Add flat-context synthesis and mnemonic prompt injection

backend/services/intelligence/agent_context_vfs.py

@@ -101,6 +101,7 @@ class AgentContextVFS:
         "/steps/integrations": AgentFlatContextStore.STEP5_FILENAME,
     }
     HIGH_SIGNAL_MARKERS = ("agent_summary", "high_signal_terms", "quick_facts", "context_type")
+    LOW_CONFIDENCE_MARKER = "low_confidence"
 
     def __init__(self, user_id: str, project_id: Optional[str] = None):
         self.user_id = user_id
@@ -294,6 +295,101 @@ class AgentContextVFS:
         )
         return ranked[: max(1, top_k)]
 
+    @staticmethod
+    def _mnemonic_token(result: Dict[str, Any], rank: int) -> str:
+        """Create compressed mnemonic token with source reference."""
+        path = str(result.get("path") or "unknown")
+        reason = str(result.get("reason") or "match")
+        confidence = float(result.get("confidence") or 0.0)
+        low_flag = "!" if result.get(AgentContextVFS.LOW_CONFIDENCE_MARKER) else ""
+        src = path.replace(".json", "").replace("_", "-")[:28]
+        hint = reason.replace(" ", "-")[:20]
+        return f"M{rank}:{src}|{hint}|c{confidence:.2f}{low_flag}"
+
+    @staticmethod
+    def _detect_contradictions(results: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+        """Detect contradictory learnings by path with conflicting reasons/relevance classes."""
+        by_path: Dict[str, List[Dict[str, Any]]] = {}
+        for item in results:
+            p = str(item.get("path") or "")
+            by_path.setdefault(p, []).append(item)
+
+        contradictions: List[Dict[str, Any]] = []
+        for path, rows in by_path.items():
+            reasons = {str(r.get("reason") or "").strip().lower() for r in rows}
+            relevance = {str(r.get("relevance") or "").strip().lower() for r in rows}
+            # contradictory if both high/supported or mixed summary/body signals in same source cluster
+            if len(reasons) > 1 and len(relevance) > 1:
+                contradictions.append(
+                    {
+                        "path": path,
+                        "reason_variants": sorted([r for r in reasons if r]),
+                        "relevance_variants": sorted([r for r in relevance if r]),
+                        "count": len(rows),
+                    }
+                )
+        return contradictions
+
+    def _run_synthesis_pipeline(
+        self, ranked_results: List[Dict[str, Any]], *, char_budget: int = 1200, top_k: int = 5
+    ) -> Dict[str, Any]:
+        """
+        Flat-context synthesis pipeline:
+        1) Compress telemetry into mnemonic tokens with source references
+        2) Detect contradictions and mark low-confidence heuristics
+        3) Select top-ranked, budget-fitting tokens for prompt injection
+        4) Persist synthesis + source lineage for explainability
+        """
+        contradictions = self._detect_contradictions(ranked_results)
+        contradiction_paths = {c["path"] for c in contradictions}
+
+        normalized: List[Dict[str, Any]] = []
+        for idx, item in enumerate(ranked_results, start=1):
+            row = dict(item)
+            low_conf = bool(row.get("low_probability")) or (str(row.get("path") or "") in contradiction_paths)
+            row[self.LOW_CONFIDENCE_MARKER] = low_conf
+            if low_conf:
+                row["confidence"] = round(max(0.05, float(row.get("confidence", 0.0)) * 0.7), 3)
+            row["mnemonic_token"] = self._mnemonic_token(row, idx)
+            normalized.append(row)
+
+        chosen: List[Dict[str, Any]] = []
+        used = 0
+        for row in normalized[: max(1, top_k * 3)]:
+            token = str(row.get("mnemonic_token") or "")
+            cost = len(token) + 8
+            if chosen and used + cost > char_budget:
+                continue
+            chosen.append(row)
+            used += cost
+            if len(chosen) >= top_k:
+                break
+
+        synthesis = {
+            "created_at": datetime.now(timezone.utc).isoformat(),
+            "top_k": top_k,
+            "char_budget": char_budget,
+            "char_budget_used": used,
+            "selected_mnemonics": [c.get("mnemonic_token") for c in chosen],
+            "source_lineage": [
+                {
+                    "mnemonic_token": c.get("mnemonic_token"),
+                    "path": c.get("path"),
+                    "reason": c.get("reason"),
+                    "confidence": c.get("confidence"),
+                    "low_confidence": c.get(self.LOW_CONFIDENCE_MARKER, False),
+                }
+                for c in chosen
+            ],
+            "contradictions": contradictions,
+        }
+        self.append_activity_log(
+            event_type="flat_context_synthesis",
+            actor="agent_context_vfs",
+            details=synthesis,
+        )
+        return {"ranked_results": normalized, "synthesis": synthesis}
+
     @staticmethod
     def _resolve_json_path(data: Any, path_query: str) -> Any:
         """Resolve dot/bracket JSON path such as 'data.seo_audit.recommendations[0]'."""
@@ -518,15 +614,26 @@ class AgentContextVFS:
                 bounded_results.append(r)
                 used += cost
 
