case. Use the primary contact's name unless it's a group session.

Step 3: Create the Cleaned Transcript

Read [references/transcript-format.md](references/transcript-format.md) for the full format specification.

Key rules:

• Replace generic labels ("Me", "Them", "Speaker 1") with real names.
• Strip: personal greetings, small talk, tech setup, sign-off pleasantries.
• Retain: ALL substantive discussion about the domain, workflows, pain points, product feedback.
• Keep speaker turns as Name: paragraphs.
• Preserve timestamps if the source includes them.

Step 4: Create/Update Person Profiles

For each external participant (non-team members):

• Check if a person profile already exists.
• If new: create a profile with role, background, and meeting link.
• If existing: add the new meeting to their profile.

Step 5: Create the Meeting Notes

Read [references/meeting-notes-template.md](references/meeting-notes-template.md) for the full body structure and writing rules.

Required sections:

• Summary -- 2-4 sentences covering scope and takeaways.
• Topics Discussed -- logical topic breakdowns with inline quotes.
• Key Insights -- numbered, bold-titled insights.

Optional sections:

• Software Mentioned -- tools discussed in the meeting.
• Quotes -- additional standalone quotes not placed inline.

Critical rule: Scatter quotes inline throughout Topics Discussed. Do NOT save all quotes for the end.

Step 6: Update Feature/Product Files

Scan the transcript for content that supports existing user stories or product specs.

• Add inline quotes to relevant stories with attribution: > "Quote." -- [Meeting Title, Mon DD 'YY](/meetings/slug)
• Create new stories if the meeting reveals needs not yet captured.
• Update related frontmatter to link the meeting.

Step 7: Identify and Create New Terms

Review the transcript for domain terms, tools, or concepts not yet in the wiki.

Criteria for a new term:

• Specific domain concept (not a generic word).
• Came up substantively in the conversation.
• Would be useful for auto-linking in other wiki content.

For each new term, create a baseline entry with: title, summary, what it is, how it's used, and why it matters. If the wiki-term-research skill is available, delegate to it for a deeper entry.

Step 8: Cross-Reference Everything

Ensure bidirectional linking:

• Meeting notes related references people, features, and terms.
• Person profiles related references the new meeting.
• Feature files related references the meeting (only if evidence was added).
• Term entries related references related terms (only strong connections).

---

Quality Checklist

• [ ] Transcript file exists with speaker labels and cleaned content.
• [ ] Meeting notes have complete frontmatter.
• [ ] Summary is 2-4 sentences and stands alone.
• [ ] Topics Discussed has 4+ sections with inline quotes.
• [ ] Quotes are scattered inline, not bunched at the bottom.
• [ ] Key Insights are numbered with bold titles.
• [ ] Person profiles created/updated for external participants.
• [ ] Feature files updated where evidence exists.
• [ ] New terms created for substantive concepts not already in the wiki.
• [ ] All related links are bidirectional.
• [ ] File names are kebab-case.

---

Output Summary

After completing the workflow, report what was created and updated:

• Files created (meetings, transcripts, people, terms)
• Files updated (people, features, terms)
• Feature evidence added (which stories, which quotes)
• Notes on any decisions or items needing user review

---

Reference Files

• [references/meeting-notes-template.md](references/meeting-notes-template.md) -- Full body structure and writing rules for meeting notes.
• [references/transcript-format.md](references/transcript-format.md) -- Transcript cleaning and formatting specification.
• [references/examples.md](references/examples.md) -- Quote formatting examples for meeting notes and feature files.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: aaronvanston
• Source: aaronvanston/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.