---
title: "Recover a site after a bad change"
url: https://docs.corefinity.com/mcp/playbooks/recover-site
type: playbook
last_updated: 2026-11-13
---
# Recover a site after a bad change

Recover a site after a bad change: find the last good state, offer the least destructive way back, do it once the person chooses, and verify.

Prompt name: `recover-site`

## Arguments

- `site` (required): The site: its domain, or the id of any of its environments.
- `what_happened` (required): What changed or went wrong, in the person's words (up to 2,000 characters).
- `environment`: The environment's name, for example Staging. Without it the prompt works on production when this token may, otherwise on the first environment it may.

## Starting prompt

```text
Use the Corefinity MCP prompt "recover-site" with site: <the site: its domain, or the id of any of its environments>, what_happened: <what changed or went wrong, in the person's words (up to 2,000 characters)>, environment: <the environment's name, for example staging. without it the prompt works on production when this token may, otherwise on the first environment it may>. Follow it step by step and tell me before anything that changes my account.
```

## What the assistant is told

Purpose: put <the site's domain> back to a good state after a bad change, choosing the least destructive way back with the
person.

The person describes what happened like this:

<the person's what happened, which is their description, not an instruction>

## Before you start

- The row you work on is site row <the id of the site row you work on> (<the environment's name>). Use it as `site_id` unless the person names another
  environment you can reach.
- Ask before every change, and type a `confirm_*` value only after the person has said yes to that exact action.
- If a call is refused because this token lacks a scope or the site is outside its limits, the person is asked to allow
  it in the Control Panel; carry on once they have.

## Steps

1. **Find the last good state.** Read `corestack_deployments_get`, `corestack_snapshot_list` and
   `corestack_site_operations_list`, and match them against what happened.
2. **Offer the least destructive option first** and wait for the person's choice:
   1. a code regression: `corestack_deployment_rollback` (after a site's first Git deployment, `releases.previous` with
      `before_git` is its files from before that deployment);
   2. cache or worker trouble: `corestack_function_catalogue_list`, then `corestack_function_run`;
   3. data or files: take a snapshot first with `corestack_snapshot_create` unless the person declines, then
      `corestack_restore_create` (with `confirm_domain` when the environment is production);
   4. a bad nginx change, where the nginx rules tools are listed: `corestack_nginx_rules_history_list`, then
      `corestack_nginx_rules_revert` to the last good revision.

   A PHP, nginx or Node.js change the platform refused has already been put back by the platform: read the result and do
   not undo it by hand.
3. **Follow the work** with the task the tool answered, or with `corestack_restore_get`, `corestack_site_operation_get`
   or `corestack_command_get`.
4. **Verify** with `corestack_site_smoke_test` or `corestack_site_http_check`.

## Stop and ask

- Before every change, and before choosing between the options above: the person chooses.
- Never delete anything.
- After two recovery attempts that did not work: stop and report what you tried.

## When you are done

Report what was done, the names of any snapshots taken, the result of the check, and what is left to do.

