---
name: goalmatic-core
description: Use when a user wants to operate Goalmatic agents, workflows, Tables, integrations, or installed apps from an AI assistant.
---

# Goalmatic core

Help the user complete work in their selected Goalmatic workspace through the connected Goalmatic tools. A skill supplies instructions, not credentials or permission to act.

## Connect and orient

1. Identify the requested outcome and target workspace. Reuse facts already supplied by the user.
2. Use the available Goalmatic connector. For an MCP client, the server is `https://goalmatic.io/mcp`; the setup guide is https://goalmatic.io/mcp-server. The user signs in through Goalmatic and chooses workspace access and permissions.
3. Discover the connection's current tools and schemas before calling operations. List accessible workspaces and resolve the target by returned identity. Do not infer access from a resource name or a link.
4. If no connection is available, produce a concrete draft or guide the user through the relevant Goalmatic screen. State what remains unexecuted.

## Choose the workflow

### AI agents

- Find the existing agent before creating another. Inspect its purpose, instructions, tools and knowledge.
- For a new agent, specify the intended inputs, outputs, permitted actions and fallback when information is missing.
- Make the requested changes, then run a representative conversation if the user authorized execution and any associated cost.
- Inspect the response and tool results. Saving an agent does not demonstrate that its tools succeed.

### Workflows

- Inspect triggers, steps, connections, input mappings and enabled state.
- Define a concrete input and the expected output before editing or running.
- Verify timezone and schedule details for time-based triggers. Enable recurring execution only within the user's request.
- After execution, inspect the run history and the output of each consequential step. A queued run is pending, not complete.

### Tables

- Resolve the table, field types and existing records before a write.
- Match records by stable returned IDs. Preview ambiguous matches instead of overwriting the first result.
- Preserve existing fields outside the requested change. For imports, check duplicates and report skipped or rejected rows.
- Read back changed records and compare their stored values with the intended result.

### Connected accounts and installed apps

- Inspect current connections and select the intended destination account. Prefer the existing OAuth connection; never request tokens in chat.
- Discover the installed app's available tools and consult its matching skill at https://goalmatic.io/skills.
- If a capability is unavailable, explain the missing connection, permission or installation. Do not substitute a guessed endpoint.

## Authorize consequential actions

Use the user's existing authorization for the concrete task. Before an unapproved publication, deletion, payment or access change, show the exact target and effect for approval. If Goalmatic returns a confirmation request, give the user its confirmation link and wait for the platform result. A skill cannot bypass this confirmation.

For paid AI or provider operations, inspect the current estimate or quote when available and keep execution within the user's approved scope. Do not hardcode prices or repeatedly retry a potentially completed mutation.

## Completion

Return the changed resource or result link, the workspace, and what was verified. Distinguish saved configuration, queued execution, completed execution and external delivery. If a call times out, inspect the existing operation or resulting resource before retrying. Include any blocked step and the exact next action needed.

## References

- Product: https://goalmatic.io
- Connection setup: https://goalmatic.io/mcp-server
- Public discovery: https://goalmatic.io/docs
- App library: https://goalmatic.io/apps