feat(thinking): Phase 1 complete - parser, model, tests
This commit is contained in:
@@ -1,13 +1,11 @@
|
||||
# Implementation Plan: Rich Thinking Trace Handling
|
||||
|
||||
## Phase 1: Core Parsing & Model Update
|
||||
- [~] Task: Audit `src/models.py` and `src/project_manager.py` to identify current message serialization schemas.
|
||||
- [ ] Task: Write Tests: Verify that raw AI responses with `<thinking>`, `<thought>`, and `Thinking:` markers are correctly parsed into segmented data structures (Thinking vs. Response).
|
||||
- [ ] Task: Implement: Add `ThinkingSegment` model and update `ChatMessage` schema in `src/models.py` to support optional thinking traces.
|
||||
- [ ] Task: Implement: Update parsing logic in `src/ai_client.py` or a dedicated utility to extract segments from raw provider responses.
|
||||
- [x] Task: Audit `src/models.py` and `src/project_manager.py` to identify current message serialization schemas.
|
||||
- [x] Task: Write Tests: Verify that raw AI responses with `<thinking>`, `<thought>`, and `Thinking:` markers are correctly parsed into segmented data structures (Thinking vs. Response).
|
||||
- [x] Task: Implement: Add `ThinkingSegment` model and update `ChatMessage` schema in `src/models.py` to support optional thinking traces.
|
||||
- [x] Task: Implement: Update parsing logic in `src/ai_client.py` or a dedicated utility to extract segments from raw provider responses.
|
||||
- [ ] Task: Conductor - User Manual Verification 'Phase 1: Core Parsing & Model Update' (Protocol in workflow.md)
|
||||
|
||||
## Phase 2: Persistence & History Integration
|
||||
- [ ] Task: Write Tests: Verify that `ProjectManager` correctly serializes and deserializes messages with thinking segments to/from TOML history files.
|
||||
- [ ] Task: Implement: Update `src/project_manager.py` to handle the new `ChatMessage` schema during session save/load.
|
||||
- [ ] Task: Implement: Ensure `src/aggregate.py` or relevant context builders include thinking traces in the "Discussion History" sent back to the AI.
|
||||
|
||||
@@ -1,92 +1,59 @@
|
||||
from dataclasses import dataclass
|
||||
from typing import Optional
|
||||
import re
|
||||
|
||||
|
||||
@dataclass
|
||||
class ThinkingSegment:
|
||||
content: str
|
||||
marker_type: str
|
||||
|
||||
|
||||
def parse_thinking_trace(raw_response: str) -> tuple[Optional[ThinkingSegment], str]:
|
||||
if not raw_response:
|
||||
return None, raw_response
|
||||
|
||||
patterns = [
|
||||
(r"<thinking>\s*(.*?)\s*</thinking>", "xml"),
|
||||
(r"<thought>\s*(.*?)\s*</thought>", "xml"),
|
||||
(r"^Thinking:\s*\n(.+?)(?:\n\n|\n?$)", "text", re.MULTILINE),
|
||||
(r"^thinking:\s*\n(.+?)(?:\n\n|\n?$)", "text", re.MULTILINE),
|
||||
]
|
||||
|
||||
for i, pattern_info in enumerate(patterns):
|
||||
pattern = pattern_info[0]
|
||||
flags = pattern_info[2] if len(pattern_info) > 2 else re.DOTALL
|
||||
match = re.search(pattern, raw_response, flags)
|
||||
if match:
|
||||
thinking_content = match.group(1).strip()
|
||||
remaining = raw_response[: match.start()] + raw_response[match.end() :]
|
||||
remaining = remaining.strip()
|
||||
return ThinkingSegment(
|
||||
content=thinking_content, marker_type=pattern_info[1]
|
||||
), remaining
|
||||
|
||||
return None, raw_response
|
||||
from src.thinking_parser import parse_thinking_trace
|
||||
|
||||
|
||||
def test_parse_xml_thinking_tag():
|
||||
raw = "<thinking>\nLet me analyze this problem step by step.\n</thinking>\nHere is the answer."
