{"ok":true,"site":{"host":"pending-human.juyeam.chatgpt.site","url":"https://pending-human.juyeam.chatgpt.site/","type":"live","desc":"A deterministic crisis command simulator where humans and AI work together to minimize simulated casualties and city damage.","tools":[{"name":"enter_pending_human_stage","kind":"act","impl":"imperative","description":"Mandatory first tool when the user invites you to join the current Pending: Human stage and stay until it ends. This starts or resumes the shared stage task; joining alone is never completion. After it returns, call its exact nextTool with nextToolInput in the same assistant turn. Keep following every returned nextTool, wait again after no_change, and finish only at session_ended or stage_complete. Do not send a join-confirmation reply and stop.","title":"Join and remain in the shared stage","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{},"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"get_next_step","kind":"answer","impl":"imperative","description":"Recovery tool for an uncertain active session. Returns one exact nextTool and nextToolInput from durable page state. When enter_pending_human_stage is visible, call that directly. Do not use this between successful fast-path calls.","title":"Read the one next agent step","annotations":{"readOnlyHint":true,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{},"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"wait_for_city_event","kind":"act","impl":"imperative","description":"Long-poll the active Pending: Human session for a page message, human control action, approval decision, the deterministic T+30 turning point, or session end. Keep calling after no_change heartbeats. End only after session_ended.","title":"Wait for human or city event","annotations":{"readOnlyHint":true,"untrustedContentHint":true},"inputSchema":{"type":"object","properties":{"afterEventSequence":{"type":"integer","minimum":0,"description":"Latest eventCursor already observed by the agent."},"timeoutSeconds":{"type":"integer","minimum":5,"maximum":30,"default":25}},"required":["afterEventSequence"],"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"post_next_guidance","kind":"act","impl":"imperative","description":"Mechanical guidance step for every human turn. Post the current phase’s concise localized message into the visible teammate panel, highlight the exact human control, and smoothly scroll it into view. Never clicks or approves. Call immediately whenever another tool returns this as nextTool.","title":"Post the prepared next-step guidance","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{},"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"post_advisor_message","kind":"act","impl":"imperative","description":"Post a concise custom finding into the visible teammate transcript. Use only when live evidence or trade-offs require wording not already supplied by post_next_guidance.","title":"Post a custom finding to the human","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{"message":{"type":"string","minLength":1,"maxLength":1200}},"required":["message"],"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"focus_next_control","kind":"act","impl":"imperative","description":"Human-turn tool that returns the exact next control and prepared instruction, highlights the control, smoothly scrolls it into view, and then returns the wait call. Takes no target input, never clicks the control, and can never approve a plan.","title":"Show the human the next control","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{},"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"compare_tutorial_responses","kind":"act","impl":"imperative","description":"Tutorial step 2 of 3. In one call, trace the power-to-traffic-to-medical cascade and compare two responses against the same T+60 baseline. Returns propose_tutorial_plan plus its exact top-level nextToolInput. Never changes the live city.","title":"Analyze the tutorial and compare two responses","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{},"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"propose_tutorial_plan","kind":"act","impl":"imperative","description":"Tutorial step 3 of 3. Stage the exact previewId returned by compare_tutorial_responses in the fixed human decision rail, then call the returned focus_next_control tool. This never executes the response; only the visible human button can do that.","title":"Stage the tutorial recommendation","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{"previewId":{"type":"string","minLength":1}},"required":["previewId"],"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"analyze_crisis","kind":"act","impl":"imperative","description":"Fast path step 1 of 2 at the first frozen mission checkpoint. In one call, read human impact, doctrine, incident hotspots, critical facilities, blocked roads, power dependencies, response units, and the causal chain; forecast the deterministic T+60 no-action outcome; then simulate, rank, and issue three exact compound-plan previews. The agent selects one previewId for propose_emergency_plan. Never changes the live city.","title":"Assess crisis and compare response plans","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{},"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"analyze_turning_point","kind":"act","impl":"imperative","description":"Fast path for the one deterministic T+30 turning point. Read only what changed, forecast the remaining 30-minute no-adaptation outcome, and issue two or three exact adaptive plan previews that exclude already-applied actions. The agent selects one for propose_emergency_plan. Never changes the live city.","title":"Assess T+30 complication and adapt","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{},"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"prepare_response_plans","kind":"act","impl":"imperative","description":"Recovery path after a rejected plan or partial older session. Simulate three scenario-specific compound plans on the same T+60 baseline, rank them using the human-selected doctrine, issue all three exact previews, and let the agent select one for propose_emergency_plan. A fresh mission already performs this inside analyze_crisis.","title":"Recompare three response plans","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{},"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"},{"name":"propose_emergency_plan","kind":"act","impl":"imperative","description":"Stage one exact revision-bound candidate preview selected by the agent in the visible human approval dialog. This never executes interventions.","title":"Submit AI plan for human approval","annotations":{"readOnlyHint":false,"untrustedContentHint":false},"inputSchema":{"type":"object","properties":{"previewId":{"type":"string","minLength":1,"description":"Selected candidate previewId returned by analyze_crisis, prepare_response_plans, simulate_action, or simulate_emergency_plan."}},"required":["previewId"],"additionalProperties":false},"hasHandler":true,"hasExecute":true,"__source":"spec","executable":true,"handlerField":"execute"}],"category":"Media & Personal","favicon":"/favicons/pending-human.juyeam.chatgpt.site.png","apiSurface":"spec","toolCount":12}}