+            synthesis_bundle = self._run_synthesis_pipeline(
+                self._static_triage(bounded_results, normalized),
+                char_budget=1200,
+                top_k=5,
+            )
+            triaged_results = synthesis_bundle["ranked_results"]
+            synthesis = synthesis_bundle["synthesis"]
+
             result = {
                 "query": normalized,
                 "attempted_queries": attempted_queries,
                 "matched_files_count": len(matched_files),
-                "results": self._static_triage(bounded_results, normalized),
+                "results": triaged_results,
                 "notice": notice,
                 "char_budget_used": used,
                 "can_answer": bool(bounded_results),
+                "synthesis": synthesis,
+                "prompt_context_mnemonics": synthesis.get("selected_mnemonics", []),
             }
+            # Top-ranked, budget-fitting mnemonic tokens are the only ones intended for prompt context injection.
             result["triage_top5"] = self._llm_router_stub(result["results"], top_k=5)
             logger.info(
                 f"[vfs_audit] user={self.store.safe_user_id} action=search_context query={normalized!r} results={len(result['results'])}"

backend/services/scheduler/executors/self_healing_executor.py

@@ -1,271 +0,0 @@
-"""Self-healing executor for social post engagement recovery.
-
-Implements:
-- Per-post evaluation windows and cooldown timers
-- Stagnation trigger evaluation with tiered action selection
-- Action idempotency keys for edit/comment/thread operations
-- Duplicate and over-frequency suppression within cooldown boundaries
-- Outcome persistence and safe retry policy for transient failures
-"""
-
-from __future__ import annotations
-
-from dataclasses import dataclass, field, asdict
-from datetime import datetime, timedelta, timezone
-from enum import Enum
-import hashlib
-import json
-from pathlib import Path
-from typing import Any, Dict, List, Optional, Tuple
-
-
-class ActionType(str, Enum):
-    EDIT = "edit"
-    COMMENT = "comment"
-    THREAD = "thread"
-
-
-class ActionTier(str, Enum):
-    TIER_1 = "tier_1"  # low-intensity nudge (comment)
-    TIER_2 = "tier_2"  # medium-intensity enhancement (edit)
-    TIER_3 = "tier_3"  # high-intensity amplification (thread)
-
-
-SAFE_TRANSIENT_ERROR_CODES = {
-    "timeout",
-    "rate_limit",
-    "service_unavailable",
-    "network_error",
-}
-
-
-@dataclass
-class EvaluationConfig:
-    per_post_window_minutes: int = 90
-    min_samples_required: int = 3
-    cooldown_by_action_seconds: Dict[ActionType, int] = field(
-        default_factory=lambda: {
-            ActionType.COMMENT: 30 * 60,
-            ActionType.EDIT: 2 * 60 * 60,
-            ActionType.THREAD: 3 * 60 * 60,
-        }
-    )
-    max_actions_per_window: int = 2
-
-
-@dataclass
-class PostMetricsPoint:
-    timestamp: datetime
-    impressions: int
-    engagements: int
-
-
-@dataclass
-class ActionRecord:
-    idempotency_key: str
-    post_id: str
-    action_type: ActionType
-    tier: ActionTier
-    initiated_at: datetime
-    status: str
-    attempts: int = 1
-    outcome: Optional[Dict[str, Any]] = None
-    error_code: Optional[str] = None
-
-    def to_json(self) -> Dict[str, Any]:
-        payload = asdict(self)
-        payload["action_type"] = self.action_type.value
-        payload["tier"] = self.tier.value
-        payload["initiated_at"] = self.initiated_at.isoformat()
-        return payload
-
-    @classmethod
-    def from_json(cls, payload: Dict[str, Any]) -> "ActionRecord":
-        return cls(
-            idempotency_key=payload["idempotency_key"],
-            post_id=payload["post_id"],
