Connect to the workflow WebSocket at:
ws://<host>:<port>/workflows/ws
| """ | |
| Slack HITL — Confirmation | |
| ========================= | |
| Billing ops agent that can cancel customer subscriptions. Cancellation is | |
| irreversible, so the destructive tool is wrapped with | |
| `@tool(requires_confirmation=True)` — Slack pauses with Approve / Deny | |
| buttons before the cancellation runs. The agent also has read-only | |
| lookup tools so it can show the customer's context before asking to | |
| confirm. |
| Parameter | Type | Required | Description |
|---|---|---|---|
step_requirements |
string (JSON array) |
Yes | Resolved StepRequirement objects from the paused response |
session_id |
string |
Yes | Session ID from the paused run |
| """ | |
| E2E test: Custom session_table and memory_table via Components API. | |
| 1. Starts an AgentOS server with a workflow (no custom tables on the workflow itself) | |
| 2. Registers a workflow config via POST /components with custom table names | |
| 3. Runs the workflow via POST /workflows/{id}/runs | |
| 4. Checks Postgres to see if custom tables were created | |
| This replicates the exact user flow from the issue. | |
| """ |
| """ | |
| Team Approval: member agent tool with @approval decorator. | |
| Demonstrates the approval system on a team — the tool pauses for both | |
| HITL confirmation AND creates an approval record in the DB that an admin | |
| can review. | |
| Based on cookbook/02_agents/11_approvals/approval_team.py | |
| """ |
| """Test Team HITL API flow with streaming everywhere. | |
| Tests: | |
| 1. Team run with streaming - pauses when member tool requires confirmation | |
| 2. Continue run with streaming - member agent streams after continue (fix for #7003) | |
| 3. Events are stored in DB (store_events=True auto-enabled by router) | |
| 4. Member responses are stored (store_member_responses=True auto-enabled by router) | |
| """ | |
| import json |
| """ | |
| Executor HITL via /continue API (Streaming) | |
| ============================================ | |
| Tests executor-level HITL when an agent inside a Step has a tool | |
| with requires_confirmation=True, using the AgentOS /continue API endpoint | |
| with streaming enabled. | |
| Flow: | |
| gather_data -> detailed_analysis (agent with HITL tool) -> report |
| """ | |
| Condition with Executor HITL via /continue API | |
| ================================================ | |
| Tests executor-level HITL when an agent inside a Condition step has a tool | |
| with requires_confirmation=True, using the AgentOS /continue API endpoint. | |
| Flow: | |
| gather_data -> Condition(evaluator=True) -> report | |
| | |
| [ | |
| { | |
| "input": { | |
| "input_content": "Deploy the payments app version 2.1 to the production environment. Ensure all pre-deployment checks are performed, including testing and security compliance. Provide a summary of the deployment process and any issues encountered." | |
| }, | |
| "model": "gpt-4o", | |
| "tools": [ | |
| { | |
| "result": "Successfully deployed payments v2.1 to production", | |
| "metrics": null, |
| [ | |
| { | |
| "input": { | |
| "input_content": "Call the break_model tool" | |
| }, | |
| "model": "gpt-4o", | |
| "tools": [ | |
| { | |
| "result": "Tool executed successfully.", | |
| "metrics": { |