feat(tools): operationalize dispatch router into triage loop

- Add tools/dispatch_router.py: audit-backed agent router based on
  April 2026 contribution audit (Timmy 95%, Gemini 70%, Perplexity 82%,
  Allegro 82.4%, Codex 100%, Claude 52%).
- Integrate into bin/timmy-orchestrator.sh: when AUTO_ASSIGN_UNASSIGNED=1,
  unassigned issues are now routed by category with logging.
- Log each dispatch decision (issue, agent, category, score, reason) to
  support future audit cycles and accuracy validation.

Closes #476

This operationalizes PR #465's dispatch router into the daily triage loop.

bin/timmy-orchestrator.sh

@@ -129,15 +129,35 @@ Preserved by timmy-orchestrator to prevent loss." 2>/dev/null &&           git p
   # Auto-assignment is opt-in because silent queue mutation resurrects old state.
   if [ "$unassigned_count" -gt 0 ]; then
     if [ "$AUTO_ASSIGN_UNASSIGNED" = "1" ]; then
-      log "Assigning $unassigned_count issues to claude..."
+      log "Routing $unassigned_count issues via dispatch_router..."
+      DISPATCHER="$SCRIPT_DIR/../tools/dispatch_router.py"
       while IFS= read -r line; do
         local repo=$(echo "$line" | sed 's/.*REPO=\([^ ]*\).*/\1/')
         local num=$(echo "$line" | sed 's/.*NUM=\([^ ]*\).*/\1/')
+        local title=$(echo "$line" | sed 's/.*TITLE=//')
+        # Get routing recommendation from dispatch_router
+        local agent="timmy" category="unknown" score="N/A" reason="unavailable"
+        if [ -x "$DISPATCHER" ]; then
+          local result
+          result=$(python3 "$DISPATCHER" "$title" 2>/dev/null)
+          if [ $? -eq 0 ]; then
+            agent=$(echo "$result" | python3 -c "import sys,json; print(json.load(sys.stdin).get('primary',{}).get('agent','timmy'))" 2>/dev/null)
+            category=$(echo "$result" | python3 -c "import sys,json; print(json.load(sys.stdin).get('category','unknown'))" 2>/dev/null)
+            local score_raw
+            score_raw=$(echo "$result" | python3 -c "import sys,json; print(json.load(sys.stdin).get('primary',{}).get('score','N/A'))" 2>/dev/null)
+            score=$(printf "%.2f" "$score_raw" 2>/dev/null || echo "$score_raw")
+            reason=$(echo "$result" | python3 -c "import sys,json; print(json.load(sys.stdin).get('primary',{}).get('reason',''))" 2>/dev/null)
+            [ -z "$agent" ] && agent="timmy"
+          fi
+        else
+          log "WARNING: dispatch_router.py not found or not executable at $DISPATCHER"
+        fi
+        # Apply assignment
         curl -sf -X PATCH "$GITEA_URL/api/v1/repos/$repo/issues/$num" \
           -H "Authorization: token $GITEA_TOKEN" \
           -H "Content-Type: application/json" \
-          -d '{"assignees":["claude"]}' >/dev/null 2>&1 && \
+          -d "{\"assignees\":[\"$agent\"]}" >/dev/null 2>&1 && \
-          log "  Assigned #$num ($repo) to claude"
+          log "  Routed #$num ($repo) to $agent | category=$category score=$score | $reason"
       done < "$state_dir/unassigned.txt"
     else
       log "Auto-assign disabled: leaving $unassigned_count unassigned issues untouched"

