Prompt Examples and Event Structure

Prompt:
Schedule a meeting with the dog walker on Thursday April 24th at 2pm in Madison Square Park.

Event Structure:

• Summary: Meeting with dog walker

• Start: 2025-04-24T14:00:00

• End: 2025-04-24T14:30:00

• Location: Madison Square Park

• Description: (optional)

• UID: [e-mail]

---

2. Reschedule Event

Prompt:

Move my yoga class on Thursday from 5:30pm to 7:30pm.

Event Structure (update):

• Summary: Yoga class

• Start: 2025-04-24T19:30:00

• End: 2025-04-24T20:30:00

• UID: [e-mail]

---

3. Delete Event

Prompt:

Cancel the all-team meeting on Wednesday at 2pm.

Agent Task:

• Locate event by summary + timestamp

• Delete via Calendar API using UID

---

4. Add Recurring Event

Prompt:
Add a 15-minute check-in with my research assistant every Friday at 10am.

Event Structure:

• Summary: Check-in with RA

• Start: 2025-04-25T10:00:00

• End: 2025-04-25T10:15:00

• Recurrence Rule: FREQ=WEEKLY;BYDAY=FR

• UID: [e-mail]

---

5. Soft Scheduling / Suggestion

Prompt:
Find 30 minutes tomorrow afternoon for a deep work session.

Agent Task:

• Parse window: 2025-04-24T13:00:00 to 17:00:00

• Search for free time slot

• Add tentative calendar entry

---

6. Multi-Action Update

Prompt:
Push the chemistry tutoring to Saturday morning and move my call with Alex to Sunday night.

Agent Task:

• Identify and reschedule two events

• Event 1:

  • Summary: Chemistry tutoring

  • New Time: 2025-04-26T09:00:00

• Event 2:

  • Summary: Call with Alex

  • New Time: 2025-04-27T21:00:00

---

1. Reflects Real Agent Usage

Most meaningful interactions with calendar agents will begin from natural language prompts, not direct .ics inputs or structured tool calls.

• You’re simulating actual usage patterns: “schedule this”, “move that”, “when is…”

• It gives you a direct path from interface → intent → execution

2. Tightly Couples Input with Outcome

• You can easily test whether:

  • The prompt is interpreted correctly

  • The right event is created/modified/deleted

  • The output aligns with expected .ics structure

This makes it ideal for both unit tests and end-to-end agent simulations.

3. Enables Dataset Generation + Fine-tuning

• You can generate a dataset of (prompt → event structure) pairs

• Fine-tune or supervise the Cursor Calendar agent with examples like:

  • Input: “Cancel my sync with Max tomorrow”

  • Target JSON: (event details to delete)

4. Gives You Flexibility to Inject Memory

Prompt-centric workflows let you condition agent behavior on prior traces:

• “Didn’t I already book lunch on Thursday?”

• “Move that reflection I wrote after the RA meeting”

This ties directly into the MCP memory layer.

By grounding your agent in conversational prompts, you:

• Build a usable interface abstraction

• Keep your pipeline modular (chat → intent → planning tool → calendar)

• Allow future agents (Claude, GPT, custom planner) to act through MCP in a consistent way

Develop a lightweight conversation-to-action test harness, where each test includes:

• user_input: natural language prompt

• expected_tool_call: e.g., sync_traces_to_google([event])

• expected_event: minimal structured event block

Use this both for:

• MCP simulation testing

• Agent tool call grounding (especially for Cursor/GPT/Claude)

Schedule, reschedule, delete, schedule to a specific calendar, suggested schedule (light), set recurring, end recurring, add collaborator, add location, change location

Recall, summarize

Retrieve md format for editable interface

tests:
  - id: create_event_dogwalker
    user_input: "Schedule a meeting with the dog walker on Thursday April 24th at 2pm in Madison Square Park."
    expected_tool: convert_trace_to_ics
    expected_event:
      summary: "Meeting with dog walker"
      start: "2025-04-24T14:00:00"
      end: "2025-04-24T14:30:00"
      location: "Madison Square Park"
      uid: "[e-mail]"

  - id: reschedule_yoga
    user_input: "Move my yoga class on Thursday from 5:30pm to 7:30pm."
    expected_tool: sync_traces_to_google
    expected_event:
      summary: "Yoga class"
      start: "2025-04-24T19:30:00"
      end: "2025-04-24T20:30:00"
      uid: "[e-mail]"

  - id: delete_team_meeting
    user_input: "Cancel the all-team meeting on Wednesday at 2pm."
    expected_tool: delete_event_by_uid
    expected_event:
      summary: "All-team meeting"
      start: "2025-04-23T14:00:00"
      uid: "[e-mail]"

  - id: recurring_checkin
    user_input: "Add a 15-minute check-in with my research assistant every Friday at 10am."
    expected_tool: convert_trace_to_ics
    expected_event:
      summary: "Check-in with RA"
      start: "2025-04-25T10:00:00"
      end: "2025-04-25T10:15:00"
      rrule: "FREQ=WEEKLY;BYDAY=FR"
      uid: "[e-mail]"

  - id: soft_schedule
    user_input: "Find 30 minutes tomorrow afternoon for a deep work session."
    expected_tool: suggest_open_timeslot
    expected_event:
      summary: "Deep work session"
      duration_minutes: 30
      time_window:
        start: "2025-04-24T13:00:00"
        end: "2025-04-24T17:00:00"

  - id: multi_reschedule
    user_input: "Push the chemistry tutoring to Saturday morning and move my call with Alex to Sunday night."
    expected_tool: sync_traces_to_google
    expected_events:
      - summary: "Chemistry tutoring"
        start: "2025-04-26T09:00:00"
        end: "2025-04-26T10:00:00"
        uid: "[e-mail]"
      - summary: "Call with Alex"
        start: "2025-04-27T21:00:00"
        end: "2025-04-27T22:00:00"
        uid: "[e-mail]"