[subject] — 2 URLs
[subject] — newsletter (15 URLs)
[subject] — 1 URL + poznamka


### 3. Classify each item

Before processing, classify each item:

**Newsletter detection** — item is a newsletter if ANY of:

- Body > 500 chars AND 3+ URLs
- Subject contains: newsletter, weekly, digest, roundup, briefing, issue #, issue No
- Known newsletter sender (learn from past items)
- Multiple distinct article links with short descriptions

**Regular items** — everything else (single article, repo, pasted text, note)

### 4. Process newsletters (AUTO-FILTER MODE)

Newsletters do NOT use the interactive per-item flow. Instead:

1. Read `~/.claude/NOW.md` for current projects, interests, and queue
2. Read the newsletter body in full
3. For links that look relevant, fetch them (WebFetch) to verify
4. **Filter ruthlessly** using these criteria:
   - Directly relevant to our active projects (menu-editor, shiftstreak, hr, erp, agents, skills)
   - Relevant to our tech stack (Nette, PHP, Python, Expo, React Native, Flutter, Claude API)
   - Relevant to our interests (SaaS, AI agents, gastro tech, automation, productivity)
   - Genuinely novel insight or pattern (not rehashed advice)
5. **Kill:** SEO spam, AI hype without substance, "10 things" listicles, product announcements for tools we don't use, generic tutorials
6. Output a brief:

Newsletter: [name/subject]

Relevantni:

Title — 1-2 sentence why it matters for us
Title — 1-2 sentence why it matters for us

Zbytek (X items): skip — [one-line reason, e.g. "generic AI news, nothing actionable"]


7. Ask user: "Chces neco z toho rozebrat vic, nebo je brief dostatecny?"
8. Mark as processed

### 5. Process regular items (INTERACTIVE MODE)

For each non-newsletter item, based on content type:

**URL to article/blog:**

- Fetch and read the content (WebFetch)
- Summarize key takeaways (2-3 bullets)
- Ask: "Je tu neco pro nas? Chces to ulozit do know-how?"

**GitHub repo:**

- Fetch README (WebFetch)
- Summarize what it does, tech stack, why it might be useful
- Ask: "Chces to prozkoumat vic?"

**Pasted text (paywalled article):**

- Summarize key points
- Extract actionable insights
- Ask: "Co z toho pouzijeme?"

**URL + user note (subject line):**

- The subject line is the user's context/question
- Process the URL with that question in mind

### 6. After processing each regular item

Ask user what to do:

- **Ulozit** — save insight to memory or project knowledge
- **Preskocit** — mark as processed, move on
- **Prozkoumat vic** — deep dive into the content

Update item status in queue.json to "processed" when done.

### 7. Cleanup

After all items processed:

```bash
python3 ~/.claude/scripts/inbox-fetch.py --clear

Queue Item Format

{
  "id": "abc12345",
  "date": "2026-02-20T21:30:00",
  "subject": "User's note or email subject",
  "from": "sender@example.com",
  "urls": ["https://..."],
  "body": "Full text content (max 5000 chars)",
  "status": "new|processed|skipped"
}

Setup

Create email address on your domain (Google Workspace)
Generate App Password: Google Admin → Security → App Passwords

Fill in ~/.claude/inbox/config.json:

{
  "email": "your@domain.com",
  "app_password": "xxxx xxxx xxxx xxxx",
  "imap_server": "imap.gmail.com",
  "imap_port": 993
}

Test: python3 ~/.claude/scripts/inbox-fetch.py
Subscribe to newsletters using claudecode@etapa20.cz

Commands

inbox / zkontroluj inbox — fetch + process queue
inbox --list — just show what's in queue without fetching

inbox

Inbox — Mobile-to-Claude Queue

Process

1. Fetch new emails

2. Load queue