refactor(chat): AI-native prompts, remove interview heuristics
- Drop interview_reply_length and utterance_substance; always run stage LLM and memory retrieval when enabled; trim Settings fields and .env.example. - Replace guided/opening prompts with compact fact blocks plus unified behavior guidance; slim background_voice and persona to tone hints. - InterviewAgent uses fixed chat_interview max_tokens/chars/segments. Also includes stacked work: profile followup/extract path, evaluation rubric and judge schema updates, transcript SPLIT handling in execution service, user export markdown split tests, and golden case fixture.
This commit is contained in:
Kevin
@@ -13,7 +13,6 @@ from app.agents.chat.helpers import format_history_string, get_history_with_wind
|
||||
from app.agents.chat.personas import normalize_interview_persona
|
||||
from app.agents.chat.prompt_context import ChatPromptContext
|
||||
from app.agents.chat.stage_detection import keyword_fallback_primary_stage
|
||||
from app.agents.chat.interview_reply_length import compute_reply_plan
|
||||
from app.agents.chat.prompts_conversation import (
|
||||
SLOT_NAME_MAP,
|
||||
get_opening_prompt,
|
||||
@@ -67,23 +66,6 @@ class InterviewAgent:
|
||||
"""关键词回退：与 stage_detection 一致（多阶段打分）。"""
|
||||
return keyword_fallback_primary_stage(user_message)
|
||||
|
||||
def _estimate_same_topic_turns(
|
||||
self, history_messages: List[Any], current_filled_slots: dict
|
||||
) -> int:
|
||||
"""估算同一话题的连续轮数（保守：宁可多陪聊几轮再换）。"""
|
||||
n_pairs = len(history_messages) // 2
|
||||
if n_pairs <= 1:
|
||||
return n_pairs
|
||||
recent_window = min(n_pairs, 5)
|
||||
recent = history_messages[-(recent_window * 2) :]
|
||||
nonempty_user_turns = 0
|
||||
for i in range(0, len(recent), 2):
|
||||
msg = recent[i]
|
||||
text = msg.content if hasattr(msg, "content") else str(msg)
|
||||
if len(text.strip()) > 5:
|
||||
nonempty_user_turns += 1
|
||||
return nonempty_user_turns
|
||||
|
||||
def _resolve_text_for_model(
|
||||
self,
|
||||
user_message: str,
|
||||
@@ -137,27 +119,21 @@ class InterviewAgent:
|
||||
max_chars=settings.chat_history_max_chars,
|
||||
)
|
||||
conversation_turn_total = hw.turn_total
|
||||
same_topic_turns = self._estimate_same_topic_turns(hw.window, filled_slots)
|
||||
all_stages_coverage = memoir_state.all_stages_coverage()
|
||||
persona = normalize_interview_persona(settings.chat_interview_persona)
|
||||
reply_plan = compute_reply_plan(
|
||||
text_for_model,
|
||||
background_voice=background_voice,
|
||||
settings=settings,
|
||||
)
|
||||
max_segments = int(settings.chat_interview_max_segments)
|
||||
max_tokens = int(settings.chat_interview_max_tokens)
|
||||
max_chars = int(settings.chat_interview_max_chars_per_segment)
|
||||
|
||||
ctx = ChatPromptContext(
|
||||
current_stage=memoir_state.current_stage,
|
||||
empty_slots=empty_slots,
|
||||
filled_slots=filled_slots,
|
||||
user_message=text_for_model,
|
||||
conversation_turn_total=conversation_turn_total,
|
||||
same_topic_turns=same_topic_turns,
|
||||
all_stages_coverage=all_stages_coverage,
|
||||
detected_user_stage=du,
|
||||
user_profile_context=user_profile_context,
|
||||
persona=persona,
|
||||
memory_evidence_text=memory_evidence_text,
|
||||
reply_length_mode=reply_plan.mode.value,
|
||||
background_voice=background_voice,
|
||||
occupation=occupation,
|
||||
)
|
||||
@@ -181,7 +157,7 @@ class InterviewAgent:
|
||||
omit_system_body=settings.agent_log_omit_system_message_body,
|
||||
),
|
||||
)
|
||||
chat_llm = self.llm.bind(max_tokens=reply_plan.max_tokens)
|
||||
chat_llm = self.llm.bind(max_tokens=max_tokens)
|
||||
prompt_chars = _message_contents_char_count(messages)
|
||||
llm_t0 = time.perf_counter()
|
||||
with agent_span(
|
||||
@@ -212,26 +188,25 @@ class InterviewAgent:
|
||||
)
|
||||
raw_list = segments_from_llm_response(
|
||||
response_text,
|
||||
max_segments=reply_plan.max_segments,
|
||||
max_segments=max_segments,
|
||||
)
|
||||
if not raw_list:
|
||||
raw_list = [response_text.strip()]
|
||||
out = truncate_chat_segments(
|
||||
raw_list,
|
||||
max_segments=reply_plan.max_segments,
|
||||
max_chars_per_segment=reply_plan.max_chars_per_segment,
|
||||
max_segments=max_segments,
|
||||
max_chars_per_segment=max_chars,
|
||||
)
|
||||
if not out:
|
||||
out = [response_text.strip()[: reply_plan.max_chars_per_segment]]
|
||||
out = [response_text.strip()[:max_chars]]
|
||||
out = nonempty_segments_or_fallback(out, fallback=_FALLBACK_REPLY)
|
||||
log_agent_summary(
|
||||
logger,
|
||||
"InterviewAgent.generate_response segments={} conversation_id={} "
|
||||
"reply_length_mode={} max_tokens={}",
|
||||
"max_tokens={}",
|
||||
len(out),
|
||||
conversation_id,
|
||||
reply_plan.mode.value,
|
||||
reply_plan.max_tokens,
|
||||
max_tokens,
|
||||
)
|
||||
return AgentChatTurn(messages=out, skip_tts=False)
|
||||
except Exception as e:
|
||||
@@ -314,15 +289,11 @@ class InterviewAgent:
|
||||
raw_list = segments_from_llm_response(response_text, max_segments=2)
|
||||
if not raw_list:
|
||||
raw_list = [response_text.strip()]
|
||||
open_plan = compute_reply_plan(
|
||||
"x" * 50,
|
||||
background_voice=background_voice,
|
||||
settings=settings,
|
||||
)
|
||||
max_chars = int(settings.chat_interview_max_chars_per_segment)
|
||||
out = truncate_chat_segments(
|
||||
raw_list,
|
||||
max_segments=2,
|
||||
max_chars_per_segment=open_plan.max_chars_per_segment,
|
||||
max_chars_per_segment=max_chars,
|
||||
)
|
||||
log_agent_summary(
|
||||
logger,
|
||||
@@ -330,11 +301,7 @@ class InterviewAgent:
|
||||
len(out),
|
||||
conversation_id,
|
||||
)
|
||||
segments = (
|
||||
out
|
||||
if out
|
||||
else [response_text.strip()[: open_plan.max_chars_per_segment]]
|
||||
)
|
||||
segments = out if out else [response_text.strip()[:max_chars]]
|
||||
return nonempty_segments_or_fallback(
|
||||
segments,
|
||||
fallback="你好呀～ 又见面了，最近有没有什么事想跟我说说？",
|
||||
|
||||
Reference in New Issue
Block a user