feat: bidirectional research dashboard + agent config fixes

- Integrate apaw_agent_model_research_v3.html as standalone dashboard - Add model-benchmarks.json with 32 agents, 11 scored models, 11 recommendations - Add build-research-dashboard.ts: inject live data into template → standalone HTML - Add rebuild-template.cjs: regenerate template from v3.html source - Add sync-benchmarks-from-yaml.cjs: sync YAML → JSON round-trip - Add sync-model-research.ts: apply recommendation matrix to config files - Add model-benchmarks.schema.json and model-research.schema.json for validation - Add bidirectional-data-flow.md architecture documentation - Add log-execution.cjs pipeline hook - Update capability-index.yaml: add fallback_models, failover_strategy - Update kilo-meta.json, kilo.jsonc, KILO_SPEC.md with synced models - Update evolution.md / research.md / self-evolution.md / evolutionary-sync.md docs - Fix security-auditor.md: quote YAML color (#DC2626) - Fix orchestrator.md: remove duplicate devops-engineer key - Build research-dashboard.html (106KB standalone) + dated archive
2026-04-29 21:04:22 +01:00
parent 2ae7789802
commit 3badb259cc
29 changed files with 13779 additions and 992 deletions
--- a/agent-evolution/README.md
+++ b/agent-evolution/README.md
@@ -117,6 +117,9 @@ bun run evolution:run     # Запустить контейнер
 bun run evolution:stop    # Остановить
 bun run evolution:dev      # Docker Compose
 bun run evolution:logs     # Логи
+bun run research:dashboard    # Build research dashboard
+bun run research:watch        # Watch mode for dashboard
+bun run research:sync         # Sync model research to agents
 ```

 ## Структура
@@ -132,6 +135,50 @@ agent-evolution/
 └── README.md                    # Этот файл
 ```

+## Research Dashboard (Model Benchmarks)
+
+### Generate from live data
+
+```bash
+# Build research dashboard from model-benchmarks.json
+bun run agent-evolution/scripts/build-research-dashboard.ts
+
+# Watch mode — auto-rebuild on data changes
+bun run agent-evolution/scripts/build-research-dashboard.ts --watch
+
+# Open in browser
+start agent-evolution/research-dashboard.html
+```
+
+### Output files
+
+| File | Description |
+|------|-------------|
+| `research-dashboard.html` | Latest interactive dashboard (all 6 tabs) |
+| `dist/research-dashboard-YYYY_MM_DD.html` | Dated archive |
+| `research-dashboard.template.html` | Template for generation |
+
+### Dashboard tabs
+
+1. **Обзор** — stat cards, current config table, agent count, model count
+2. **Groq** — free tier models with RPM/RPD/TPM/TPD limits, speed indicators
+3. **Модели** — filterable cards with SWE-bench, IF scores, context windows, tags
+4. **Матрица** — Agent×Model heatmap with IF adjustment, tooltips, color coding
+5. **Рекомендации** — selectable cards with JSON export, impact analysis
+6. **Анализ профита** — before/after comparison, canvas charts, closed-source comparison
+
+### Source data
+
+The dashboard reads from `agent-evolution/data/model-benchmarks.json`:
+- 15 models with benchmarks (SWE-bench, IF scores)
+- 36 agent configurations
+- 33 agent×model score matrices
+- 11 recommendations
+- 5 Groq models with rate limits
+- Closed-source comparison data
+
+Refresh: run `/research models` or `/evolution research` to update
+
 ## Быстрый старт

 ```bash
@@ -231,6 +278,22 @@ git log --all --oneline -- ".kilo/agents/"
 **Files**: src/auth.ts, src/user.ts
 ```

+### 6. Model Benchmarks (agent-evolution/data/model-benchmarks.json)
+
+Research data extracted from `apaw_agent_model_research_v3.html`:
+- Static benchmark scores (SWE-bench, IF scores, context windows)
+- Heatmap compatibility matrix
+- Provider rate limits
+- Recommendation history
+
+### 7. Model Research Output (agent-evolution/data/model-research-latest.json)
+
+Dynamic research results:
+- Fresh model data from provider APIs
+- IF-adjusted agent×model scores
+- Pending recommendations with impact levels
+- Ready-to-apply YAML patches
+
 ## JSON Schema

 Формат `agent-versions.json`:
@@ -271,6 +334,76 @@ git log --all --oneline -- ".kilo/agents/"
 }
 ```

+## Model Research Data
+
+### model-benchmarks.json
+
+Comprehensive benchmark data from the HTML research file:
+
+```json
+{
+  "version": "1.0.0",
+  "generated": "2026-04-27T17:44:44Z",
+  "total_agents": 36,
+  "total_models_tracked": 11,
+  "models": [
+    {
+      "id": "ollama-cloud/qwen3-coder:480b",
+      "name": "Qwen3-Coder 480B",
+      "organization": "Qwen",
+      "swe_bench": 66.5,
+      "if_score": 88,
+      "context_window": "256K→1M",
+      "categories": ["coding", "agent"],
+      "provider": "ollama"
+    }
+  ],
+  "agent_current_config": [
