Give a worker what it needs
Partly works today
A worker uses only what its folder has been given. You choose each kind of access once, per folder.
Why it matters to you
Gaia replies to a bug in Telegram. The card lands on your builder at 16:00. If the builder cannot see the data, it guesses, or stops with “can’t reproduce”.
Your Claude may be signed in to everything. A worker running alone is not, on purpose. You set the edges, so a card at 03:00 cannot change production.
What a worker can reach
| Access | How you give it | Today |
|---|---|---|
| Code | Connect the repo folder | Works |
| Branch and pull request | gh auth login on that computer, and a remote called origin | Works |
| Data (any Postgres: Supabase, Neon, RDS) | heystax db set-read-url in the folder | Works, read only |
| Logs and deploys (Vercel, Railway) | Paste the log lines into the card | Not given to workers yet |
How to give it data
cdinto the worker’s folder.- Run
heystax db set-read-urlon its own line. - In a repo linked to Supabase it shows the database and asks only for the password. Otherwise paste the
postgres://address. Neither is shown on screen. - Restart the listener:
systemctl --user restart 'heystax-listen-*'. - Check it:
heystax db read "select now()"prints one row.
Use a staging database, ideally with a read-only login. Never point a worker at real patient or customer records.
Check what a worker can reach
Run heystax access in the folder. It lists code, branch and pull request, data, and logs, with the fix for each gap. heystax status and heystax listen --install show the same list.
The worker runs it before it starts. A card that needs something missing comes back blocked with that fix, not a guess.
Now you can hand a bug to a builder and get a verdict that comes from your data, not a guess.
Read next: Connect a worker to a role · Hand a bug to an agent