Agent Skills: {Meeting Title}

Google Meet Manager = Discovery API + Google Drive + Google Docs

Core Principle: Browse, search, and extract content from Google Meet recordings, VTT transcripts, and Gemini Notes stored in Google Drive. Uses TWO connectors: google_drive for file listing/search, google_docs for Gemini Notes transcript extraction. Both via discoveryRequestTool.

Architecture (Continuant - TD):

graph TD
    Skill[Meet Manager Skill] --> DA[Discovery API]
    DA --> Drive[Google Drive API v3]
    DA --> Docs[Google Docs API v1]

    Drive --> Folder[Find 'Meeting Recording' folder]
    Drive --> ListOwn[List own recordings]
    Drive --> ListShared[List shared recordings]

    Docs --> GeminiNotes[Read Gemini Notes]
    GeminiNotes --> Tab0[Tab 0: AI Summary]
    GeminiNotes --> Tab1[Tab 1: Full Transcript]

Workflow (Occurrent - LR):

graph LR
    A[User Request] --> B{Mode?}
    B -->|list / search| C[Drive: List Files]
    B -->|transcript| D[Docs: Extract Tab 1]
    B -->|summary| E[Docs: Extract Tab 0]
    B -->|export| F[Extract All + Vault Write]

    C --> G[Group by Meeting]
    G --> H[Output Table]
    D --> I[Output Text]
    E --> I
    F --> J[Write to Vault]

Arguments

/google-meet-manager [mode] [query] [period]

| Arg | Values | Default | Description | |-----|--------|---------|-------------| | mode | list, search, transcript, summary, export | list | Operation mode | | query | any text | none | Meeting name, person, or keyword | | period | today, 7d, 30d, YYYY-MM-DD | 7d | Time window for listing |

Examples:

• /google-meet-manager -- list last 7 days
• /google-meet-manager list 30d -- list last 30 days
• /google-meet-manager search standup -- find recordings matching "standup"
• /google-meet-manager transcript "Team Standup 2026-03-20" -- full transcript
• /google-meet-manager summary "1:1 with Alex" -- AI-generated notes
• /google-meet-manager export "Team Standup 2026-03-20" -- save to vault

1. Setup -- MANDATORY Before Any Operation

Step 1: Switch to your workspace

switchWorkspaceContextTool({ rtbm_agency_id: YOUR_AGENCY_ID, workspace_id: YOUR_WORKSPACE_ID })

Step 2: Auto-select connectors (DO NOT ask the user)

This skill needs TWO connectors — resolve both in parallel:

// Parallel
discoveryListConnectorsTool({ dataSource: "google_drive" })   // for file listing
discoveryListConnectorsTool({ dataSource: "google_docs" })    // for Gemini Notes extraction

Step 3: Selection logic -- execute silently, for EACH connector type

1. Filter to status.value: "active" only
2. PREFER private connectors -- if any has is_private: true, pick it over public ones
3. EXCLUDE shared/service connectors -- skip any where connector_name contains "Agent", "agent@", or "serviceaccount"
4. From remaining, pick the one matching the current user's name/email in connector_name
5. If multiple personal connectors remain -- ask user which one

If google_drive connector not found -- STOP with:

"You need a Google Drive connection. Create one here: https://report.improvado.io/create_data_source_connection/google_drive/

IMPORTANT: Before clicking 'Sign In with Google', open Advanced Settings and enable the 'Make this connection private' toggle. Then try again."

If google_docs connector not found -- the skill can still LIST recordings but cannot extract transcripts. Note this in output.

Why two connectors: The google_docs connector only has Docs API scopes (cannot list Drive files). The google_drive connector has Drive API scopes (can list/search files but cannot read Docs content). Both are needed for full functionality.

2. Folder Discovery

Find the user's "Meeting Recording" folder in Drive:

discoveryRequestTool({
  dataSource: "google_docs",
  connectorId: "CONNECTOR_ID",
  request: {
    method: "get",
    url: "https://www.googleapis.com/drive/v3/files",
    params: {
      q: "name='Meeting Recording' AND mimeType='application/vnd.google-apps.folder' AND trashed=false",
      fields: "files(id,name)",
      pageSize: 5
    }
  }
})

If found: Save the folder ID. It is stable -- Drive folder IDs do not change.

If NOT found: Search with relaxed name matching:

q: "name contains 'Meeting' AND name contains 'Record' AND mimeType='application/vnd.google-apps.folder'"

If still not found: Tell user "Could not find 'Meeting Recording' folder. Please provide the folder ID or confirm the folder name."