+    { "agent": "lead-developer", "model": "ollama-cloud/qwen3-coder:480b", "fit_score": 92, "status": "optimal" }
+  ],
+  "recommendations": [
+    {
+      "agent": "planner",
+      "current_model": "nemotron-3-super",
+      "recommended_model": "deepseek-v4-pro-max",
+      "impact": "high",
+      "expected_improvement": { "quality": "+10%", "speed": "~1x", "context_window": "1M" }
+    }
+  ]
+}
+```
+
+### model-research-latest.json
+
+Latest research output (overwritten each cycle):
+- Generated by `/research models` or `/evolution Step 0`
+- Validated against `model-research.schema.json`
+- Consumed by `sync-model-research.ts`
+
+### sync-model-research.ts
+
+Applies model recommendations to configuration:
+
+```bash
+# Dry-run first
+bun run agent-evolution/scripts/sync-model-research.ts --dry-run
+
+# Apply all pending recommendations
+bun run agent-evolution/scripts/sync-model-research.ts
+
+# Apply for single agent
+bun run agent-evolution/scripts/sync-model-research.ts --agent planner
+```
+
+Updates:
+1. `.kilo/capability-index.yaml` — model assignments
+2. `kilo-meta.json` — source of truth
+3. `kilo.jsonc` — agent config
+4. `agent-evolution/data/agent-versions.json` — history tracking
+5. `.kilo/agents/*.md` frontmatter (via sync-agents.js --fix)
+
+After applying, rebuilds dashboard automatically.
+
 ## Интеграция

 ### В Pipeline
@@ -406,4 +539,50 @@ cp agent-evolution/data/backup/agent-versions-20260405.json agent-evolution/data
 4. **Integration**:
   - Slack/Telegram уведомления
   - Автоматическое применение рекомендаций
-   - A/B testing моделей
+   - A/B testing моделей
+
+## Bidirectional Data Flow
+
+```
+[/research models] OR [/evolution Step 0]
+       ↓
+[agent-evolution/data/model-research-latest.json]
+       ↓
+[bun run sync-model-research.ts]
+       ↓
+[.kilo/capability-index.yaml] → updated model assignments
+[kilo-meta.json]              → updated source of truth
+[kilo.jsonc]                  → updated config
+[agent-versions.json]         → history entries
+[.kilo/agents/*.md]           → frontmatter updated
+       ↓
+[sync-agents.js --fix]        → propagate to all files
+       ↓
+[bun run build-research-dashboard.ts]
+       ↓
+[research-dashboard.html]     → live dashboard
+[dist/dashboard-YYYY_MM_DD.html] → dated archive
+       ↓
+[/research models]            ← loop continues
+```
+
+### Data staleness check
+
+```bash
+# Check if benchmarks need refresh
+node -e "
+const d = require('./agent-evolution/data/model-benchmarks.json');
+const days = (Date.now() - new Date(d.generated)) / (1000*60*60*24);
+console.log(days > 7 ? 'STALE: needs refresh' : 'FRESH', Math.round(days), 'days old');
+"
+```
+
+### Auto-refresh pipeline
+
+```yaml
+# In capability-index.yaml
+evolution:
+  auto_trigger: true
+  max_evolution_attempts: 3
+  dashboard_rebuild: true  # new: auto-rebuild on model changes
+```
--- a/agent-evolution/data/agent-versions.json
+++ b/agent-evolution/data/agent-versions.json
@@ -1,12 +1,12 @@
 {
  "version": "1.0.0",
-  "lastUpdated": "2026-04-23T06:24:32.543Z",
+  "lastUpdated": "2026-04-27T20:28:58.592Z",
  "agents": {
    "lead-developer": {
      "current": {
        "description": "Primary code writer for backend and core logic. Writes implementation to pass tests",
        "mode": "subagent",
-        "model": "ollama-cloud/qwen3-coder:480b",
+        "model": "ollama-cloud/nemotron-3-super",
        "provider": "Ollama",
        "variant": "thinking",
        "color": "\"#DC2626\"",
@@ -27,6 +27,24 @@
          "to": "ollama-cloud/qwen3-coder:480b",
          "reason": "Initial configuration from capability-index.yaml",
          "source": "git"
+        },
+        {
+          "date": "2026-04-27T16:56:09.013Z",
+          "commit": "model-research-sync",
+          "type": "model_change",
+          "from": "ollama-cloud/qwen3-coder:480b",
+          "to": "ollama-cloud/nemotron-3-super",
+          "reason": "Nemotron 3 Super has better reasoning for core development tasks and RULER@1M context window. SWE-bench 68% vs Qwen's 66.5%.",
+          "source": "research"
+        },
+        {
+          "date": "2026-04-27T20:28:58.592Z",
+          "commit": "model-research-sync",
+          "type": "model_change",
+          "from": "ollama-cloud/qwen3-coder:480b",
+          "to": "ollama-cloud/nemotron-3-super",
+          "reason": "Nemotron 3 Super has better reasoning for core development tasks and RULER@1M context window. SWE-bench 68% vs Qwen's 66.5%.",
+          "source": "research"
        }
      ],
      "performance_log": []
@@ -255,7 +273,7 @@
      "current": {
        "description": "Designs technical specifications, data schemas, and API contracts before implementation",
