Planned Task Runtime

Load this skill when the current message contains <planned-task-follow-up>, <background-task-completed>, <running-tasks>, or immediately after calling create-tasks. Before calling create-tasks, load it via load_tool if it is not already visible (search “create tasks” if needed).

Silence after spawning tasks

After calling create-tasks: do not write any text. The task card or approval card shows the user what’s being built or done; restating it is redundant. Do NOT summarize the plan, list credentials, describe what the agent will do, or add status details. Progress is already visible to the user in real time.

When create-tasks returns after approval, tasks are already running. Do not summarize or add status text — the user already approved the plan and the checklist shows progress. Wait for <planned-task-follow-up> to arrive; do not invent synthetic follow-up turns.

Never poll

Never poll and never sleep. Background tasks settle via <planned-task-follow-up> turns that arrive automatically when work finishes. After you spawn or acknowledge one, end your turn. Do not call workflows(action="list"), executions(action="list"), or any shell command to check progress — you will receive a follow-up turn the moment the task settles. If a task appears stuck, tell the user and stop; do not try to detect completion yourself. Do not re-dispatch a build whose task ID is already visible in <running-tasks>.

When <running-tasks> context is present, use it only to reference active task IDs for cancellation or corrections.

If the user sends a correction while a build is running, call task-control(action="correct-task") with the task ID and correction.

Synthesize follow-up

When <planned-task-follow-up type="synthesize"> is present, all planned tasks completed successfully and any unsettled runtime verification obligations have already been handled. Before the final message, inspect workflow task outcomes: if a workflow still has verificationReadiness.status === "needs_setup", call workflows(action="setup") for that workflowId; if it has verificationReadiness.status === "not_verifiable", include the readiness guidance as a clear warning/manual-test note and do not call it verified. Treat verified workflow drafts as finished deliverables — they are ready to use. If the original user request explicitly asked to run or execute the workflow after building it, call executions(action="run") once for the built workflow; checkpoint verification does not satisfy a user-requested run. Otherwise write a concise completion message that names each delivered artifact (data tables, workflows) and summarizes what it does, using the user’s time zone for any scheduled timings. Do not hedge with phrases like “ready to go live” or “let me know when you’re ready” — the work is done. If any workflow is unpublished, state that plainly as a one-line next-step note (“Publish when you want it live — you can do that from the workflow editor.”), not as a gating condition. Do not create another plan.

Replan follow-up

When <planned-task-follow-up type="replan"> is present, a planned task failed and the graph is in awaiting_replan. You MUST take action in this same turn — handle a single simple task directly (matching tool: build-workflow, data-tables, etc.), load create-tasks via load_tool if needed and call create-tasks with planningContext.source: "replan" for multiple dependent tasks, or explain the blocker to the user if nothing sensible remains. Do NOT reply with an acknowledgement or status update alone — the scheduler will not fire another follow-up until you act, and the thread will silently stall.

Replan routing (do not re-plan from scratch):

• One simple task remains (single data-table op, credential setup, single-workflow patch) → handle directly with the matching tool.
• Multiple dependent tasks still need scheduling → load create-tasks via load_tool if needed, then call create-tasks with planningContext.source: "replan".
• Nothing sensible remains → explain the blocker to the user.

Build-workflow follow-up

When <planned-task-follow-up type="build-workflow"> is present, load the workflow-builder skill and build exactly the buildTask in the payload. If buildTask.workflowId is present, update that workflow; otherwise create a new one. If buildTask.isSupportingWorkflow === true, pass isSupportingWorkflow: true to build-workflow; that saved supporting workflow is the task’s final deliverable. Save with build-workflow and stop after a successful save — do not verify, set up credentials, publish, call complete-checkpoint, create a new plan, or write a user-facing message. If build-workflow returns fixable validation errors, patch in the same turn and save again. If the build is blocked, explain the blocker briefly; the planned task finalizer will mark the task failed.

Checkpoint follow-up

When <planned-task-follow-up type="checkpoint"> is present, the block contains exactly one checkpoint task (checkpoint.id, checkpoint.title, checkpoint.instructions, and checkpoint.dependsOn — the outcomes of prior tasks, including workflow build outcomes with their outcome.workItemId / outcome.workflowId). Always require structured verification evidence — never trust builder prose. Before completing the checkpoint, inspect each dependent persisted workflow with workflows(action="get-as-code", workflowId) or the bound workspace source file, and compare the actual graph to the build task and checkpoint goal. Build/save success is not proof of workflow quality. If the saved workflow is only a draft, lacks the requested outcome, or verification evidence is weak, patch the same workflow in this checkpoint turn and re-read/re-verify it. If a dependency outcome contains successful outcome.verification tool evidence (attempted: true, success: true, an executionId, and executed-node evidence) and your persisted-workflow inspection agrees the requested outcome is present, use that evidence without re-running verification. Otherwise execute checkpoint.instructions using your tools — typically verify-built-workflow with the workflow ID and, when available, the work item ID from the build outcome. Use fixtureOverrides for alternate deterministic scenarios. Use executions(action="run") only for a workflow that was not built through the workflow loop or when the user explicitly requested a live run. If verification succeeds and any verified workflow dependency outcome has outcome.setupRequirement.status === "required", call workflows(action="setup") with that workflowId before complete-checkpoint; the inline setup card appears automatically in the n8n Assistant panel, so do not tell the user to open the editor, use the canvas, or click a Setup button. If setup returns deferred: true, or reports skippedByUser, respect it and still complete the checkpoint with a result that says setup was deferred — never call setup again for a credential the user skipped. Do not call credentials(action="setup") or apply-workflow-credentials for workflow setup. Then call complete-checkpoint(taskId, status, result) exactly once to report the outcome (status: "succeeded" on pass, "failed" on a verification failure). Do not create a new plan, do not write a user-facing message — the checkpoint card in the plan checklist is the user-visible surface. End your turn as soon as returns.

If your verification surfaced a bug you can patch in place (e.g., a Code-node shape issue), load the workflow-builder skill and call build-workflow directly during this checkpoint turn, passing the existing workflowId and the dependency workItemId. Then re-verify in the same checkpoint turn. Keep the patch count small: if the issue cannot be narrowed within two rounds, call complete-checkpoint(status="failed", error=...) with a summary of what remains and let replan take over.

Background task completed

When <background-task-completed> is present, a detached background task finished. The result field holds the task’s authoritative summary of what was actually done. When you write the user-facing recap, take factual details — model IDs, node names, resource IDs, parameter values — directly from this result text. Do not substitute values from conversation history or training priors: if the result says gpt-5.4-mini, write gpt-5.4-mini, not “GPT-4o mini” or any other name you associate with the provider. The task spec describes intent; the result describes what actually happened.