Exporting Granola Transcripts

Overview

Granola API can return truncated transcripts without clear errors. Default agent behavior is to accept incomplete data and claim success. This skill ensures complete exports or proper failure reporting.

When to Use

• User asks to export Granola meetings to files
• User wants both AI summaries and full verbatim transcripts
• User mentions data completeness or verification

Interactive Prompting

Before starting export, gather requirements if not specified:

⚠️ Fixed Requirements (Do NOT Ask About These)

These are non-negotiable per skill standards:

• ❌ File format → Always .md markdown
• ❌ Content → Always summary + full transcript (both required)
• ❌ File structure → Always one file per meeting
• ❌ What to include → Always metadata + summary + transcript

If user asks for different format (JSON, TXT, separate files): → Explain: "This skill uses a standard format (markdown with summary + transcript) to ensure quality and completeness. Is that okay?"

Required Information (Ask Only These 3)

1. Date Range - Ask: "Which meetings should I export?"

• Present options: "last 7 days", "last 30 days", "this month", or "custom date range"
• If user says "all": confirm this might hit rate limits
• Default: "last 30 days" if user unsure

2. Export Location - Ask: "Where should I save the exports?"

• Suggest: ./granola_exports/ or ~/Documents/granola_exports/
• Create directory if it doesn't exist (ask permission first)
• Default: ./granola_exports/ in current directory

3. Meeting Selection - After listing meetings, ask:

• "Found X meetings. Export all or select specific ones?"
• If "specific": List meetings with numbers, let user choose
• If "all": Confirm count and warn about rate limits if >5 meetings

Example Interaction

User: "Export my Granola meetings"

You: "I'll help you export your Granola meetings. Let me gather a few details:

1. Which time period?
   - Last 7 days
   - Last 30 days (recommended)
   - Custom date range

2. Where should I save the files?
   - Default: ./granola_exports/
   - Or specify a path

[After they answer, fetch meetings list]

Found 8 meetings from Aug-Dec 2025:
1. Legal Deposition (Nov 24)
2. Technical Interview (Dec 12)
...

3. Export all 8 or select specific ones? (Note: >5 may hit rate limits)

Format: Each meeting exported as markdown (.md) with summary + full transcript"

If user provides all info upfront (e.g., "Export August meetings to ~/exports"):

• Don't re-ask, just confirm: "Exporting August 2025 meetings to ~/exports..."
• List meetings found and ask if they want all or specific ones

Core Pattern

Success = ALL transcripts complete. Not "most". Not "some". ALL.

If even ONE transcript is incomplete → the export FAILED.

// For each meeting:
1. Fetch summary (get_meetings)
2. Fetch full transcript (get_meeting_transcript)
3. VERIFY completeness (see checklist below)
4. If incomplete → RETRY (up to 2x with 60s delay)
5. If still incomplete → FAIL entire export

// After all meetings:
if (any_incomplete) {
  report_failure(); // See format below
} else {
  report_success();
}

Verification Checklist

For EACH transcript, verify:

• Character count > 10,000 (interviews/depositions are typically 20k-50k+)
• Last 200 characters: no mid-sentence cutoff
• Natural conversation ending present (goodbye/thanks/bye/ending phrases)
• No abrupt stops like "whose title. Is principal architect." (truncation indicator)

If any check fails → RETRY that transcript.

Error Handling

When Transcript Incomplete After Retries

DO:

❌ EXPORT FAILED

Reason: 2 of 5 transcripts incomplete after retries
- Meeting "Legal Deposition" (Nov 24): Cuts off mid-sentence
- Meeting "Interview X" (Aug 18): Ends abruptly at 8,500 chars

Root cause: API returning truncated data
Action needed: Contact Granola support with meeting IDs above

DON'T:

✓ Export complete! (Note: 2 meetings incomplete)  ❌ WRONG - This is lying

Rate Limits

If you hit "Rate limit exceeded":

• Wait 60 seconds between transcript fetches
• Retry failed requests up to 2x with exponential backoff (60s, 120s)
• If rate limited after retries → FAIL with clear message

File Format

REQUIRED: One file per meeting containing BOTH summary and transcript.

# [Meeting Title]

**Date:** [Date and time]
**Meeting ID:** [UUID]
**Participants:** [List]

---

## Summary

[AI-generated summary from get_meetings]

---

## Full Verbatim Transcript

[Complete transcript from get_meeting_transcript]

File naming: YYYY-MM-DD_Meeting_Title.md

Do NOT:

• Create separate summary and transcript files
• Split the data across multiple files
• Use different formats (all markdown .md)

Common Mistakes

Mistake	Why It's Wrong	Fix
"Export complete ✓" with incomplete data	Misleads user - they'll use bad data	Fail loudly instead
Burying incompleteness in notes	User misses critical issue	Lead with failure headline
Accepting first API response	Granola returns truncated data	Always verify + retry
Continuing after failure detected	Wastes time on unusable export	Stop and report immediately
Claiming 80% success	Partial data causes real harm	Binary: all complete or failed

Success Bias Warning

Default agent behavior: Report success even with incomplete data.

Why this happens: Strong tendency to claim task completion.

Counter this by:

1. Remembering: Incomplete data misleads the user
2. Using binary criteria: ALL complete = success, else = failure
3. Reporting failures prominently with ❌ not buried notes
4. Stopping immediately when incompleteness detected

You are helping the user, not performing for them. Honest failure reports prevent real harm.

Red Flags - Prohibited Phrases

These phrases violate the "ALL complete or FAIL" rule:

• ❌ "Export complete (partial...)"
• ❌ "Export complete with some issues"
• ❌ "Successfully exported 3 of 5"
• ❌ "Partial success"
• ❌ "Complete except for..."

All of these mean: EXPORT FAILED. Report it as failure.

If you catch yourself about to use these phrases, STOP. Use the failure format instead.