        "mode": "subagent",
-        "model": "ollama-cloud/glm-5.1",
+        "model": "ollama-cloud/nemotron-3-super",
        "provider": "Ollama",
        "variant": "thinking",
        "color": "\"#0891B2\"",
@@ -285,6 +303,15 @@
          "to": "ollama-cloud/glm-5.1",
          "reason": "Model update from sync",
          "source": "git"
+        },
+        {
+          "date": "2026-04-27T16:59:52.825Z",
+          "commit": "model-research-sync",
+          "type": "model_change",
+          "from": "ollama-cloud/glm-5.1",
+          "to": "ollama-cloud/nemotron-3-super",
+          "reason": "Test recommendation for model research sync script",
+          "source": "research"
        }
      ],
      "performance_log": []
--- a/agent-evolution/data/model-benchmarks.json
+++ b/agent-evolution/data/model-benchmarks.json
--- a/agent-evolution/data/model-benchmarks.schema.json
+++ b/agent-evolution/data/model-benchmarks.schema.json
@@ -0,0 +1,553 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$id": "https://app.kilo.ai/model-benchmarks.schema.json",
+  "title": "APAW Model Benchmarks Data",
+  "description": "Schema for static model benchmarks extracted from HTML sources",
+  "type": "object",
+  "required": [
+    "version",
+    "generated",
+    "source",
+    "metadata",
+    "models",
+    "groq_models",
+    "agent_model_scores",
+    "if_scores",
+    "agent_current_config",
+    "recommendations",
+    "impact_data",
+    "benchmark_comparison"
+  ],
+  "properties": {
+    "version": {
+      "type": "string",
+      "const": "1.0.0"
+    },
+    "generated": {
+      "type": "string",
+      "format": "date-time"
+    },
+    "source": {
+      "type": "string",
+      "description": "Source of benchmark data (e.g. HTML scraping, API, manual entry)"
+    },
+    "metadata": {
+      "type": "object",
+      "properties": {
+        "scrape_date": {
+          "type": "string",
+          "format": "date-time"
+        },
+        "source_urls": {
+          "type": "array",
+          "items": {
+            "type": "string"
+          }
+        },
+        "notes": {
+          "type": "string"
+        },
+        "data_quality": {
+          "type": "string",
+          "enum": [
+            "high",
+            "medium",
+            "low",
+            "estimated"
+          ]
+        }
+      }
+    },
+    "models": {
+      "type": "array",
+      "description": "All benchmarked models from various providers",
+      "items": {
+        "type": "object",
+        "required": [
+          "id",
+          "name",
+          "provider",
+          "category"
+        ],
+        "properties": {
+          "id": {
+            "type": "string",
+            "description": "Model identifier"
+          },
+          "name": {
+            "type": "string"
+          },
+          "organization": {
+            "type": "string"
+          },
+          "provider": {
+            "type": "string",
+            "enum": [
+              "ollama",
+              "ollama-cloud",
+              "openrouter",
+              "groq",
+              "anthropic",
+              "openai",
+              "meta",
+              "cohere",
+              "google",
+              "microsoft",
+              "unknown"
+            ]
+          },
+          "category": {
+            "type": "string",
+            "enum": [
+              "big",
+              "medium",
+              "small",
+              "coder",
+              "reasoning",
+              "creative"
+            ]
+          },
+          "parameters": {
+            "type": "string"
+          },
+          "benchmarks": {
+            "type": "object",
+            "properties": {
+              "swe_bench": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              },
+              "swe_bench_pro": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              },
+              "terminal_bench": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              },
+              "live_codebench": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              },
+              "gpqa": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              },
+              "hle": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              },
+              "browse_comp": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              },
+              "m_mlu": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              },
+              "m_mlu_pro": {
+                "type": [
+                  "number",
+                  "null"
+                ]
+              }
+            }
+          },
+          "description": {
