Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Install and configure Granola AI meeting notes with calendar and audio permissions. Use when setting up Granola for the first time, connecting Google/Outlook calendars, granting macOS Screen Recording permission, or configuring Windows audio capture. Trigger: "install granola", "setup granola", "granola calendar", "granola permissions".
.claude/skills/jeremylongshore-granola-install-auth/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 83% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 117% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 81% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 71% | 0% |
Install Granola, the AI notepad that captures meeting audio directly from your device (no bot joins the call), transcribes with GPT-4o/Claude, and produces structured notes with action items. Supports Zoom, Google Meet, Teams, Slack Huddles, and WebEx.
bash# macOS via Homebrew brew install --cask granola # Or download directly open "https://www.granola.ai/download"
Windows: download the installer from granola.ai/download and run the .exe.
macOS (critical — both required):
System Settings > Privacy & Security > Microphone
→ Enable Granola
System Settings > Privacy & Security > Screen & System Audio Recording
→ Enable GranolaThe Screen Recording permission is required because macOS bundles system audio capture under that category — Granola does not record your screen.
Windows: Microphone permissions are granted automatically. Confirm at:
Settings > Privacy & Security > Microphone → Granola enabledbash# macOS — confirm Granola is running pgrep -l Granola # Check installed version defaults read /Applications/Granola.app/Contents/Info.plist CFBundleShortVersionString 2>/dev/null || echo "Check Granola > About"
Join or start any meeting (Zoom, Meet, Teams). Granola shows a floating notepad when it detects a calendar event with a conferencing link. Verify the live transcription indicator appears.
| Setting | Location | Recommended | |---------|----------|-------------| | Auto-start with calendar | Preferences > General | On | | Default template | Preferences > Templates | Match your meeting type | | AI model | Uses GPT-4o/Claude | No configuration needed | | Auto-update | Preferences > General | On |
| Error | Cause | Fix | |-------|-------|-----| | No audio captured | Missing Screen & System Audio permission (macOS) | System Settings > Privacy & Security > Screen & System Audio Recording > enable Granola, then restart | | Calendar not syncing | OAuth token expired or wrong account | Disconnect calendar in Settings, re-authenticate | | App won't launch | macOS Gatekeeper block | Right-click Granola.app > Open, or xattr -cr /Applications/Granola.app | | Meeting not detected | Event has no video link | Add a Zoom/Meet/Teams link to the calendar event | | Bluetooth audio drops | BT device causes transcription stops | Switch to built-in mic or wired headset |
Your Device Audio ──→ Granola Desktop App ──→ Granola Cloud (transcription)
│ │
Local notepad GPT-4o / Claude
(your typed notes) (enhance + summarize)
│ │
└────── Merged Output ───┘
│
Structured meeting notes
with action itemsAfter installation, proceed to granola-hello-world for your first meeting capture.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,044 | 14,336 | +30% | 1 | 1 | 0% | 2,413 | 3,107 | +29% | 0 | 0 | — |
case-02 | fail→pass | 16,343 | 12,916 | -21% | 1 | 1 | 0% | 2,383 | 3,644 | +53% | 0 | 0 | — |
case-03 | fail→fail | 10,948 | 12,694 | +16% | 1 | 1 | 0% | 2,087 | 2,629 | +26% | 0 | 0 | — |
case-04 | pass→pass | 4,724 | 7,086 | +50% | 1 | 1 | 0% | 816 | 1,475 | +81% | 0 | 0 | — |
case-05 | fail→pass | 10,004 | 2,511 | -75% | 1 | 1 | 0% | 784 | 1,438 | +83% | 0 | 0 | — |
case-06 | pass→pass | 14,400 | 12,869 | -11% | 1 | 1 | 0% | 1,324 | 2,269 | +71% | 0 | 0 | — |
case-07 | fail→fail | 13,312 | 5,004 | -62% | 1 | 1 | 0% | 1,383 | 1,813 | +31% | 0 | 0 | — |
case-08 | pass→pass | 6,109 | 6,954 | +14% | 1 | 1 | 0% | 924 | 1,343 | +45% | 0 | 0 | — |
case-09 | pass→pass | 5,766 | 7,273 | +26% | 1 | 1 | 0% | 846 | 1,406 | +66% | 0 | 0 | — |
case-10 | pass→pass | 15,960 | 5,264 | -67% | 1 | 1 | 0% | 1,724 | 1,914 | +11% | 0 | 0 | — |
case-11 | fail→pass | 3,960 | 2,619 | -34% | 1 | 1 | 0% | 691 | 1,501 | +117% | 0 | 0 | — |
case-12 | pass→pass | 16,256 | 6,414 | -61% | 1 | 1 | 0% | 1,780 | 2,193 | +23% | 0 | 0 | — |
case-13 | pass→pass | 13,027 | 9,983 | -23% | 1 | 1 | 0% | 2,007 | 2,687 | +34% | 0 | 0 | — |
case-14 | pass→pass | 11,553 | 2,853 | -75% | 1 | 1 | 0% | 1,029 | 1,523 | +48% | 0 | 0 | — |
case-15 | pass→pass | 10,895 | 5,924 | -46% | 1 | 1 | 0% | 1,827 | 2,120 | +16% | 0 | 0 | — |
case-16 | pass→pass | 10,670 | 5,257 | -51% | 1 | 1 | 0% | 1,660 | 1,943 | +17% | 0 | 0 | — |
case-17 | pass→pass | 14,879 | 7,159 | -52% | 1 | 1 | 0% | 1,336 | 1,354 | +1% | 0 | 0 | — |
case-18 | pass→pass | 5,697 | 3,801 | -33% | 1 | 1 | 0% | 1,008 | 1,662 | +65% | 0 | 0 | — |
case-19 | pass→pass | 11,333 | 6,979 | -38% | 1 | 1 | 0% | 987 | 1,355 | +37% | 0 | 0 | — |
case-20 | pass→pass | 13,012 | 16,421 | +26% | 1 | 1 | 0% | 1,888 | 2,936 | +56% | 0 | 0 | — |
case-21 | pass→pass | 14,167 | 9,852 | -30% | 1 | 1 | 0% | 1,518 | 2,687 | +77% | 0 | 0 | — |
case-22 | pass→pass | 18,973 | 12,950 | -32% | 1 | 1 | 0% | 2,217 | 3,080 | +39% | 0 | 0 | — |
case-23 | pass→pass | 9,909 | 7,511 | -24% | 1 | 1 | 0% | 1,712 | 1,497 | -13% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.