tools/dispatch_router.py

@@ -0,0 +1,224 @@
+#!/usr/bin/env python3
+"""
+tools/dispatch_router.py — Audit-backed agent dispatch router
+
+Routes incoming tasks to the best agent based on proven merge rates
+from the Perplexity contribution audit (April 2026).
+
+Usage:
+    python tools/dispatch_router.py "fix: broken CI pipeline"
+    python tools/dispatch_router.py "add feature architecture for user vault"
+"""
+
+import sys
+import json
+import re
+from pathlib import Path
+from dataclasses import dataclass, asdict
+from typing import List, Optional
+
+
+@dataclass
+class AgentRecommendation:
+    """Single agent recommendation with confidence score."""
+    agent: str
+    score: float
+    reviewer: str
+    reason: str
+    category: str
+
+
+@dataclass
+class DispatchResult:
+    """Complete routing result."""
+    task: str
+    category: str
+    primary: AgentRecommendation
+    candidates: List[AgentRecommendation]
+
+    def to_dict(self):
+        return {
+            "task": self.task,
+            "category": self.category,
+            "primary": asdict(self.primary),
+            "candidates": [asdict(c) for c in self.candidates],
+        }
+
+
+# ── Agent profiles ──────────────────────────────────────────────────────────
+# Score percentages are from the April 2026 audit (SOVEREIGN_AUDIT.md)
+AGENTS = {
+    "timmy": {
+        "categories": ["ci_cd", "infra", "security", "hotfix", "governance"],
+        "merge_rate": 0.95,
+        "reviewer": "allegro",
+        "notes": "Sovereign review, architecture, release judgment",
+    },
+    "gemini": {
+        "categories": ["feature_architecture", "research", "design"],
+        "merge_rate": 0.70,
+        "reviewer": "timmy",
+        "notes": "Frontier architecture, research-heavy prototypes",
+    },
+    "perplexity": {
+        "categories": ["code_review", "qa", "integration", "research_triage"],
+        "merge_rate": 0.82,
+        "reviewer": "timmy",
+        "notes": "Research triage, integration evaluation, architecture memos",
+    },
+    "allegro": {
+        "categories": ["spec_writing", "docs", "triage", "queue_hygiene"],
+        "merge_rate": 0.824,
+        "reviewer": "timmy",
+        "notes": "Tempo-and-dispatch, queue hygiene, operational next-move",
+    },
+    "codex": {
+        "categories": ["high_precision", "migration", "verification", "cleanup"],
+        "merge_rate": 1.00,
+        "reviewer": "timmy",
+        "notes": "Workflow hardening, cleanup, migration verification",
+    },
+    "claude": {
+        "categories": ["frontend_3d", "nexus", "refactor", "test_heavy"],
+        "merge_rate": 0.52,
+        "reviewer": "perplexity",
+        "notes": "Hard refactors, deep implementation, test-heavy changes",
+    },
+}
+
+
+# ── Keyword → category mapping ──────────────────────────────────────────────
+KEYWORD_MAP = {
+    # CI / CD / Infra / Security
+    "ci": "ci_cd",
+    "cd": "ci_cd",
+    "pipeline": "ci_cd",
+    "build": "ci_cd",
+    "deploy": "ci_cd",
+    "infra": "infra",
+    "infrastructure": "infra",
+    "server": "infra",
+    "vps": "infra",
+    "ssh": "infra",
+    "security": "security",
+    "vulnerability": "security",
+    "auth": "security",
+    "authentication": "security",
+    "hotfix": "hotfix",
+    "emergency": "hotfix",
+    "governance": "governance",
+    "policy": "governance",
+    # Architecture / Design
+    "architecture": "feature_architecture",
+    "architectural": "feature_architecture",
+    "design": "feature_architecture",
+    "prototype": "feature_architecture",
+    "research": "research",
+    # Code review / QA
+    "review": "code_review",
+    "qa": "qa",
+    "test": "qa",
+    "quality": "qa",
+    "bug": "qa",
+    "integration": "integration",
+    # Docs / Specs
+    "spec": "spec_writing",
+    "specification": "spec_writing",
+    "documentation": "docs",
+    "docs": "docs",
+    "readme": "docs",
+    "triage": "triage",
+    "dispatch": "triage",
+    # High-precision
+    "migration": "migration",
+    "verify": "verification",
+    "verification": "verification",
+    "cleanup": "cleanup",
+    "refactor": "refactor",
+    # 3D / Frontend
+    "nexus": "nexus",
+    "three": "frontend_3d",
+    "3d": "frontend_3d",
+    "frontend": "frontend_3d",
+    "ui": "frontend_3d",
+    # Misc
+    "agent": "high_precision",
+    "orchestrat": "high_precision",
+}
+
+
+def categorize(task: str) -> str:
+    """Categorize a task description into one of the known categories."""
+    task_lower = task.lower()
+    # Score keywords by presence
+    scores = {}
+    for keyword, category in KEYWORD_MAP.items():
+        if keyword in task_lower:
+            scores[category] = scores.get(category, 0) + 1
+    if scores:
+        return max(scores, key=scores.get)
+    # Default: treat as generic task → Timmy (sovereign decision-maker)
+    return "ci_cd"
+
+
+def agents_for_category(category: str) -> List[str]:
+    """Return agent names that handle the given category, ordered by preference."""
+    agents_in_category = []
+    for agent, profile in AGENTS.items():
+        if category in profile["categories"]:
+            agents_in_category.append(agent)
+    # If no agents match, default to timmy
+    return agents_in_category or ["timmy"]
+
+
+def rank_agents(task: str, category: str) -> List[AgentRecommendation]:
+    """Rank all agents by suitability for this task."""
+    candidates = []
+    for agent, profile in AGENTS.items():
+        score = profile["merge_rate"]
+        # Boost score if agent handles the category
+        if category in profile["categories"]:
+            score += 0.1  # category fit bonus
+        else:
+            score *= 0.5  # penalty for wrong category
+        candidates.append(
+            AgentRecommendation(
+                agent=agent,
+                score=round(score, 3),
+                reviewer=profile["reviewer"],
+                reason=profile["notes"],
+                category=category,
+            )
+        )
+    # Sort descending by score
+    candidates.sort(key=lambda x: x.score, reverse=True)
+    return candidates
+
+
+def dispatch(task: str) -> DispatchResult:
+    """Main entry: route a task description to the best agent."""
+    category = categorize(task)
+    ranked = rank_agents(task, category)
+    primary = ranked[0] if ranked else AgentRecommendation(
+        agent="timmy", score=0.5, reviewer="allegro",
+        reason="default fallback", category=category
+    )
+    return DispatchResult(
+        task=task,
+        category=category,
+        primary=primary,
+        candidates=ranked,
+    )
+
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: python tools/dispatch_router.py "<task description>"")
+        sys.exit(1)
+    task = " ".join(sys.argv[1:])
+    result = dispatch(task)
+    print(json.dumps(result.to_dict(), indent=2))
+
+
+if __name__ == "__main__":
+    main()