htmlgraph 0.9.3__py3-none-any.whl → 0.27.5__py3-none-any.whl

htmlgraph/cigs/messaging.py

@@ -0,0 +1,365 @@
+"""
+Imperative Message Generation for CIGS
+
+Generates escalating imperative messages for delegation enforcement.
+
+Reference: .htmlgraph/spikes/computational-imperative-guidance-system-design.md (Part 4)
+"""
+
+from .models import OperationClassification
+
+
+class ImperativeMessageGenerator:
+    """Generate imperative messages with escalation for delegation violations.
+
+    The generator supports 4 escalation levels:
+    - Level 0: Guidance (informative, no cost shown)
+    - Level 1: Imperative (commanding, includes cost)
+    - Level 2: Final Warning (urgent, includes consequences)
+    - Level 3: Circuit Breaker (blocking, requires acknowledgment)
+
+    Each level includes progressively more urgent messaging and additional context.
+    """
+
+    ESCALATION_LEVELS = {
+        0: {
+            "prefix": "💡 GUIDANCE",
+            "tone": "informative",
+            "includes_cost": False,
+            "includes_suggestion": True,
+            "includes_consequences": False,
+            "requires_acknowledgment": False,
+        },
+        1: {
+            "prefix": "🔴 IMPERATIVE",
+            "tone": "commanding",
+            "includes_cost": True,
+            "includes_suggestion": True,
+            "includes_consequences": False,
+            "requires_acknowledgment": False,
+        },
+        2: {
+            "prefix": "⚠️ FINAL WARNING",
+            "tone": "urgent",
+            "includes_cost": True,
+            "includes_suggestion": True,
+            "includes_consequences": True,
+            "requires_acknowledgment": False,
+        },
+        3: {
+            "prefix": "🚨 CIRCUIT BREAKER",
+            "tone": "blocking",
+            "includes_cost": True,
+            "includes_suggestion": True,
+            "includes_consequences": True,
+            "requires_acknowledgment": True,
+        },
+    }
+
+    # Tool-specific core messages
+    TOOL_MESSAGES = {
+        "Read": "YOU MUST delegate file reading to Explorer subagent",
+        "Grep": "YOU MUST delegate code search to Explorer subagent",
+        "Glob": "YOU MUST delegate file search to Explorer subagent",
+        "Edit": "YOU MUST delegate code changes to Coder subagent",
+        "Write": "YOU MUST delegate file writing to Coder subagent",
+        "NotebookEdit": "YOU MUST delegate notebook editing to Coder subagent",
+        "Bash": "YOU MUST delegate command execution to appropriate subagent",
+    }
+
+    # Category-specific rationales (WHY delegation is mandatory)
+    RATIONALES = {
+        "direct_exploration": (
+            "Exploration operations have unpredictable scope. "
+            "What looks like 'one Read' often becomes 3-5 reads. "
+            "Each direct read pollutes your strategic context with tactical details."
+        ),
+        "direct_implementation": (
+            "Implementation requires iteration (write → test → fix → test). "
+            "Delegating keeps your context focused on architecture, "
+            "while subagent handles the edit-test cycle."
+        ),
+        "exploration_sequence": (
+            "You have already executed multiple exploration operations. "
+            "This pattern indicates research work that should be delegated. "
+            "Subagent can explore comprehensively and return a summary."
+        ),
+        "direct_git": (
+            "Git operations cascade unpredictably (hooks, conflicts, push failures). "
+            "Copilot specializes in git AND costs 60% less than Task()."
+        ),
+        "direct_testing": (
+            "Test execution requires multiple iterations and debugging cycles. "
+            "Delegating test runs keeps test output out of your strategic context."
+        ),
+    }
+
+    def generate(
+        self,
+        tool: str,
+        classification: OperationClassification,
+        violation_count: int,
+        autonomy_level: str = "strict",
+    ) -> str:
+        """Generate imperative message based on context.
+
+        Args:
+            tool: Tool being used (Read, Grep, Edit, etc.)
