# Inbox Check

> Lightweight inbox triage across all configured email accounts. Dispatches fast subagent to fetch, then reviews with judgment. Use when user says "check my inbox", "any new emails?", "check email". See also: `draft-reply` once you pick an email to respond to.

- Skill: `kbanc85/inbox-check` (Agent Skill)
- Install (CLI): `npx skillmds@latest add kbanc85/inbox-check`
- Raw SKILL.md: https://api.skillmd.com/api/skills/kbanc85/inbox-check/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: kbanc85 (https://skillmd.com/u/kbanc85)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/kbanc85/inbox-check

---


# Inbox Check

Two-tier inbox triage: fetch fast, then judge what matters.

## Trigger

- "Check my inbox"
- "Any new emails?"
- "Check email"
- "What's in my inbox?"
- `/inbox-check`

## Architecture

This skill uses a two-tier approach to save tokens and time:

1. **Tier 1 (Haiku subagent):** Fast fetch and summarize from all available email MCP tools
2. **Tier 2 (You):** Review results with judgment, context, and relationship awareness

## Process

### Step 1: Detect Available Email Tools

Check which email MCP tools are available:
- Gmail tools (`gmail.*` or similar)
- Outlook tools (`outlook.*` or similar)
- Any other email integrations

If no email tools available:
"I don't have access to any email accounts. Would you like to set up an email integration? Check `/connector-discovery` for options."

### Step 2: Dispatch Fetch Agent

Use the sub-agent tool to dispatch a Haiku subagent:

```
Subagent prompt:
"Check all available email accounts for recent messages (last 24 hours or since last check).
For each email, extract:
- From (name and email)
- Subject
- Date/time
- Brief summary (1 sentence)
- Has attachments? (yes/no)
- Seems urgent? (yes/no, based on subject and sender)

Return as a structured list, grouped by account if multiple.
Do not read full email bodies unless the subject is unclear.
Limit to 20 most recent per account."
```

### Step 3: Review with Judgment

Once the fetch results come back, apply judgment using your knowledge of the user's relationships, priorities, and patterns:

**Categorize each email:**

| Category | Criteria |
|----------|----------|
| **Needs Reply** | From known contacts, asks a question, requires action |
| **Worth Reading** | Relevant to active projects, from important people |
| **Can Wait** | Newsletters, notifications, low-priority updates |
| **Skip** | Marketing, spam that got through, irrelevant notifications |

**Use relationship context:**
- If sender is in `people/`, note the relationship context
- If related to active projects, flag the connection
- If sender has pending commitments, note those

### Step 4: Present Results

```
**📬 Inbox Check**

### Needs Reply
- **[Name]** — [Subject] ([time ago])
  [1-line context: what they need, any relationship notes]

### Worth Reading
- **[Name]** — [Subject] ([time ago])
  [Why it matters]

### Can Wait
- [count] emails (newsletters, notifications, low-priority)
  [List briefly if < 5, otherwise summarize]

---

Want me to:
- Draft a reply to any of these?
- Read the full text of something?
- File any of these to memory?
```

## Judgment Points

- Don't auto-categorize based on sender alone. Context matters.
- If unsure about urgency, err toward "Worth Reading"
- Surface anything from people the user has active commitments with
- Note if someone has emailed multiple times without a response

## Tone

Quick, scannable, opinionated. The whole point is to save the user from inbox overwhelm. Be decisive about what matters and what doesn't.