{ "_meta": { "description": "D6 fixtures for langgraph-python / hitl-in-app", "sourceFile": "d5-all.json", "created": "2026-05-21" }, "fixtures": [ { "match": { "userMessage": "Issue a $50 refund to customer #12345", "hasToolResult": false, "context": "langgraph-python" }, "response": { "toolCalls": [ { "id": "call_d5_request_approval_001", "name": "request_user_approval", "arguments": "{\"message\":\"Issue a $50 refund to customer #12345.\",\"context\":\"Per the standard goodwill-credit policy for shipping delays.\"}" } ] } }, { "match": { "userMessage": "Issue a $50 refund to customer #12345", "hasToolResult": true, "context": "langgraph-python" }, "response": { "content": "Approved — processing the $50 refund to customer #12345 now." } }, { "_comment": "hitl-in-app — refund pill (#12345), post-tool-result response. Content includes BOTH approve and reject key phrases so both test assertions can match (the test checks for a substring, not the full string). This avoids sequenceIndex which is fragile across aimock restarts.", "match": { "userMessage": "$50 refund to Jordan Rivera on ticket #12345", "toolCallId": "call_d5_hitl_refund_12345_001", "hasToolResult": true, "context": "langgraph-python" }, "response": { "content": "I am processing the $50 refund to Jordan Rivera on ticket #12345 now. The refund request was not approved by default — your explicit approval or rejection determines the outcome." } }, { "_comment": "hitl-in-app — refund pill (#12345). 1st turn: emit request_user_approval tool call. The post-tool-result fixture above (with toolCallId + hasToolResult: true) is more specific and wins on the 2nd turn, so hasToolResult: false is NOT needed here — omitting it allows this fixture to match even in multi-pill conversations where prior tool results exist.", "match": { "userMessage": "$50 refund to Jordan Rivera on ticket #12345", "context": "langgraph-python" }, "response": { "toolCalls": [ { "id": "call_d5_hitl_refund_12345_001", "name": "request_user_approval", "arguments": "{\"message\":\"Issue a $50 refund to Jordan Rivera on ticket #12345 for the duplicate charge.\",\"context\":\"Ticket #12345 — duplicate-charge goodwill credit.\"}" } ] } }, { "_comment": "hitl-in-app — downgrade pill (#12346), 2nd turn. Same pattern as refund/escalate. Without this entry, the generic 'plan' catch-all in feature-parity.json hijacks the downgrade message.", "match": { "userMessage": "downgrade Priya Shah (#12346) to the Starter plan", "toolCallId": "call_d5_hitl_downgrade_12346_001", "hasToolResult": true, "context": "langgraph-python" }, "response": { "content": "Downgrade confirmed — Priya Shah (#12346) will move to the Starter plan effective next billing cycle." } }, { "_comment": "hitl-in-app — downgrade pill (#12346). 1st turn: emit request_user_approval tool call. Specific userMessage substring takes precedence over the bare 'plan' fixture in feature-parity.json (which lives later in the load order). hasToolResult: false omitted so multi-pill conversations still match.", "match": { "userMessage": "downgrade Priya Shah (#12346) to the Starter plan", "context": "langgraph-python" }, "response": { "toolCalls": [ { "id": "call_d5_hitl_downgrade_12346_001", "name": "request_user_approval", "arguments": "{\"message\":\"Downgrade Priya Shah (#12346) to the Starter plan effective next billing cycle.\",\"context\":\"Ticket #12346 — voluntary downgrade per customer request.\"}" } ] } }, { "_comment": "hitl-in-app — escalate pill (#12347), post-tool-result response. Content includes both approve and reject key phrases.", "match": { "userMessage": "escalate ticket #12347 to the payments team", "toolCallId": "call_d5_hitl_escalate_12347_001", "hasToolResult": true, "context": "langgraph-python" }, "response": { "content": "Escalated ticket #12347 to the payments team for Morgan Lee. Not escalated by default — your explicit approval determines the outcome." } }, { "_comment": "hitl-in-app — escalate pill (#12347). 1st turn: emit request_user_approval tool call. hasToolResult: false omitted so multi-pill conversations still match (the post-tool-result fixture above with toolCallId is more specific and wins on the 2nd turn).", "match": { "userMessage": "escalate ticket #12347 to the payments team", "context": "langgraph-python" }, "response": { "toolCalls": [ { "id": "call_d5_hitl_escalate_12347_001", "name": "request_user_approval", "arguments": "{\"message\":\"Escalate ticket #12347 to the payments team for Morgan Lee's stuck payment.\",\"context\":\"Ticket #12347 — payment stuck, needs payments-team triage.\"}" } ] } } ] }