OfferCatcher

What It Does

Scans Apple Mail for recruiting emails, extracts important events (interviews, assessments, deadlines) with LLM, and syncs them to native Apple Reminders on iPhone/Mac.

Execution Boundary

• OpenClaw is responsible for orchestration: scan mail, ask the LLM to parse events, and decide whether anything should be written.
• scripts/apple_reminders_bridge.py is the only reminder write path.
• The bridge prefers remindctl (Swift + EventKit) and only falls back to AppleScript if remindctl is unavailable.
• Do not rely on node -> Reminders.app Automation as the primary path. On macOS this permission is often less stable than a native Reminders bridge.

How To Use

Trigger Phrases

• "Check my recruiting emails"
• "Any interviews coming up?"
• "Sync interview emails to reminders"
• "Don't let me miss my coding test"

Workflow

1. Scan:
   - OpenClaw heartbeat path: direct top-level `osascript -e ...` command
   - Manual/local CLI path: `python3 scripts/recruiting_sync.py --scan-only`
2. Parse: OpenClaw LLM extracts structured recruiting events
3. Apply: `--apply-events` → sends validated events to the native reminders bridge

Step 1: Scan Emails

osascript \
  -e 'tell application "Mail"' \
  -e 'set acc to account "谷歌"' \
  -e 'set mbx to mailbox "INBOX" of acc' \
  -e 'set output to ""' \
  -e 'set processedCount to 0' \
  -e 'repeat with m in messages of mbx' \
  -e 'if processedCount is greater than or equal to 300 then exit repeat' \
  -e 'set msgDate to date received of m' \
  -e 'if msgDate > ((current date) - (2 * days)) then' \
  -e 'set msgId to (id of m) as string' \
  -e 'set subj to subject of m as string' \
  -e 'set sndr to sender of m as string' \
  -e 'set ts to (date received of m) as string' \
  -e 'set c to content of m as string' \
  -e 'if (length of c) > 2000 then set c to text 1 thru 2000 of c' \
  -e 'set lineText to "谷歌" & (character id 31) & "INBOX" & (character id 31) & msgId & (character id 31) & subj & (character id 31) & sndr & (character id 31) & ts & (character id 31) & c' \
  -e 'set output to output & lineText & (character id 30)' \
  -e 'set processedCount to processedCount + 1' \
  -e 'end if' \
  -e 'end repeat' \
  -e 'return output' \
  -e 'end tell'

or for local debugging:

python3 scripts/recruiting_sync.py --scan-only

The heartbeat path returns raw mail records separated by character id 30 and
character id 31. The local debugging path returns JSON.

Step 2: LLM Parses

For each email, extract:

• company: Company name
• event_type: interview / ai_interview / written_exam / assessment / authorization / deadline
• timing: {"start": "YYYY-MM-DD HH:MM", "end": "..."} or {"deadline": "..."}
• role: Job title
• link: Event URL

Step 3: Apply Events

python3 scripts/recruiting_sync.py --apply-events /tmp/events.json

This does not write Reminders directly from OpenClaw itself. It always routes through scripts/apple_reminders_bridge.py.

For Mail reads, prefer the top-level osascript -e ... entry when OpenClaw is
the caller. This keeps Mail automation attached to the host process instead of a
Python child process.

LLM Parsing Prompt

Extract recruiting event information from this email. Return JSON.

Email:
{body}

Extract:
- company: Company name
- event_type: interview / ai_interview / written_exam / assessment / authorization / deadline
- timing: {"start": "YYYY-MM-DD HH:MM", "end": "..."} or {"deadline": "..."}
- role: Job title
- link: Event URL
- notes: Additional info

Output Rules

• Reminder title: Company + Event type (e.g., "Google Interview", "Meta Coding Test")
• Include: Time, role, link in notes
• Prefer native bridge writes through remindctl; if remindctl is unavailable, let the bridge use its AppleScript fallback
• If no new events: respond HEARTBEAT_OK

Configuration

~/.openclaw/offercatcher.yaml:

mail_account: "Gmail"    # Apple Mail account name
mailbox: INBOX           # Folder to scan
days: 2                  # Scan last N days
max_results: 60          # Max emails

Supported Languages

The LLM parser works with any language—Chinese, English, Japanese, German, etc. No regex, no language-specific rules.