+            "type": "string"
+          },
+          "availability": {
+            "type": "object",
+            "properties": {
+              "rpm": {
+                "type": [
+                  "integer",
+                  "null"
+                ]
+              },
+              "rpd": {
+                "type": [
+                  "integer",
+                  "string",
+                  "null"
+                ]
+              },
+              "tpm": {
+                "type": [
+                  "integer",
+                  "string",
+                  "null"
+                ]
+              },
+              "tpd": {
+                "type": [
+                  "integer",
+                  "string",
+                  "null"
+                ]
+              }
+            }
+          },
+          "free": {
+            "type": "boolean"
+          },
+          "cost_per_1m_input": {
+            "type": [
+              "number",
+              "string",
+              "null"
+            ]
+          },
+          "tier": {
+            "type": "string",
+            "enum": [
+              "free",
+              "trial",
+              "paid",
+              "enterprise"
+            ]
+          }
+        }
+      }
+    },
+    "groq_models": {
+      "type": "array",
+      "description": "Groq-specific models with performance data",
+      "items": {
+        "type": "object",
+        "required": [
+          "id",
+          "name",
+          "speed_tps",
+          "provider"
+        ],
+        "properties": {
+          "id": {
+            "type": "string"
+          },
+          "name": {
+            "type": "string"
+          },
+          "speed_tps": {
+            "type": [
+              "number",
+              "string"
+            ]
+          },
+          "provider": {
+            "type": "string",
+            "const": "groq"
+          },
+          "benchmarks": {
+            "type": "object"
+          },
+          "availability": {
+            "type": "object"
+          }
+        }
+      }
+    },
+    "agent_model_scores": {
+      "type": "array",
+      "description": "Agent × Model compatibility scoring matrices",
+      "items": {
+        "type": "object",
+        "required": [
+          "agent",
+          "model_id",
+          "score",
+          "category"
+        ],
+        "properties": {
+          "agent": {
+            "type": "string"
+          },
+          "model_id": {
+            "type": "string"
+          },
+          "score": {
+            "type": "number",
+            "minimum": 0,
+            "maximum": 100
+          },
+          "category": {
+            "type": "string",
+            "enum": [
+              "performance",
+              "instruction_following",
+              "creativity",
+              "code_generation"
+            ]
+          },
+          "reason": {
+            "type": "string"
+          },
+          "timestamp": {
+            "type": "string",
+            "format": "date-time"
+          },
+          "current_model_id": {
+            "type": "string",
+            "description": "Current model ID string (replaces index)"
+          }
+        }
+      }
+    },
+    "if_scores": {
+      "type": "object",
+      "description": "Instruction Following scores mapping",
+      "additionalProperties": {
+        "type": "number",
+        "minimum": 0,
+        "maximum": 100
+      }
+    },
+    "agent_current_config": {
+      "type": "array",
+      "description": "Current agent model configurations",
+      "items": {
+        "type": "object",
+        "required": [
+          "agent",
+          "model",
+          "provider",
+          "status"
+        ],
+        "properties": {
+          "agent": {
+            "type": "string"
+          },
+          "model": {
+            "type": "string"
+          },
+          "provider": {
+            "type": "string"
+          },
+          "status": {
+            "type": "string",
+            "enum": [
+              "active",
+              "testing",
+              "deprecated",
+              "pending"
+            ]
+          },
+          "reasoning_effort": {
+            "type": "string",
+            "enum": [
+              "L",
+              "M",
+              "H"
+            ]
+          },
+          "fit_score": {
+            "type": "number"
+          },
+          "date_applied": {
+            "type": "string",
+            "format": "date-time"
+          }
+        }
+      }
+    },
+    "recommendations": {
+      "type": "array",
+      "description": "Model change recommendations based on benchmarks",
