Mortdecai
924f16b9da
New tools implemented and deployed to dev gateway: - log.query: focused event queries (chat/deaths/joins/actions), replaces 200-line dump - user.ask: risk-scaled clarifying questions, async with tellraw - journal.read/write: per-player files, cross-mode (God+Sudo share) All wired into langgraph_gateway.py _execute_tool and model-driven tool loop. Tool schemas updated (22 total). Deployed to CT 644 dev server. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
80 lines
2.6 KiB
Python
80 lines
2.6 KiB
Python
"""
|
|
User Ask — clarifying questions sent to the player in-game.
|
|
|
|
The model sends a question via tellraw and the gateway stores the pending
|
|
question state. The player's next chat message gets routed back as the
|
|
tool result.
|
|
|
|
Risk-scaled: model should exhaust journal/state/log queries before asking.
|
|
Low risk = just act creatively. High risk = ask first.
|
|
|
|
Implementation:
|
|
1. Model emits: <tool_call>{"name": "user.ask", "arguments": {"question": "..."}}</tool_call>
|
|
2. Gateway sends tellraw to the player
|
|
3. Gateway stores pending_question in session state
|
|
4. Player's next chat message becomes the tool result
|
|
5. Model continues with the answer
|
|
|
|
For training: simulate the ask/answer flow with synthetic responses.
|
|
For production: gateway handles the async wait.
|
|
|
|
Usage:
|
|
from agent.tools.user_ask import handle_user_ask, format_ask_tellraw
|
|
"""
|
|
|
|
import json
|
|
from typing import Any, Dict
|
|
|
|
|
|
def format_ask_tellraw(player: str, question: str, prefix: str = "[MORTDECAI]") -> str:
|
|
"""Format a clarifying question as a tellraw command."""
|
|
safe_q = question.replace('"', '\\"').replace("\\", "\\\\")
|
|
return (
|
|
f'tellraw {player} ['
|
|
f'{{"text":"{prefix} ","color":"gold","bold":true}},'
|
|
f'{{"text":"{safe_q}","color":"yellow","italic":true}}'
|
|
f']'
|
|
)
|
|
|
|
|
|
def handle_user_ask(config: dict, arguments: dict, rcon_fn=None) -> Dict[str, Any]:
|
|
"""
|
|
Send a clarifying question to the player.
|
|
|
|
In production: sends tellraw and returns a pending state.
|
|
The gateway is responsible for waiting for the player's response
|
|
and feeding it back as the tool result.
|
|
|
|
In training: the response is simulated in the training data.
|
|
|
|
Args:
|
|
config: server config
|
|
arguments: {"player": str, "question": str}
|
|
rcon_fn: function to execute RCON commands
|
|
|
|
Returns:
|
|
{"ok": True, "status": "pending", "question": question}
|
|
In production, the gateway replaces this with the actual player response.
|
|
"""
|
|
player = arguments.get("player", "")
|
|
question = arguments.get("question", "")
|
|
|
|
if not player or not question:
|
|
return {"ok": False, "error": "player and question required"}
|
|
|
|
# Send the question in-game
|
|
if rcon_fn:
|
|
prefix = config.get("god_chat_prefix", "[MORTDECAI]")
|
|
cmd = format_ask_tellraw(player, question, prefix)
|
|
try:
|
|
rcon_fn(cmd)
|
|
except Exception as e:
|
|
return {"ok": False, "error": f"Failed to send question: {e}"}
|
|
|
|
return {
|
|
"ok": True,
|
|
"status": "pending",
|
|
"player": player,
|
|
"question": question,
|
|
}
|