lamatic-flows

name	lamatic-flows
description	Run tasks by discovering and executing deployed Lamatic flows through the GraphMCP server. Use this whenever the user asks to run, execute, or "do a task" with a Lamatic flow / project — e.g. "run the Changelog flow", "ask my RAG flow a question", "execute the flow that summarizes text". Handles picking the right flow, building a schema-matched payload, and reporting the result.

Lamatic Flows (GraphMCP)

This skill teaches you how to turn a natural-language task ("do a task for me") into a concrete Lamatic flow execution using the GraphMCP tools. GraphMCP gives you the tools; this skill gives you the procedure for using them well.

The tools you have

• graph_auth_login — one-time authentication (only if not pre-configured).
• graph_load_project_flows — LIST: every active flow + its input schema.
• graph_refresh_flows — re-run LIST to pick up newly deployed flows.
• graph_execute_flow — EXECUTE: run one flow with a payload.

The core loop

Follow these steps in order. Do not skip LIST — never guess a flowId.

1. Understand the task. Restate what the user wants in one sentence. If a project ID is neither given nor pre-configured, ask for it (or run graph_load_project_flows if a default project is configured).

2. LIST — which flow? Call graph_load_project_flows (or graph_refresh_flows if the user just deployed something). Read the returned flows, their triggers, and their Input Fields schema.

   • Match the user's request to a flow by name and by what its inputs suggest.
   • If two flows plausibly match, ask the user to pick — don't assume.
   • Skip flows tagged ⚠️ Not manually executable; tell the user why if that's the only match (they need trigger data, e.g. a cron or Google Sheets event).

3. How to call — build the payload. Construct a payload object whose keys exactly match the chosen flow's Input Fields.

   • Map the user's words to the right fields (e.g. their question → prompt).
   • Only include fields that exist in the schema. If a required field is missing from the user's request, ask for it before executing.
   • If a flow has no input schema, an empty payload is fine.

4. EXECUTE. Call graph_execute_flow with the flowId (and projectId if not pre-configured) and the payload you built.

5. Report. Summarize result for the user in plain language. If status is not success, surface the error and suggest the likely fix (see below).

Modes — this changes what LIST can do

• Full mode (org API key + project key + org ID): LIST and EXECUTE both work.
• Execution-only mode (Studio-injected PROJECT_API_KEY / PROJECT_ID / ENDPOINT): LIST now tries to discover flows directly from the project endpoint. If it succeeds, proceed normally. If LIST reports discovery is unavailable for this endpoint, ask the user for the flowId (or rely on the single pre-configured flow) and go straight to EXECUTE.

Handling common failures

• "Not authenticated" → run graph_auth_login, or tell the user to set the PROJECT_API_KEY / PROJECT_ID / ENDPOINT environment variables.
• "Project endpoint not available" → the project isn't deployed. Tell the user to deploy it first.
• "No flowId provided and no flow pre-configured" → run LIST and pick a flow, or ask the user which flow.
• Payload rejected / empty result → re-check your payload keys against the flow's Input Fields from LIST; a key mismatch is the usual cause.

Examples

• "run the Changelog flow with sampleInput 'hello world'" → LIST → find "Changelog" → payload { "sampleInput": "hello world" } → EXECUTE.
• "ask my RAG flow what Lamatic is" → LIST → find the RAG flow → map question to its prompt/query field → EXECUTE.
• "what flows can I run?" → LIST and summarize the executable ones.