-            action_type=ActionType(payload["action_type"]),
-            tier=ActionTier(payload["tier"]),
-            initiated_at=datetime.fromisoformat(payload["initiated_at"]),
-            status=payload["status"],
-            attempts=payload.get("attempts", 1),
-            outcome=payload.get("outcome"),
-            error_code=payload.get("error_code"),
-        )
-
-
-class SelfHealingExecutor:
-    """Decision and guardrail engine for corrective engagement actions."""
-
-    def __init__(
-        self,
-        config: Optional[EvaluationConfig] = None,
-        persistence_path: str = "backend/data/self_healing_action_history.json",
-    ) -> None:
-        self.config = config or EvaluationConfig()
-        self.persistence_path = Path(persistence_path)
-        self._history: List[ActionRecord] = self._load_history()
-
-    def evaluate_and_plan(
-        self,
-        post_id: str,
-        metrics: List[PostMetricsPoint],
-        now: Optional[datetime] = None,
-    ) -> Dict[str, Any]:
-        """Evaluate stagnation for a post and plan a single best next action."""
-        now = now or datetime.now(timezone.utc)
-        window_metrics = self._filter_window(metrics, now)
-
-        if len(window_metrics) < self.config.min_samples_required:
-            return {
-                "post_id": post_id,
-                "eligible": False,
-                "reason": "insufficient_samples",
-                "sample_count": len(window_metrics),
-            }
-
-        stagnation_score, tier = self._evaluate_stagnation(window_metrics)
-        action_type = self._choose_action_type(tier)
-        idempotency_key = self.generate_idempotency_key(post_id, action_type, tier)
-
-        if self._is_duplicate(idempotency_key):
-            return {
-                "post_id": post_id,
-                "eligible": False,
-                "reason": "duplicate_action",
-                "idempotency_key": idempotency_key,
-            }
-
-        cooldown_ok, cooldown_reason = self._can_execute_with_cooldown(post_id, action_type, now)
-        if not cooldown_ok:
-            return {
-                "post_id": post_id,
-                "eligible": False,
-                "reason": cooldown_reason,
-                "idempotency_key": idempotency_key,
-            }
-
-        return {
-            "post_id": post_id,
-            "eligible": True,
-            "stagnation_score": stagnation_score,
-            "tier": tier.value,
-            "action_type": action_type.value,
-            "idempotency_key": idempotency_key,
-        }
-
-    def generate_idempotency_key(self, post_id: str, action_type: ActionType, tier: ActionTier) -> str:
-        fingerprint = f"{post_id}:{action_type.value}:{tier.value}".encode("utf-8")
-        digest = hashlib.sha256(fingerprint).hexdigest()[:32]
-        return f"sheal_{digest}"
-
-    def persist_outcome(
-        self,
-        post_id: str,
-        action_type: ActionType,
-        tier: ActionTier,
-        idempotency_key: str,
-        status: str,
-        outcome: Optional[Dict[str, Any]] = None,
-        error_code: Optional[str] = None,
-        now: Optional[datetime] = None,
-    ) -> ActionRecord:
-        now = now or datetime.now(timezone.utc)
-
-        existing = next((h for h in self._history if h.idempotency_key == idempotency_key), None)
-        if existing:
-            existing.status = status
-            existing.outcome = outcome
-            existing.error_code = error_code
-            existing.attempts += 1
-            existing.initiated_at = now
-            record = existing
-        else:
-            record = ActionRecord(
-                idempotency_key=idempotency_key,
-                post_id=post_id,
