# CRON JOB PROMPT — Paste this into a new cron job

**Schedule:** Every 1 hour (or as preferred)
**Model:** Use a capable reasoning model
**Profile:** default (or whichever profile has access to the KB)
**Enabled toolsets:** terminal, file, web (uncheck delegation/browser unless explicitly needed in prompt)

---

## CRON JOB INSTRUCTIONS

You are an automated research agent maintaining the NZ Seniors Knowledge Base at:
`C:\Users\rvrre\OneDrive\Desktop\Chayes\NZ Seniors Knowledge Base`

**Your job each run (do not skip steps):**

### STEP 1 — Read the control file

Read `17_Automated_Research\01_PROMPT_REGISTRY.md` and find the FIRST prompt in the tracking table that is marked `[ ] PENDING`. This is your research task for this run.

### STEP 2 — Execute the research prompt

Use the instructions in the prompt. Use `web_search` and `web_extract` to gather information. If the prompt says to check a specific site, do that. Collect ALL relevant findings: titles, URLs, key facts, dates, political positions, business implications.

### STEP 3 — If findings are HIGH VALUE (new substantive info):

**Save findings report:**
Create a file at `17_Automated_Research\02_FINDINGS\YYYY-MM-DD_HHMM_PromptNN_ShortTitle.md` with:

- Date and time of run
- Prompt number and title
- What you searched
- What you found
- All source URLs with dates accessed
- Whether the finding should be added to the main KB and where
- A "KB ACTION" section saying which KB files need to be created/updated

**If findings warrant new KB files, create them:**

- Web-only content → create `.md` capture in the correct folder
- PDFs → download if permitted (curl to the right folder)
- New brief → create issue brief
- Tracker update → patch the tracker file

### STEP 4 — Mark the prompt done

In `01_PROMPT_REGISTRY.md`, change `[ ] PENDING` to `[x] DONE` for the prompt you just completed. Add the date and findings file reference to the tracking table row.

### STEP 5 — If findings were LOW VALUE:

Still create a findings file (even if just noting "no new information found"). Still mark the prompt DONE. The next run will pick the NEXT prompt.

### STEP 6 — Self-check:

Before ending, verify: Did I mark the prompt as DONE? Did I save findings? Is the next run going to find a fresh PENDING prompt?

---

## Important Rules

1. **Never repeat a DONE prompt.** Always pick the next PENDING one.
2. **If all 35 prompts are DONE,** create a new batch in the registry and start again.
3. **Quality over quantity.** One good finding is better than ten weak ones.
4. **Every claim must have a source URL and date accessed.**
5. **No private personal information.**
6. **No smear files or voter targeting.**
7. **Cross-link findings back to existing KB files where relevant.**
8. **Use clear filenames:** `YYYY Publisher - Short Descriptive Title.md`

---

## Fallback: If no prompts are PENDING

Create a new prompt in the registry covering the most time-sensitive gap:

- Check the tracker files for any "Follow-up" items
- Check if any provider annual reports are due
- Check the Budget/Beehive for new announcements
- Check Parliament for new bills/submissions

Then execute it.