chihlasm/resolutionflow
1
0
Fork 0
Code Issues 23 Pull Requests 1 Actions Packages Projects Releases Wiki Activity

feat(pilot): [FIX_OUTCOME] marker parser + AI outcome proposal

Browse Source 
The AI emits [FIX_OUTCOME] when the engineer indicates in chat that a
prior suggested fix worked, didn't work, or was partially applied. The
marker writes to session_suggested_fixes.ai_outcome_proposal (JSONB),
which the frontend surfaces as a "confirm outcome?" banner. The status
column is only updated when the engineer clicks confirm (via PATCH
/outcome endpoint from Task 3).

Placeholder-only system prompt wiring comes in Task 5.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Michael Chihlas
2026-04-23 15:03:36 -04:00
parent 8988dbc885
commit c0112f8bee
2 changed files with 193 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

103
backend/app/services/unified_chat_service.py
View File

@@ -354,6 +354,56 @@ def _parse_suggest_fix_marker(
return cleaned, parsed
def _parse_fix_outcome_marker(
ai_content: str,
) -> tuple[str, dict[str, Any] | None]:
"""Extract a single [FIX_OUTCOME]...[/FIX_OUTCOME] JSON block.
Block shape:
{"fix_id": "<uuid>", "outcome": "success"|"failure"|"partial",
"reason": "<one-line>"}
Emitted by the AI when the engineer clearly indicates in chat that a
prior suggested fix worked, didn't work, or was partially applied.
The marker PROPOSES an outcome — the engineer confirms via the UI.
Only the last block in a response is honored.
"""
blocks = list(re.finditer(
r"\[FIX_OUTCOME\]\s*([\s\S]*?)\s*\[/FIX_OUTCOME\]", ai_content,
))
if not blocks:
return ai_content, None
last = blocks[-1]
raw = last.group(1).strip()
if raw.startswith("```"):
raw = re.sub(r"^```(?:json)?\s*", "", raw)
raw = re.sub(r"\s*```$", "", raw)
cleaned = re.sub(
r"\[FIX_OUTCOME\]\s*[\s\S]*?\s*\[/FIX_OUTCOME\]", "", ai_content,
).strip()
try:
data = json.loads(raw)
except (json.JSONDecodeError, ValueError) as e:
logger.warning("Failed to parse [FIX_OUTCOME] block: %s", e)
return cleaned, None
if not isinstance(data, dict):
return cleaned, None
fix_id = str(data.get("fix_id") or "").strip()
outcome = str(data.get("outcome") or "").strip().lower()
reason = str(data.get("reason") or "").strip()
if not fix_id or outcome not in {"success", "failure", "partial"}:
logger.warning("[FIX_OUTCOME] missing/invalid fields, dropping")
return cleaned, None
return cleaned, {"fix_id": fix_id, "outcome": outcome, "reason": reason}
async def _persist_suggested_fix(
*,
db: AsyncSession,
@@ -415,6 +465,39 @@ async def _persist_suggested_fix(
await db.flush()
async def _record_ai_outcome_proposal(
*,
db: AsyncSession,
session: AISession,
proposal: dict[str, Any],
) -> None:
"""Persist the AI's proposed outcome on the active fix.
Writes to session_suggested_fixes.ai_outcome_proposal. Frontend polls
the active fix and renders the AI-confirming banner state when this is
non-null. Does NOT mutate the fix's status — the engineer's confirmation
click via PATCH /outcome is what changes the status.
Drops silently when the fix_id isn't a valid UUID or doesn't belong to
this session.
"""
try:
fix_uuid = UUID(proposal["fix_id"])
except (ValueError, KeyError, TypeError):
logger.warning("[FIX_OUTCOME] invalid fix_id, dropping")
return
await db.execute(
update(SessionSuggestedFix)
.where(
SessionSuggestedFix.id == fix_uuid,
SessionSuggestedFix.session_id == session.id,
)
.values(ai_outcome_proposal=proposal)
)
await db.flush()
async def _persist_promote_items(
*,
db: AsyncSession,
@@ -566,6 +649,7 @@ async def send_chat_message(
branch_display, branch_questions_data = _parse_questions_marker(branch_display)
branch_display, branch_promote_items = _parse_promote_marker(branch_display)
branch_display, branch_suggest_fix = _parse_suggest_fix_marker(branch_display)
branch_display, branch_outcome_proposal = _parse_fix_outcome_marker(branch_display)
if branch_display != ai_content:
# Store stripped content in branch history
msgs[-1] = {"role": "assistant", "content": branch_display}
@@ -629,6 +713,12 @@ async def send_chat_message(
db=db, session=session, fix=branch_suggest_fix,
)
# Persist a [FIX_OUTCOME] proposal if the branch turn included one.
if branch_outcome_proposal is not None:
await _record_ai_outcome_proposal(
db=db, session=session, proposal=branch_outcome_proposal,
)
suggested_flows = extract_suggested_flows(
await rag_search(query=message, account_id=account_id, db=db, limit=8)
)
@@ -681,11 +771,16 @@ async def send_chat_message(
# Check for a [SUGGEST_FIX] marker — supersedes the prior active fix.
display_content, suggest_fix_data = _parse_suggest_fix_marker(display_content)
# Check for a [FIX_OUTCOME] proposal — AI confirms a prior fix's outcome.
display_content, outcome_proposal = _parse_fix_outcome_marker(display_content)
logger.info(
"Marker parsing results — actions: %s, questions: %s, fork: %s, "
"promote: %d, suggest_fix: %s, raw_length: %d, display_length: %d",
"promote: %d, suggest_fix: %s, outcome_proposal: %s, "
"raw_length: %d, display_length: %d",
bool(actions_data), bool(questions_data), bool(fork_data),
len(promote_items or []), bool(suggest_fix_data),
bool(outcome_proposal),
len(ai_content), len(display_content),
)
@@ -774,6 +869,12 @@ async def send_chat_message(
if suggest_fix_data:
await _persist_suggested_fix(db=db, session=session, fix=suggest_fix_data)
# Persist a [FIX_OUTCOME] proposal if this turn included one.
if outcome_proposal is not None:
await _record_ai_outcome_proposal(
db=db, session=session, proposal=outcome_proposal,
)
suggested_flows = extract_suggested_flows(rag_results)
return display_content, suggested_flows, session, fork_metadata, actions_data, questions_data

