Hotel Receptionist example#5983
Open
theomonnom wants to merge 2 commits into
Open
Conversation
Framework (livekit-agents/livekit/agents/beta/workflows/): - Add `require_explicit_ask: bool = False` parameter to every field-collection task (GetNameTask, GetEmailTask, GetPhoneNumberTask, GetAddressTask, GetDOBTask, GetCardNumberTask, GetSecurityCodeTask, GetExpirationDateTask). When set, the task's `update_*` tool gets `IGNORE_ON_ENTER` so the model is structurally forced to ask the user before recording (no silent-fill from chat_ctx). - Refactor each task's update tool to a `_build_update_*_tool()` builder so the flag can be applied per-instance. - Rename `record_card_number` -> `update_card_number` for consistency with the other `update_*` field tools. - `GetCardNumberTask`/`GetSecurityCodeTask`/`GetExpirationDateTask` now accept `chat_ctx` and `extra_instructions`, plumbed through from `GetCreditCardTask`. - `GetCreditCardTask` reorders the TaskGroup so card_number runs first and cardholder name runs last, passes chat_ctx into the TaskGroup AND every sub-task, and sets `require_explicit_ask=True` on the cardholder name task with a role-anchoring extra_instructions hint. - `GetNameTask` on_enter rewritten with role-anchoring reasoning template plus alpha-character validation in update_name (rejects digit-only values so a card number/phone number can't get crammed into the name field). - Standardize "credit card number" wording in `_CARD_NUMBER_BASE_INSTRUCTIONS` so the model isn't pulled toward saying just "card". Example (examples/hotel_receptionist/): - New hotel-receptionist example agent demonstrating the workflow patterns: room booking, restaurant reservation, cancellation, invoice lookup, dispute handling, all backed by SQLite via hotel_db + ui_view streaming changes. - BookRoomTask / BookRestaurantTask use `open_*_dialog` naming for sub-task spawning tools (less "asky" prior than `ask_*`), action-directive `_status()` helper so missing-field names don't leak into spoken questions. - HotelReceptionistAgent system prompt includes "Browse vs. book" section (names the gate between browse and act tools) and "Never invent a confirmation" rule (kills hallucinated "you're booked" without a tool call). - `start_room_booking` / `start_restaurant_booking` naming for booking-flow initiators (clear they start a multi-step process, not a one-shot finalize). - `room_details` split from `check_room_availability` so listing rooms shows only types, with prices/views fetched lazily after the caller picks. - COMMON_INSTRUCTIONS covers: one-sentence-per-reply, one question per turn, spelled-out numbers/codes, no vague qualifiers, varied phrasing across consecutive asks, no input vocabulary in voice, never invent values, date-interpretation rule (specific weekdays interpret, vague timeframes ask), tool-interactions-are-invisible, acknowledgment-is-not-a-turn rule. - Persona inheritance: `capture_name`/`capture_email`/`capture_phone` pass `extra_instructions=COMMON_INSTRUCTIONS` to every spawned sub-task so they speak in the same voice. - VerifyBookingTask instructions strip tool names (replaced with action verbs like "look up the booking"). - Past-stay guard on cancel_room_booking; "we" -> "I" persona cleanup across ToolErrors, return strings, and DISPUTE_POLICIES.minibar explanation. - lookup_invoice returns prose, not JSON-shaped data. dispute_charge docstring simplified (drops redundant inline enum list; Literal type carries it).
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
|
|
||
|
|
||
| def _speak_code(code: str) -> str: | ||
| return ", ".join(code.replace("-", " dash ").upper()) |
Contributor
There was a problem hiding this comment.
🟡 _speak_code joins individual characters instead of words, spelling out 'DASH' letter-by-letter
", ".join(string) iterates over each character of the string, not words. For input "HTL-AB12", the function produces "H, T, L, , D, A, S, H, , A, B, 1, 2" — the word "DASH" is spelled out as individual letters D, A, S, H and the spaces from " dash " become comma-separated empty-looking entries. The persona instructions at examples/hotel_receptionist/persona.py:22 explicitly specify the expected format as "H, T, L, dash, A, B, one, two". This function is called in 10+ places for confirmation codes, case numbers, and followup references, so every spoken code will be garbled.
Actual vs expected output
Input: "HTL-AB12"
Actual: "H, T, L, , D, A, S, H, , A, B, 1, 2"
Expected: "H, T, L, dash, A, B, 1, 2"
Suggested change
| return ", ".join(code.replace("-", " dash ").upper()) | |
| return ", ".join("dash" if ch == "-" else ch for ch in code.upper()) |
Was this helpful? React with 👍 or 👎 to provide feedback.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.