from __future__ import annotations

from app.agents.schemas.learning import LearningSignal, SessionRetrospective


class RetrospectiveSignalExtractor:
    def extract(self, retrospective: SessionRetrospective) -> list[LearningSignal]:
        signals: list[LearningSignal] = []

        if retrospective.outcome == "completed":
            signals.append(
                LearningSignal(
                    signal_type="workflow",
                    confidence=0.8,
                    evidence_refs=retrospective.evidence_refs[:3],
                    explanation="Completed runs can be mined as workflow hints later.",
                    payload={
                        "task_type": retrospective.task_type,
                        "execution_mode": retrospective.execution_mode,
                    },
                )
            )

        if len(retrospective.task_refs) > 1:
            context_snapshot = retrospective.context_snapshot or {}
            merge_report = dict(context_snapshot.get("merge_report") or {})
            verification_report = dict(context_snapshot.get("verification_report") or {})
            effectiveness_score = 1.0
            if merge_report.get("status") == "conflicted":
                effectiveness_score = 0.45
            elif merge_report.get("status") == "fallback":
                effectiveness_score = 0.25
            elif verification_report.get("status") == "failed":
                effectiveness_score = 0.3
            signals.append(
                LearningSignal(
                    signal_type="decomposition",
                    confidence=0.7,
                    evidence_refs=retrospective.task_refs[:3],
                    explanation="Multiple completed task refs indicate a decomposition pattern.",
                    payload={
                        "task_count": len(retrospective.task_refs),
                        "scheduled_subtask_count": context_snapshot.get("scheduled_subtask_count", 0),
                        "effectiveness_score": effectiveness_score,
                        "merge_status": merge_report.get("status"),
                    },
                )
            )

        if retrospective.used_skill_names:
            signals.append(
                LearningSignal(
                    signal_type="tool_success",
                    confidence=0.65 if retrospective.outcome == "completed" else 0.35,
                    evidence_refs=retrospective.evidence_refs[:2],
                    explanation="Task-scoped skill shortlist was available during this run.",
                    payload={"skills": retrospective.used_skill_names[:3]},
                )
            )

        if retrospective.outcome == "failed":
            signals.append(
                LearningSignal(
                    signal_type="correction",
                    confidence=0.75,
                    evidence_refs=retrospective.evidence_refs[:2],
                    explanation="Failed retrospectives should remain auditable before any promotion.",
                    payload={"verification_status": retrospective.verification_status},
                )
            )

        return signals