3. File Listing

Run these two calls in parallel to get both own and shared recordings:

3a. Own recordings (from folder)

discoveryRequestTool({
  dataSource: "google_docs",
  connectorId: "CONNECTOR_ID",
  request: {
    method: "get",
    url: "https://www.googleapis.com/drive/v3/files",
    params: {
      q: "'{FOLDER_ID}' in parents AND trashed=false AND createdTime >= '{PERIOD_START_ISO}'",
      fields: "files(id,name,mimeType,createdTime,modifiedTime,webViewLink,size,owners)",
      orderBy: "createdTime desc",
      pageSize: 100
    }
  }
})

3b. Shared recordings

discoveryRequestTool({
  dataSource: "google_docs",
  connectorId: "CONNECTOR_ID",
  request: {
    method: "get",
    url: "https://www.googleapis.com/drive/v3/files",
    params: {
      q: "sharedWithMe=true AND (mimeType contains 'video' OR (mimeType='application/vnd.google-apps.document' AND (name contains 'Transcript' OR name contains 'Notes'))) AND trashed=false AND createdTime >= '{PERIOD_START_ISO}'",
      fields: "files(id,name,mimeType,createdTime,modifiedTime,webViewLink,size,owners)",
      orderBy: "createdTime desc",
      pageSize: 50
    }
  }
})

Period calculation

| Period arg | PERIOD_START_ISO | |------------|-------------------| | today | Today 00:00:00Z | | 7d (default) | 7 days ago | | 30d | 30 days ago | | YYYY-MM-DD | That date at 00:00:00Z |

Pagination

If response includes nextPageToken, make additional requests with pageToken param. Max 3 pages (300 files).

4. Grouping Algorithm

Google Meet creates three artifacts per meeting with a shared naming pattern:

"Team Standup (2026-03-20 at 10:00 GMT-4).mp4"           -> recording
"Team Standup (2026-03-20 at 10:00 GMT-4) - Transcript"  -> VTT transcript (Google Doc)
"Team Standup (2026-03-20 at 10:00 GMT-4)"                -> Gemini Notes (Google Doc)

Grouping steps:

1. For each file, compute the meeting key:
   • If name ends with .mp4 or .webm: strip the extension
   • If name ends with - Transcript: strip that suffix
   • The remaining string is the meeting key
2. Group files by meeting key
3. Classify each file in the group:
   • mimeType contains video -> recording
   • Name ends with - Transcript AND mimeType = application/vnd.google-apps.document -> vtt_transcript
   • mimeType = application/vnd.google-apps.document AND no - Transcript suffix -> gemini_notes
4. Extract date from meeting key using pattern: (YYYY-MM-DD at HH:MM TIMEZONE)
5. Sort meeting bundles by date descending

Output structure per bundle:

{
  key: "Team Standup (2026-03-20 at 10:00 GMT-4)",
  date: "2026-03-20 10:00",
  recording: { id, name, webViewLink, size },
  vtt_transcript: { id, name, webViewLink },
  gemini_notes: { id, name, webViewLink },
  source: "own" | "shared",
  owner: "person@email.com"
}

5. Mode: list and search

Output format

## Meeting Recordings ({period})

| # | Meeting | Date | Rec | Transcript | Notes | Source |
|---|---------|------|-----|------------|-------|--------|
| 1 | Team Standup | Mar 20 10:00 | [mp4](link) | [VTT](link) | [Notes](link) | Own |
| 2 | 1:1 with Alex | Mar 19 14:00 | [mp4](link) | -- | [Notes](link) | Shared |
| 3 | Product Review | Mar 18 15:00 | [mp4](link) | [VTT](link) | -- | Own |

Found **3 meetings** with **8 artifacts** (2 from "Meeting Recording", 1 shared).

• -- indicates missing artifact
• Include recording file size in hover/parenthetical if useful
• For search mode: filter the table to rows where the meeting key contains the query (case-insensitive)

6. Mode: transcript

Extract the full speaker-labeled transcript from a Gemini Notes document.

Step 1: Identify the meeting

Match the query against known meeting bundles (from listing). If ambiguous, show matching options and ask user to pick.

