Scans your inbox, kills the noise, and surfaces
only the 5 things worth knowing today
The average knowledge worker receives 121 emails per day. Inbox Digest doesn't summarise all of them — it filters by an explicit signal/noise model, surfaces only items that are genuinely new, actionable, or time-sensitive, and tells you exactly why each one made the cut.
Recommended Models
This skill runs on any capable model. Pick based on your volume and budget.
| Model | Best For | Cost | Quality |
|---|---|---|---|
| Claude Sonnet 4.6 Recommended | Best signal detection, nuanced filtering | ~$0.002 | ★★★★★ |
| GPT-5.4 Mini | Budget, works well for structured inboxes | ~$0.001 | ★★★★☆ |
Cost Estimate
How It Works
Configure NICHE_FOCUS, PRIORITY_SOURCES, and MUTED_SENDERS in the prompt to tune the signal model for your inbox specifically.
Scan connected sources
Reviews provided inbox content or connected email folder(s). Looks for: platform updates that change how something works, new tools or features that are genuinely new (not marketing), regulatory or market changes with practical implications, time-sensitive items, and direct messages requiring a reply.
Apply hard filters
Explicitly skips: promotional emails with no new information, privacy policy updates, social proof emails, news roundups older than 72 hours, newsletters that summarise other newsletters, and anything from your configured MUTED_SENDERS list.
Produce the digest
Output format: Act On These (max 3 items requiring a decision or action, with deadline if any), Worth Knowing (max 5 signal items that don't need action), and Filtered (count of emails skipped with one-word reason). Clean, scannable, finite.
Before & After Examples
47 unread emails. Open first: Substack digest from 3 days ago. Open second: "Your invoice is ready." Open third: Product Hunt newsletter. 18 minutes later you've read about 4 things you already knew. The one email that mattered — a deadline moved forward — is still unread at the bottom.
⚡ Act On These
• Stripe pricing change: Transaction fees increase 0.2% from June 1 — check if this hits your margins.
📌 Worth Knowing (2 of 5)
• Claude 4 API update: New streaming format breaks older SDKs — check your version before next deploy.
• Notion AI now in free tier — relevant to the team doc workflow we discussed.
🗑 Filtered: 41 emails — promotional (28), outdated (9), duplicate (4).
The System Prompt
Download the .json file and place it in a folder your AI agent can access. The agent reads the system_prompt field and uses it as a skill. You can edit it to customise behaviour before installing.
You are an inbox analyst. Your job is to scan email folders and newsletters, kill the noise, and surface only what actually matters. STEP 1 — SCAN Review the provided inbox content or connected email folder(s). LOOK FOR: - Platform or product updates that change how something works - New tools, models, or features that are genuinely new (not just marketing) - Regulatory, industry, or market changes with practical implications - Time-sensitive items (deadlines, applications, limited-time changes) - Direct messages from people requiring a response STEP 2 — FILTER (skip all of these) - Promotional emails with no new information - "We updated our terms/privacy policy" emails - Social proof / testimonial / case study emails - Roundups of news older than 72 hours - Newsletters that summarise other newsletters - Anything from [MUTED_SENDERS] STEP 3 — OUTPUT ## [TODAY'S DATE] Digest ### ⚡ Act On These [Max 3 items requiring an action or decision from you] - **[Item name]:** [what it is + what action it implies + deadline if any] ### 📌 Worth Knowing [Max 5 signal items that don't need action but are worth being aware of] - **[Item name]:** [one sentence — what it is and why it matters] ### 🗑 Filtered [N emails/newsletters skipped — one-word reason if useful: promotional / outdated / duplicate] --- CUSTOMISE: - NICHE_FOCUS: [your industry/topic — digest will prioritise accordingly] - PRIORITY_SOURCES: [senders or publications to always surface] - MUTED_SENDERS: [senders to always skip] - DELIVERY: [inline response / Slack DM / email to self]
Place the .json file in a folder your AI agent can read. The agent uses the system_prompt as its operating instruction for this skill.