feat: inject briefing article content for deep article Q&A

Add _build_briefing_article_context() helper to llm.py that reads
rss_item_ids from briefing message metadata and injects article content
into the system prompt. Pass conv_id through build_context() and
generation_task.py.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-03-28 00:31:43 -04:00
parent 35f57e0d3e
commit 260103d533
3 changed files with 105 additions and 0 deletions
@@ -185,6 +185,7 @@ async def run_generation(
rag_project_id=rag_project_id,
workspace_project_id=workspace_project_id,
user_timezone=user_timezone,
conv_id=conv_id,
))
messages, context_meta = await context_task
+82
View File
@@ -453,6 +453,7 @@ async def build_context(
rag_project_id: int | None = None,
workspace_project_id: int | None = None,
user_timezone: str | None = None,
conv_id: int | None = None,
) -> tuple[list[dict], dict]:
"""Build messages array for Ollama with system prompt and context.
@@ -676,7 +677,88 @@ async def build_context(
f"\n\n--- Earlier Conversation ---\n{history_summary}\n--- End Earlier Conversation ---"
)
# Inject briefing article content for follow-up Q&A
if conv_id is not None:
from fabledassistant.models import async_session as _async_session
from fabledassistant.models.conversation import Conversation
async with _async_session() as _sess:
_conv = await _sess.get(Conversation, conv_id)
if _conv and getattr(_conv, "conversation_type", None) == "briefing":
article_context = await _build_briefing_article_context(conv_id)
if article_context:
system_parts.append(article_context)
messages = [{"role": "system", "content": "".join(system_parts)}]
messages.extend(history)
messages.append({"role": "user", "content": user_message})
return messages, context_meta
async def _build_briefing_article_context(conv_id: int) -> str:
"""Fetch article content from today's briefing message and return a
formatted context block for injection into the system prompt.
Looks at the most recent assistant briefing messages for rss_item_ids
in their metadata, then loads those items from the DB.
Capped at 10 articles × 500 chars to keep token use reasonable.
"""
import json as _json
from sqlalchemy import select, text as _text
from fabledassistant.models import async_session as _async_session
from fabledassistant.models.conversation import Message
async with _async_session() as session:
result = await session.execute(
select(Message)
.where(
Message.conversation_id == conv_id,
Message.role == "assistant",
)
.order_by(Message.created_at.desc())
.limit(10)
)
messages = result.scalars().all()
rss_item_ids: list[int] = []
for msg in messages:
meta = msg.msg_metadata or {}
if isinstance(meta, str):
try:
meta = _json.loads(meta)
except Exception:
continue
ids = meta.get("rss_item_ids") or []
if ids:
rss_item_ids = ids
break
if not rss_item_ids:
return ""
async with _async_session() as session:
result = await session.execute(
_text("""
SELECT i.title, i.url, i.content, f.title AS feed_title
FROM rss_items i
JOIN rss_feeds f ON f.id = i.feed_id
WHERE i.id = ANY(:ids)
ORDER BY i.published_at DESC NULLS LAST
LIMIT 10
""").bindparams(ids=rss_item_ids[:10])
)
rows = result.mappings().all()
if not rows:
return ""
lines = ["\n\nARTICLE CONTEXT (source articles from today's briefing):"]
for row in rows:
lines.append(f"\n[{row['feed_title']}] {row['title']}")
if row["url"]:
lines.append(f"URL: {row['url']}")
if row["content"]:
lines.append(row["content"][:500])
return "\n".join(lines)
+22
View File
@@ -56,6 +56,28 @@ def test_list_news_item_serialisation():
assert result["published_at"] == pub.isoformat()
def test_build_briefing_article_context_metadata_extraction():
"""Verify the rss_item_ids extraction logic from message metadata."""
import json
# Simulates the logic in _build_briefing_article_context
def extract_ids(msg_metadata):
meta = msg_metadata or {}
if isinstance(meta, str):
try:
meta = json.loads(meta)
except Exception:
return []
return meta.get("rss_item_ids") or []
assert extract_ids({}) == []
assert extract_ids(None) == []
assert extract_ids({"rss_item_ids": [1, 2, 3]}) == [1, 2, 3]
assert extract_ids(json.dumps({"rss_item_ids": [4, 5]})) == [4, 5]
assert extract_ids("not json") == []
def test_list_news_null_content_serialisation():
"""Null content should produce empty snippet without error."""
item = {