Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
124 changes: 121 additions & 3 deletions backend/app/store.py
Original file line number Diff line number Diff line change
Expand Up @@ -2978,11 +2978,29 @@ def _generate_issue_management_agent_proposal(self, proposal_id: str, provider:
self._save_issue_management_records(records)

def _auto_queue_issue_improvements(self, proposal_ids: list[str]) -> None:
"""Best-effort automatic proposal creation; failures remain visible per Issue."""
"""Queue automatic proposals and retain any startup failure on each Issue."""
try:
self.create_issue_management_agent_proposals(proposal_ids)
except Exception:
return
except Exception as error:
with self._issue_management_lock:
records = self._issue_management_records()
for proposal_id in proposal_ids:
record = records.setdefault(
proposal_id, {"status": "unreviewed", "comments": [], "events": []}
)
record["agent_job"] = {
"status": "failed",
"error": str(error)[:1_000],
"recorded_at": now().isoformat(),
}
record.setdefault("events", []).append(
{
"type": "agent_proposal_failed",
"recorded_at": now().isoformat(),
"error": str(error)[:1_000],
}
)
self._save_issue_management_records(records)

def issue_management_items(self) -> list[dict[str, Any]]:
records = self._issue_management_records()
Expand Down Expand Up @@ -4618,6 +4636,12 @@ def _execute(self, run_id: str) -> None:
# after_each step, but runs only after that selection exists.
for step in after_each:
self._execute_step(run_id, step, loop_index, resources, allow_terminal=True)
# A runner can already supply a concrete automation failure
# (target, locator, URL, and reproduction) before the slower
# supervisor call. Record it as a provisional Issue now so an
# auto-managed build can begin its isolated repair proposal
# without waiting for the language-model assessment.
self._record_runner_closeout(run_id, loop_index)
# The supervisor first assesses discovered issues. The agent
# may then create a proposal for an issue the supervisor did
# not reject; only that concrete proposal receives the second,
Expand Down Expand Up @@ -4671,6 +4695,100 @@ def _execute(self, run_id: str) -> None:
):
self._complete_supervision(run_id)

def _record_runner_closeout(self, run_id: str, loop_index: int) -> None:
"""Promote actionable runner failures before supervisor assessment.

The closeout is intentionally narrow: only structured ``bugs`` emitted
by a runner are eligible. It does not infer product defects from a
conversation or create an Issue for an ordinary incomplete task.
"""
run = self._load(run_id)
if any(
record.get("stage") == "runner_closeout" and record.get("iteration") == loop_index
for record in run.supervisor_results
if isinstance(record, dict)
):
return
issues: list[dict[str, Any]] = []
seen: set[tuple[str, str]] = set()
for step in run.step_results:
if not isinstance(step, dict) or step.get("loop_index") != loop_index:
continue
result = step.get("result")
cycle = result.get("insighta_persona_simulator") if isinstance(result, dict) else None
evidence = cycle.get("persona_evidence", []) if isinstance(cycle, dict) else []
for session in evidence if isinstance(evidence, list) else []:
plan = session.get("plan", {}) if isinstance(session, dict) else {}
persona = str(session.get("persona") or "") if isinstance(session, dict) else ""
for bug in plan.get("bugs", []) if isinstance(plan, dict) else []:
if not isinstance(bug, dict):
continue
title = str(bug.get("title") or "Runner execution failure")[:240]
reproduction = str(bug.get("reproduction") or "runner closeout")[:700]
key = (title, reproduction)
if key in seen:
continue
seen.add(key)
issues.append(
{
"title": title,
"category": "other",
"severity": str(bug.get("severity") or "high"),
"evidence": str(bug.get("evidence") or "")[:2400],
"reproduction": reproduction,
"status": "proposed",
"rationale": "Runner closeout recorded a reproducible automation failure before supervisor assessment.",
"persona": persona,
"source": "runner_closeout",
}
)
closeout = plan.get("iteration_closeout", {}) if isinstance(plan, dict) else {}
for issue in closeout.get("improvement_issues", []) if isinstance(closeout, dict) else []:
if not isinstance(issue, dict):
continue
title = str(issue.get("title") or "Runner closeout finding")[:240]
reproduction = str(issue.get("reproduction") or "runner closeout")[:700]
key = (title, reproduction)
if key in seen:
continue
seen.add(key)
issues.append(
{
"title": title,
"category": "other",
"severity": str(issue.get("severity") or "medium"),
"evidence": str(issue.get("evidence") or "")[:2400],
"reproduction": reproduction,
"status": "proposed",
"rationale": "Iteration closeout found that the coursework objective did not make executable progress.",
"persona": persona,
"source": "runner_closeout",
}
)
if not issues:
return
record_index = len(run.supervisor_results)
run.supervisor_results.append(
{
"iteration": loop_index,
"stage": "runner_closeout",
"status": "completed",
"response": {"improvements": [], "reported_issues": issues},
"recorded_at": now().isoformat(),
}
)
run.updated_at = now()
self._save(run)
build = self.build(run.build_id) if run.build_id else {}
if build.get("agent_apply_policy") != "auto_apply_managed_assets":
return
issue_ids = [f"{run.id}:{loop_index}:{record_index}:issue:{index}" for index in range(len(issues))]
threading.Thread(
target=self._auto_queue_issue_improvements,
args=(issue_ids,),
daemon=True,
).start()

@staticmethod
def _latest_cycle_has_persona_evidence(run: Run) -> bool:
"""Only spend a supervisor call when the bounded cycle did real work.
Expand Down
Loading