Prompt library

Hand it to your agent.

Prompts you can paste into Claude, ChatGPT, OpenCode or any MCP client connected to veldhost. Each one lists the tools it uses, the permission it needs and what waits for your approval.

Diagnose

Your agent finds the cause. You decide the fix.

It checks live health, recent deploys and the error log, and tells you what broke and when. It changes your code, not our platform, and rolling back a bad release takes one confirmation.

Find out why a site returns errors

Prompt
shop has been returning 500 errors since this morning. Check its health and the nginx error log, tell me whether the last deploy caused it, and roll back if it did. Ask me before you roll back.

Tools: get_site_health · get_site_logs · list_deploys · rollback_release

Needs: read, deploy

Waits for you: A rollback is two-step: it shows what will change and only proceeds after you confirm.

Explain a failed deploy

Prompt
The last deploy of shop failed. Show me the step that broke and the lines that matter from its log, and suggest the fix in my code.

Tools: list_deploys · get_site_logs

Needs: read

Who changed what

Prompt
Something changed on shop yesterday. Go through the audit log and the deploy history and list every change in time order.

Tools: get_audit_log · list_deploys · get_site_settings

Needs: read

Diagnose

Ask why it’s slow. Get an answer, not a graph.

Your agent compares visitors, server errors and top pages with last week, lines them up against deploys and changes, and names the most likely cause. There’s no black-box anomaly detector. Your agent reads the same numbers you would, faster, and shows its working.

Explain a drop in traffic

Prompt
Traffic on example.com dropped yesterday afternoon. Compare it with the same day last week, check the server-error rate and recent deploys, and tell me the most likely cause.

Tools: get_site_analytics · get_site_health · list_deploys

Needs: read

Find the pages that fail

Prompt
shop feels broken this week. Find the pages with the most server errors and check the error log for what they have in common.

Tools: get_site_analytics · get_site_health · get_site_logs

Needs: read

A weekly health check

Prompt
Give me a short health report for all my sites: status, server-error rate, visitors against last week, disk use, the latest backup and anything that changed.

Tools: list_sites · get_site_health · get_site_analytics · get_site_traffic · list_backups · get_audit_log

Needs: read

Operate

Deploys, DNS and cron, by asking. Anything risky waits for you.

Deploy from git, refresh staging, set environment variables, add cron jobs and edit DNS on our EU nameservers. A zone import shows its plan before it writes, and DNSSEC asks you first.

Deploy to staging first

Prompt
Create a staging copy of shop and deploy the main branch to it. When it is healthy, show me the staging link so I can check it.

Tools: create_staging · deploy_site · get_staging_status · get_site_health

Needs: read, deploy, manage

Set up email DNS

Prompt
Add an SPF record for example.com that allows our mail provider, then check that every nameserver answers with it.

Tools: list_dns_records · add_dns_record · probe_domain_dns

Needs: read, dns

Move a DNS zone in

Prompt
Here is the zone file from my old provider. Import it for example.com as a preview first and show me what would change before you apply it.

Tools: import_dns_zone · list_dns_records

Needs: read, dns

Waits for you: The import previews by default and writes nothing until it is applied.

Schedule a nightly task

Prompt
Add a cron job on shop that runs the sitemap command every night at 03:00, and list the cron jobs afterwards so I can see it.

Tools: add_cron_job · list_cron_jobs

Needs: read, manage

Sign a domain with DNSSEC

Prompt
Check that every nameserver for example.com answers consistently, and if they do, turn on DNSSEC.

Tools: probe_domain_dns · get_domain_registrar · enable_dnssec

Needs: read, dns

Waits for you: Enabling DNSSEC is two-step: it returns a warning and only proceeds after you confirm.

Your data

From raw orders to a reporting table, every night, in SQL you can read.

Your agent writes the pipeline in plain SQL, checks that it compiles and previews the rows before anything runs, then puts it on a schedule. Runs are billed per minute under a cap you set, with a hard stop.

A nightly sales roll-up

Prompt
Build a pipeline that rolls up the orders in shop by day and product into my warehouse. Compile it, preview the first rows, then schedule it every night at 02:00.

Tools: list_databases · get_schema · create_pipeline · compile_pipeline · preview_pipeline · create_job

Needs: data

Change a pipeline safely

Prompt
Add a country column to my daily sales pipeline. Read the current definition first, change only that, and compile it before saving.

Tools: get_pipeline · update_pipeline · compile_pipeline

Needs: data

Watch a run, stop it if needed

Prompt
Run the nightly sales job now, follow it until it finishes, and cancel it if it runs longer than ten minutes.

Tools: run_job · get_job_run · cancel_job_run

Needs: data

Your data

Ask your data a question. Get a dashboard your team can open.

With the opt-in data scope it runs read-only SQL on your site databases and warehouse, and turns the answer into dashboard panels in Manage.

Top products on a dashboard

Prompt
What were my top 10 products last month by revenue? Put the answer on a dashboard called Sales, with a weekly trend panel next to it.

Tools: run_query · create_dashboard · add_panel · render_panel_data

Needs: data

A quick question, no dashboard

Prompt
How many new customers signed up in shop each week this quarter? Just show me the table.

Tools: list_databases · get_schema · run_query

Needs: data

Load a spreadsheet and report on it

Prompt
Create a warehouse database called dw with a table for the product costs I paste below, load them, and add a panel to the Sales dashboard that lists the ten most expensive products.

Tools: create_warehouse_database · run_warehouse_statement · add_panel

Needs: data

Waits for you: A statement that drops or deletes data is two-step and waits for your confirmation.

What always waits for you

Connect your assistant Missing a job? Tell us, and we build the tool.

  • An assistant can only propose an order: you approve it in Manage.
  • Restoring a backup, enabling DNSSEC, rolling back and deleting data are two-step and wait for your confirmation.
  • It can never do more than your own role allows, and every action lands in your audit log.
  • You choose the scopes, and revoke the connection whenever you like.