+            classification: Operation classification with delegation details
+            violation_count: Number of violations in current session (0-3+)
+            autonomy_level: Current autonomy level ("strict", "guided", "observer")
+
+        Returns:
+            Formatted imperative message with appropriate escalation level
+        """
+        # Determine escalation level (0-3)
+        level = min(violation_count, 3)
+        config = self.ESCALATION_LEVELS[level]
+
+        message_parts = []
+
+        # 1. Prefix with escalation indicator + core message
+        core_msg = self._get_core_message(tool, classification)
+        message_parts.append(f"{config['prefix']}: {core_msg}")
+
+        # 2. WHY this is mandatory
+        rationale = self._get_rationale(classification)
+        message_parts.append(f"\n\n**WHY:** {rationale}")
+
+        # 3. Cost impact (if applicable for this level)
+        if config["includes_cost"]:
+            cost_msg = self._get_cost_message(classification)
+            message_parts.append(f"\n\n**COST IMPACT:** {cost_msg}")
+
+        # 4. What to do instead
+        if config["includes_suggestion"]:
+            message_parts.append(
+                f"\n\n**INSTEAD:** {classification.suggested_delegation}"
+            )
+
+        # 5. Consequences (for levels 2+)
+        if config["includes_consequences"]:
+            consequence = self._get_consequence_message(level)
+            message_parts.append(f"\n\n**CONSEQUENCE:** {consequence}")
+
+        # 6. Acknowledgment requirement (level 3 only)
+        if config["requires_acknowledgment"]:
+            message_parts.append(
+                "\n\n**REQUIRED:** Acknowledge this violation before proceeding:\n"
+                "`uv run htmlgraph orchestrator acknowledge-violation`\n\n"
+                "OR disable enforcement:\n"
+                "`uv run htmlgraph orchestrator set-level guidance`"
+            )
+
+        return "".join(message_parts)
+
+    def _get_core_message(
+        self, tool: str, classification: OperationClassification
+    ) -> str:
+        """Get core imperative message based on tool.
+
+        Args:
+            tool: Tool name
+            classification: Operation classification
+
+        Returns:
+            Core imperative message
+        """
+        return self.TOOL_MESSAGES.get(
+            tool, f"YOU MUST delegate {tool} operations to appropriate subagent"
+        )
+
+    def _get_rationale(self, classification: OperationClassification) -> str:
+        """Explain WHY delegation is mandatory.
+
+        Args:
+            classification: Operation classification
+
+        Returns:
+            Rationale explaining why delegation is required
+        """
+        # Use category-specific rationale or classification reason
+        rationale = self.RATIONALES.get(
+            classification.category, "Delegation preserves your strategic context."
+        )
+
+        # If classification provides specific reason, append it
+        if classification.reason and classification.reason not in rationale:
+            rationale = f"{rationale} {classification.reason}"
+
+        return rationale
+
+    def _get_cost_message(self, classification: OperationClassification) -> str:
+        """Generate cost impact message.
+
+        Args:
+            classification: Operation classification with cost estimates
+
+        Returns:
+            Formatted cost impact message
+        """
+        if classification.predicted_cost > 0 and classification.optimal_cost > 0:
+            savings_pct = classification.waste_percentage
+            return (
+                f"Direct execution costs ~{classification.predicted_cost:,} tokens in your context. "
+                f"Delegation would cost ~{classification.optimal_cost:,} tokens "
+                f"({savings_pct:.0f}% savings)."
+            )
+        else:
+            # Generic message if costs not available
+            return (
+                "Direct execution pollutes your context with tactical details. "
+                "Delegation isolates these details in subagent context."
+            )
+
+    def _get_consequence_message(self, level: int) -> str:
+        """Get consequence message for high escalation levels.
+
+        Args:
+            level: Escalation level (0-3)
+
+        Returns:
+            Consequence message appropriate for the level
+        """
+        if level == 2:
+            return (
+                "Next violation will trigger circuit breaker, "
+                "requiring manual acknowledgment before further operations."
+            )
+        elif level == 3:
+            return (
+                "Circuit breaker is now ACTIVE. "
+                "All delegation-required operations require explicit acknowledgment until reset."