|
||||
thinking, response = parse_thinking_trace(raw)
|
||||
assert thinking is not None
|
||||
assert thinking.content == "Let me analyze this problem step by step."
|
||||
assert thinking.marker_type == "xml"
|
||||
segments, response = parse_thinking_trace(raw)
|
||||
assert len(segments) == 1
|
||||
assert segments[0].content == "Let me analyze this problem step by step."
|
||||
assert segments[0].marker == "thinking"
|
||||
assert response == "Here is the answer."
|
||||
|
||||
|
||||
def test_parse_xml_thought_tag():
|
||||
raw = "<thought>This is my reasoning process</thought>\nFinal response here."
|
||||
thinking, response = parse_thinking_trace(raw)
|
||||
assert thinking is not None
|
||||
assert thinking.content == "This is my reasoning process"
|
||||
assert thinking.marker_type == "xml"
|
||||
segments, response = parse_thinking_trace(raw)
|
||||
assert len(segments) == 1
|
||||
assert segments[0].content == "This is my reasoning process"
|
||||
assert segments[0].marker == "thought"
|
||||
assert response == "Final response here."
|
||||
|
||||
|
||||
def test_parse_text_thinking_prefix():
|
||||
raw = "Thinking:\nThis is a text-based thinking trace.\n\nNow for the actual response."
|
||||
thinking, response = parse_thinking_trace(raw)
|
||||
assert thinking is not None
|
||||
assert thinking.content == "This is a text-based thinking trace."
|
||||
assert thinking.marker_type == "text"
|
||||
segments, response = parse_thinking_trace(raw)
|
||||
assert len(segments) == 1
|
||||
assert segments[0].content == "This is a text-based thinking trace."
|
||||
assert segments[0].marker == "Thinking:"
|
||||
assert response == "Now for the actual response."
|
||||
|
||||
|
||||
def test_parse_no_thinking():
|
||||
raw = "This is a normal response without any thinking markers."
|
||||
thinking, response = parse_thinking_trace(raw)
|
||||
assert thinking is None
|
||||
segments, response = parse_thinking_trace(raw)
|
||||
assert len(segments) == 0
|
||||
assert response == raw
|
||||
|
||||
|
||||
def test_parse_empty_response():
|
||||
thinking, response = parse_thinking_trace("")
|
||||
assert thinking is None
|
||||
segments, response = parse_thinking_trace("")
|
||||
assert len(segments) == 0
|
||||
assert response == ""
|
||||
|
||||
|
||||
def test_parse_multiple_markers_prefers_first():
|
||||
def test_parse_multiple_markers():
|
||||
raw = "<thinking>First thinking</thinking>\n<thought>Second thought</thought>\nResponse"
|
||||
thinking, response = parse_thinking_trace(raw)
|
||||
assert thinking is not None
|
||||
assert thinking.content == "First thinking"
|
||||
segments, response = parse_thinking_trace(raw)
|
||||
assert len(segments) == 2
|
||||
assert segments[0].content == "First thinking"
|
||||
assert segments[1].content == "Second thought"
|
||||
|
||||
|
||||
def test_parse_thinking_with_empty_response():
|
||||
raw = "<thinking>Just thinking, no response</thinking>"
|
||||
thinking, response = parse_thinking_trace(raw)
|
||||
assert thinking is not None
|
||||
assert thinking.content == "Just thinking, no response"
|
||||
segments, response = parse_thinking_trace(raw)
|
||||
assert len(segments) == 1
|
||||
assert segments[0].content == "Just thinking, no response"
|
||||
assert response == ""
|
||||
|
||||
|
||||
@@ -96,6 +63,6 @@ if __name__ == "__main__":
|
||||
test_parse_text_thinking_prefix()
|
||||
test_parse_no_thinking()
|
||||
test_parse_empty_response()
|
||||
test_parse_multiple_markers_prefers_first()
|
||||
test_parse_multiple_markers()
|
||||
test_parse_thinking_with_empty_response()
|
||||
print("All thinking trace tests passed!")
|
||||
|
||||
Reference in New Issue
Block a user