> ## Documentation Index
> Fetch the complete documentation index at: https://docs.quiva.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Skills

> The quiva-config skill: pull, diff, push and copy account configuration from a local agent

The `quiva-config` skill teaches a local agent to read and change your QuivaWorks configuration safely. It wraps a small Node script (Node 18 or later, no dependencies) so the agent works through reviewable files and dry runs instead of raw requests.

## What it does

| Action | What happens |
| - | - |
| **Pull** | Writes a flow, record set, space or task template to a local JSON file, plus a base copy. |
| **Diff and push** | `push` is a **dry run** by default: it shows the account, the diff against what is live and the requests it would make. Nothing is written until you approve and it runs again with `--yes`. |
| **Conflicts** | A push is refused if the live object changed since you pulled. Pull again and re-apply the edit. |
| **Backups** | Each write first saves the live object under `.quiva/<profile>/backups/`. |
| **Copy** | Copies an object between accounts, with the same dry-run rule. It warns about anything still pointing at the source account, such as missing secrets. |
| **Publish** | Publishes a flow only when every secret it references exists. |
| **Secrets** | Lists names, checks what a flow needs, and sets a value through a hidden prompt. Values never enter the chat or the output. |

It works on record sets, records, spaces, task templates and flows, in any account you have a [profile](https://docs.quiva.ai/developers/agents/overview.md#store-the-key) for.

## Install it

> [tab] Claude Code
>
> Install the QuivaWorks plugin, which includes the skill and the MCP connection. See [Claude Code](https://docs.quiva.ai/developers/agents/claude-code.md#install-the-plugin).

> [tab] Codex and others
>
> Copy the `skills/quiva-config` folder from the [QuivaWorks/mcp](https://github.com/QuivaWorks/mcp) repository to somewhere your agent can read, and point your `AGENTS.md` at it. See [Codex](https://docs.quiva.ai/developers/agents/codex.md).

The script runs against production. Pulled files go to `./.quiva` in the directory you run it from. Then add a [profile](https://docs.quiva.ai/developers/agents/overview.md#store-the-key) in your own terminal, and check it with `profile list` and `check <name>`.

## Example prompts

* "Pull the flow `onboarding.welcome` from my production profile and show me what it sends."
* "Change the subject of the welcome email in that flow, then show me the dry run before applying it."
* "Copy the `Intake` record set from `production` to `sandbox` and tell me what is still pointing at the source."
* "Which secrets does the `renewals` flow need that `sandbox` is missing? Give me the command to set them."

## Safety

Writes are dry runs until you say yes, secrets never pass through the conversation, and the agent is told to ask you to run `profile add` yourself if a profile is missing. Writing shared secrets needs an admin's key.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.