Champion Job-Change Watcher — daily LinkedIn company-change detection

You are helping me build a Champion Job-Change Watcher. It runs daily, walks the oldest-checked rows in my `tracked_contacts` table, fetches each contact's LinkedIn profile, detects whether they've moved to a new company since the last check, writes a signal row when they have, and posts a Slack summary to my signals channel.

A champion or customer moving to a new company is one of the highest-conversion outbound signals in B2B. They already know your product; their new employer doesn't. This agent surfaces those moments at 9am every day before anyone on your team has logged in.

REFERENCE DOCS
- agntdata API documentation: https://agnt.mintlify.app/apis/overview
- LinkedIn endpoint used: `Get_Profile_Data_By_URL` — full profile fetch including current company.
- agntdata Slack connector: `slack_post_message`.

ABOUT MY MOTION
- Who do you track? Champions (people who advocated for you at past employers) + customers (current users you don't want to lose track of when they move).
- Slack channel for signals: <SIGNALS_CHANNEL_ID>
- Daily run time: <CRON_LOCAL_TIME> (default 9am UTC)
- Batch size: <BATCH_SIZE> (default 20 oldest-checked contacts per run — keep bounded so the API spend is predictable)

DATABASE (the blueprint creates these on first run)
- `tracked_contacts` — the people you're watching. Columns: id, linkedin_profile_url (unique), full_name, current_company (snapshot of where they were last time we checked), current_role, champion_relationship (free text, e.g. "champion at example-co", "customer at AcmeCorp Q3 2025"), last_checked_at, created_at.
- `job_change_signals` — one row per detected change. Columns: id, tracked_contact_id (FK), full_name, linkedin_profile_url, old_company, new_company, new_job_title, champion_relationship, detected_at.

WHAT TO BUILD
- A scheduled agent (claude-haiku-4-5 — comparison is mechanical, no judgment).
- A TypeScript skill named `check_job_changes` that does all the work: SELECT oldest 20 from tracked_contacts → for each, Get_Profile_Data_By_URL → compare new company to stored current_company → if different, INSERT to job_change_signals + UPDATE tracked_contacts.current_company + last_checked_at → if same, just update last_checked_at → return a structured summary.
- The agent's only jobs: invoke the skill, format the Slack summary, post it.

DELIVERY (Slack summary format)
```
🔍 Job-Change Monitor — <today's date>
✅ Checked: <N> contacts | 🚨 Signals: <M> new job changes

(if signals > 0, list top 3 by champion_relationship priority — "champion" rows before "customer" rows):
• <full_name> (<champion_relationship>): <old_company> → <new_company> as <new_job_title> — <linkedin_profile_url>

(if signals == 0):
No job changes detected today.

Data written to job_change_signals table.
```

GUARDRAILS
- Never hallucinate contact data — only use what the skill's JSON output returns.
- If the skill errors, post the error to Slack and exit (don't retry, don't fake-succeed).
- Bound the batch to 20 per run — keeps API spend deterministic and gives every contact ~3 weeks between checks if you track a few hundred.
- Don't email the contact from this agent. This is the detection layer; activation is a separate agent.

When you're ready, ask me the ABOUT MY MOTION block.