AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents
TypeScript ai-agentai-agent-toolsai-agentsai-agents-frameworkmcpmcp-servermcp-toolsmcpsn8n-alternativeno-code-automationworkflowworkflow-automationworkflows
docs(brain): pre-publish verification runs in the main chat, not a task (#16853)
Hazem Adel1409e1ef0d60
diff --git a/brain/knowledge/decisions/000052-pre-publish-verification-runs-in-the-main-chat-not-a-task.md b/brain/knowledge/decisions/000052-pre-publish-verification-runs-in-the-main-chat-not-a-task.mdnew file mode 100644index 00000000000..c7a8043e2c1--- /dev/null+++ b/brain/knowledge/decisions/000052-pre-publish-verification-runs-in-the-main-chat-not-a-task.md@@ -0,0 +1,24 @@+---+title: Pre-publish verification runs in the main chat, not a task+status: accepted+icon: ✅+---++# Pre-publish verification runs in the main chat, not a task++## Decision++The test-before-going-live check runs in the main chat. Tasks (`ap_run_task`) never test, publish or turn on a flow: `ap_test_flow`, `ap_test_step`, `ap_lock_and_publish` and `ap_change_flow_status` are left out of their tools. A task's check is validation, and it hands the flow back to the main chat to test and turn on.++## Context++000046 planned a separate verifier sub-agent, and 000048 made tasks general runners with the main chat's tools. That left open where the live test belongs. Decided 13 Oct on AI-5.++## Why++A live test can write real data, so it needs the user's approval, and `ap_test_flow` already shows that approval card in the main chat (`wrapTestFlowGate`). What comes after the test is a conversation: read the result, fix or ask, then "Turn it on?". Tasks can't ask the user questions, so a task that tests would hand the result straight back anyway. We rejected a verifier task for that reason: it adds a hop for a check that's one tool call in the main chat.++## Consequences++Each "Turn it on?" and every live test result shows up in the main conversation, so its context grows a little. Giving tasks test or publish tools later means moving the approval card and the write-check gate along with them.+