+            )
+        return ""
+
+
+class PositiveReinforcementGenerator:
+    """Generate positive feedback for correct delegation patterns.
+
+    Provides encouraging messages when Claude correctly delegates operations,
+    reinforcing good behavior and showing concrete benefits.
+    """
+
+    ENCOURAGEMENTS = [
+        "Excellent delegation pattern!",
+        "Perfect use of subagent!",
+        "Context preserved effectively!",
+        "Optimal delegation choice!",
+        "Great strategic thinking!",
+        "Correct workflow applied!",
+    ]
+
+    def generate(
+        self,
+        tool: str,
+        cost_savings: int,
+        compliance_rate: float,
+        session_waste: int = 0,
+    ) -> str:
+        """Generate positive reinforcement message.
+
+        Args:
+            tool: Delegation tool used (Task, spawn_gemini, etc.)
+            cost_savings: Estimated tokens saved by delegating
+            compliance_rate: Current session compliance rate (0.0-1.0)
+            session_waste: Total waste tokens accumulated this session
+
+        Returns:
+            Formatted positive reinforcement message
+        """
+        import random
+
+        base = random.choice(self.ENCOURAGEMENTS)
+
+        message_parts = [f"✅ {base}"]
+
+        # Impact section
+        impact_lines = [
+            "\n\n**Impact:**",
+            f"- Saved ~{cost_savings:,} tokens of context",
+            "- Subagent handled tactical details",
+            "- Your strategic view remains clean",
+        ]
+        message_parts.append("\n".join(impact_lines))
+
+        # Session stats section
+        stats_lines = [
+            "\n\n**Session Stats:**",
+            f"- Delegation compliance: {compliance_rate:.0%}",
+        ]
+
+        if session_waste > 0:
+            stats_lines.append(
+                f"- Waste avoided: {session_waste:,} tokens saved by delegating"
+            )
+
+        if compliance_rate >= 0.9:
+            stats_lines.append(
+                "- Outstanding! Keep up the excellent delegation pattern."
+            )
+        elif compliance_rate >= 0.75:
+            stats_lines.append(
+                "- Keep it up! Consistent delegation improves response quality."
+            )
+        else:
+            stats_lines.append(
+                "- Good progress! Continue delegating to improve efficiency."
+            )
+
+        message_parts.append("\n".join(stats_lines))
+
+        return "".join(message_parts)
+
+    def generate_session_summary(
+        self,
+        total_delegations: int,
+        compliance_rate: float,
+        efficiency_score: float,
+        total_savings: int,
+    ) -> str:
+        """Generate end-of-session positive summary.
+
+        Args:
+            total_delegations: Number of successful delegations
+            compliance_rate: Session compliance rate (0.0-1.0)
+            efficiency_score: Cost efficiency score (0-100)
+            total_savings: Total tokens saved through delegation
+
+        Returns:
+            Session summary message
+        """
+        message_parts = ["## ✅ Delegation Session Summary\n"]
+
+        # Overall performance
+        if compliance_rate >= 0.9 and efficiency_score >= 80:
+            message_parts.append(
+                "**Outstanding Performance!** Excellent delegation discipline.\n"
+            )
+        elif compliance_rate >= 0.75:
+            message_parts.append(
+                "**Good Performance!** Strong delegation patterns observed.\n"
+            )
+        else:
+            message_parts.append(
+                "**Room for Improvement.** Continue working on delegation.\n"
+            )
+
+        # Metrics
+        metrics = [
+            "\n**Metrics:**",
+            f"- Total delegations: {total_delegations}",
+            f"- Compliance rate: {compliance_rate:.1%}",
+            f"- Efficiency score: {efficiency_score:.0f}/100",
+            f"- Tokens saved: {total_savings:,}",
+        ]
+        message_parts.append("\n".join(metrics))
+
+        # Encouragement
+        if compliance_rate < 0.75:
+            message_parts.append(
+                "\n\n**Next Session:** Focus on delegating exploration and implementation "
+                "tasks to specialized subagents."
+            )
+
+        return "\n".join(message_parts)