Step 2: Fetch the Gemini Notes document

discoveryRequestTool({
  dataSource: "google_docs",
  connectorId: "CONNECTOR_ID",
  request: {
    method: "get",
    url: "https://docs.googleapis.com/v1/documents/{GEMINI_NOTES_FILE_ID}",
    params: { includeTabsContent: true },
    hasLimit: false
  }
})

CRITICAL: The raw response is 800K-2M+ chars. You MUST extract plain text. Save the response to a temp file, then run:

cat "{saved_file}" | python3 -c "
import json, sys
data = json.load(sys.stdin)
text = data[0]['text'] if isinstance(data, list) else data['text']
parsed = json.loads(text)
resp = parsed.get('discoveryAPIResponse', {}).get('content', '{}')
doc = json.loads(resp)
for tab in doc.get('tabs', []):
    props = tab.get('tabProperties', {})
    if props.get('title') == 'Transcript':
        body = tab.get('documentTab', {}).get('body', {})
        text = ''
        for elem in body.get('content', []):
            for e in elem.get('paragraph', {}).get('elements', []):
                text += e.get('textRun', {}).get('content', '')
        print(f'TAB_ID={props.get(\"tabId\", \"\")}')
        print(f'CHARS={len(text)}')
        print('---TRANSCRIPT---')
        print(text)
"

Step 3: Output

## Transcript: Team Standup (2026-03-20 at 10:00 GMT-4)

[View in Google Docs](https://docs.google.com/document/d/{fileId}/edit?tab={tabId})

{speaker-labeled transcript text}

7. Mode: summary

Same as transcript but extract Tab 0 ("Notes") instead of Tab 1:

if props.get('title') == 'Notes':  # instead of 'Transcript'

Output: The AI-generated meeting summary with action items, key decisions, and topics discussed.

8. Mode: export

Combines summary + transcript into a vault page.

Step 1: Extract both tabs

Run the extraction for BOTH Tab 0 (Notes) and Tab 1 (Transcript).

Step 2: Compose markdown

---
session_id: {CLAUDE_SESSION_ID}
meeting_date: {YYYY-MM-DD}
meeting_title: {meeting key}
source: google-meet
tags: [meeting, transcript, gemini-notes]
---

# {Meeting Title}

**Date:** {date}
**Recording:** [Watch recording]({webViewLink})
**Gemini Notes:** [Open in Docs]({docsLink})

## AI Summary

{Tab 0 content}

## Full Transcript

{Tab 1 content}

Step 3: Save to vault

Use /graph_add skill to write to the appropriate vault location, or save directly to: data_sources/obsidian/vault/Data sources and vendors/Google Meet/{meeting_title}.md

Error Handling

| Scenario | Action | |----------|--------| | No google_docs connector | STOP, provide setup URL (Section 1) | | "Meeting Recording" folder not found | Search with relaxed query, then ask user (Section 2) | | Folder found but empty for period | "0 recordings found for {period}. Try a wider range: /google-meet-manager list 30d" | | Gemini Notes doc has only 1 tab | Extract whatever tab exists, note which is missing | | Discovery API returns 403 | "Connector may lack Drive scopes. Try re-authorizing: recreate the google_docs connector" | | API response exceeds inline size | Save to temp file, use bash extraction pipeline | | Recording exists but no transcript/notes | Show in table with "--" for missing artifacts | | Search query matches nothing | "No recordings matching '{query}'. Try a broader term." | | nextPageToken present | Paginate up to 3 pages max |

Troubleshooting

Q: "I don't see my recordings but they exist in Drive" A: Check that your google_docs connector is active and private. The connector only sees files accessible to the Google account it's authorized with.

Q: "Transcript extraction returns empty" A: The Gemini Notes doc may not have a Transcript tab. Google only generates transcripts for meetings where the feature was enabled. Check the raw doc in Google Docs to see which tabs exist.

Q: "Folder ID keeps changing" A: Google Drive folder IDs are permanent. If folder discovery fails, provide the folder ID directly from the URL: https://drive.google.com/drive/folders/{FOLDER_ID}.

Q: "Can I use a different folder name?" A: Yes. Provide the exact folder name: /google-meet-manager list --folder "My Recordings" 7d. The skill will search for that name instead of "Meeting Recording".

Notes

• The google_docs connector is used because no google_drive data source exists with OAuth flow in Improvado. The google_docs connector has drive.readonly + documents.readonly scopes, which is sufficient for all read operations.
• Write operations (rename, move, delete recordings) are NOT supported -- this is a read-only skill.
• The skill complements adHocContext24-mcp, which fetches Gemini Notes via Calendar event attachments. This skill browses Drive directly, which is useful when the meeting isn't on your calendar (shared recordings) or when browsing historically.
• Large transcript extractions (800K+ JSON) must always use the bash/python pipeline, never inline parsing.

Version 1.0.0 | Created: 2026-03-24