91
backend/tests/test_fix_outcome_marker.py Normal file
View File

@@ -0,0 +1,91 @@
"""Unit tests for the [FIX_OUTCOME] marker parser."""
from __future__ import annotations
from app.services.unified_chat_service import _parse_fix_outcome_marker
def test_parses_success_outcome():
ai = (
"Great news — that confirms the root cause.\n\n"
"[FIX_OUTCOME]\n"
'{"fix_id":"11111111-1111-1111-1111-111111111111",'
'"outcome":"success","reason":"user said the fix worked"}\n'
"[/FIX_OUTCOME]\n"
)
cleaned, parsed = _parse_fix_outcome_marker(ai)
assert "[FIX_OUTCOME]" not in cleaned
assert "confirms the root cause" in cleaned
assert parsed == {
"fix_id": "11111111-1111-1111-1111-111111111111",
"outcome": "success",
"reason": "user said the fix worked",
}
def test_parses_failure_outcome():
ai = (
"[FIX_OUTCOME]\n"
'{"fix_id":"22222222-2222-2222-2222-222222222222",'
'"outcome":"failure","reason":"user reports still broken"}\n'
"[/FIX_OUTCOME]"
)
cleaned, parsed = _parse_fix_outcome_marker(ai)
assert "[FIX_OUTCOME]" not in cleaned
assert parsed["outcome"] == "failure"
def test_missing_marker_returns_none():
ai = "no marker here"
cleaned, parsed = _parse_fix_outcome_marker(ai)
assert cleaned == ai
assert parsed is None
def test_invalid_json_is_dropped():
ai = "[FIX_OUTCOME]\nnot-json\n[/FIX_OUTCOME]"
cleaned, parsed = _parse_fix_outcome_marker(ai)
assert "[FIX_OUTCOME]" not in cleaned
assert parsed is None
def test_unknown_outcome_rejected():
ai = (
"[FIX_OUTCOME]\n"
'{"fix_id":"33333333-3333-3333-3333-333333333333",'
'"outcome":"maybe","reason":"x"}\n'
"[/FIX_OUTCOME]"
)
_, parsed = _parse_fix_outcome_marker(ai)
assert parsed is None
def test_last_block_wins_when_multiple():
ai = (
"[FIX_OUTCOME]\n"
'{"fix_id":"44444444-4444-4444-4444-444444444444",'
'"outcome":"failure","reason":"first"}\n'
"[/FIX_OUTCOME]\n"
"[FIX_OUTCOME]\n"
'{"fix_id":"55555555-5555-5555-5555-555555555555",'
'"outcome":"success","reason":"second"}\n'
"[/FIX_OUTCOME]"
)
cleaned, parsed = _parse_fix_outcome_marker(ai)
assert "[FIX_OUTCOME]" not in cleaned
assert parsed["fix_id"] == "55555555-5555-5555-5555-555555555555"
assert parsed["outcome"] == "success"
def test_parses_partial_outcome():
ai = (
"[FIX_OUTCOME]\n"
'{"fix_id":"66666666-6666-6666-6666-666666666666",'
'"outcome":"partial","reason":"user ran cred clear only"}\n'
"[/FIX_OUTCOME]"
)
_, parsed = _parse_fix_outcome_marker(ai)
assert parsed == {
"fix_id": "66666666-6666-6666-6666-666666666666",
"outcome": "partial",
"reason": "user ran cred clear only",
}