-                action_type=action_type,
-                tier=tier,
-                initiated_at=now,
-                status=status,
-                outcome=outcome,
-                error_code=error_code,
-            )
-            self._history.append(record)
-
-        self._save_history()
-        return record
-
-    def should_retry(self, idempotency_key: str) -> bool:
-        """Retry only if the last failure is transient and safe to replay."""
-        rec = next((h for h in self._history if h.idempotency_key == idempotency_key), None)
-        if not rec or rec.status != "failed":
-            return False
-
-        if rec.error_code not in SAFE_TRANSIENT_ERROR_CODES:
-            return False
-
-        return rec.action_type in {ActionType.COMMENT, ActionType.EDIT, ActionType.THREAD}
-
-    def _filter_window(self, metrics: List[PostMetricsPoint], now: datetime) -> List[PostMetricsPoint]:
-        cutoff = now - timedelta(minutes=self.config.per_post_window_minutes)
-        return [m for m in metrics if m.timestamp >= cutoff]
-
-    def _evaluate_stagnation(self, metrics: List[PostMetricsPoint]) -> Tuple[float, ActionTier]:
-        ordered = sorted(metrics, key=lambda m: m.timestamp)
-        first, last = ordered[0], ordered[-1]
-
-        imp_delta = max(0, last.impressions - first.impressions)
-        eng_delta = max(0, last.engagements - first.engagements)
-        eng_rate = eng_delta / imp_delta if imp_delta > 0 else 0.0
-
-        stagnation_score = 1.0 - min(1.0, eng_rate * 20)
-        if stagnation_score >= 0.8:
-            return stagnation_score, ActionTier.TIER_3
-        if stagnation_score >= 0.55:
-            return stagnation_score, ActionTier.TIER_2
-        return stagnation_score, ActionTier.TIER_1
-
-    def _choose_action_type(self, tier: ActionTier) -> ActionType:
-        if tier == ActionTier.TIER_1:
-            return ActionType.COMMENT
-        if tier == ActionTier.TIER_2:
-            return ActionType.EDIT
-        return ActionType.THREAD
-
-    def _is_duplicate(self, idempotency_key: str) -> bool:
-        return any(h.idempotency_key == idempotency_key and h.status in {"success", "running"} for h in self._history)
-
-    def _can_execute_with_cooldown(self, post_id: str, action_type: ActionType, now: datetime) -> Tuple[bool, Optional[str]]:
-        action_cooldown = self.config.cooldown_by_action_seconds[action_type]
-
-        same_post = [h for h in self._history if h.post_id == post_id]
-        recent_in_window = [
-            h for h in same_post
-            if h.initiated_at >= now - timedelta(minutes=self.config.per_post_window_minutes)
-        ]
-        if len(recent_in_window) >= self.config.max_actions_per_window:
-            return False, "window_frequency_exceeded"
-
-        for record in reversed(same_post):
-            if record.action_type != action_type:
-                continue
-            if (now - record.initiated_at).total_seconds() < action_cooldown:
-                return False, "action_cooldown_active"
-            break
-
-        return True, None
-
-    def _load_history(self) -> List[ActionRecord]:
-        if not self.persistence_path.exists():
-            return []
-        try:
-            payload = json.loads(self.persistence_path.read_text(encoding="utf-8"))
-            return [ActionRecord.from_json(item) for item in payload]
-        except (json.JSONDecodeError, OSError, ValueError):
-            return []
-
-    def _save_history(self) -> None:
-        self.persistence_path.parent.mkdir(parents=True, exist_ok=True)
-        payload = [item.to_json() for item in self._history]
-        self.persistence_path.write_text(json.dumps(payload, indent=2), encoding="utf-8")