+      "items": {
+        "type": "object",
+        "required": [
+          "agent",
+          "action",
+          "current_model",
+          "recommended_model",
+          "impact"
+        ],
+        "properties": {
+          "agent": {
+            "type": "string"
+          },
+          "action": {
+            "type": "string",
+            "enum": [
+              "update_model",
+              "confirm_model",
+              "add_fallback",
+              "redesign_agent"
+            ]
+          },
+          "current_model": {
+            "type": "string"
+          },
+          "current_provider": {
+            "type": "string"
+          },
+          "recommended_model": {
+            "type": "string"
+          },
+          "recommended_provider": {
+            "type": "string"
+          },
+          "impact": {
+            "type": "string",
+            "enum": [
+              "critical",
+              "high",
+              "medium",
+              "low"
+            ]
+          },
+          "rationale": {
+            "type": "string"
+          },
+          "expected_improvement": {
+            "type": "object"
+          },
+          "applied": {
+            "type": "boolean"
+          }
+        }
+      }
+    },
+    "impact_data": {
+      "type": "array",
+      "description": "Impact analysis of model changes",
+      "items": {
+        "type": "object",
+        "required": [
+          "agent",
+          "model_change",
+          "impact_score"
+        ],
+        "properties": {
+          "agent": {
+            "type": "string"
+          },
+          "model_change": {
+            "type": "string"
+          },
+          "impact_score": {
+            "type": "number",
+            "minimum": 0,
+            "maximum": 100,
+            "description": "Impact score 0-100"
+          }
+        }
+      }
+    },
+    "benchmark_comparison": {
+      "type": "object",
+      "description": "APAW vs closed-source benchmark comparison",
+      "properties": {
+        "benchmarks": {
+          "type": "array",
+          "description": "Benchmark names used for comparison",
+          "items": {
+            "type": "string"
+          }
+        },
+        "closed_source_models": {
+          "type": "array",
+          "description": "Closed-source models included in comparison",
+          "items": {
+            "type": "object",
+            "properties": {
+              "name": {
+                "type": "string"
+              },
+              "provider": {
+                "type": "string"
+              },
+              "benchmarks": {
+                "type": "object"
+              }
+            }
+          }
+        },
+        "apaw_models": {
+          "type": "array",
+          "description": "APAW pipeline models included in comparison",
+          "items": {
+            "type": "object",
+            "properties": {
+              "name": {
+                "type": "string"
+              },
+              "provider": {
+                "type": "string"
+              },
+              "benchmarks": {
+                "type": "object"
+              }
+            }
+          }
+        },
+        "apaw_best": {
+          "type": "object",
+          "description": "Best APAW model per benchmark",
+          "additionalProperties": {
+            "type": "object",
+            "properties": {
+              "model": {
+                "type": "string"
+              },
+              "score": {
+                "type": "number"
+              },
+              "gap_to_closed": {
+                "type": [
+                  "number",
+                  "string"
+                ]
+              }
+            }
+          }
+        },
+        "closed_best": {
+          "type": "object",
+          "description": "Best closed-source model per benchmark",
+          "additionalProperties": {
+            "type": "object",
+            "properties": {
+              "model": {
+                "type": "string"
+              },
+              "score": {
+                "type": "number"
+              }
+            }
+          }
+        },
+        "summary": {
+          "type": "object",
+          "properties": {
+            "apaw_avg_score": {
+              "type": "number"
+            },
+            "closed_avg_score": {
+              "type": "number"
+            },
+            "coverage_gap": {
+              "type": "string"
+            }
+          }
+        }
+      }
+    }
+  }
+}
--- a/agent-evolution/data/model-research-latest.json
+++ b/agent-evolution/data/model-research-latest.json
@@ -0,0 +1,59 @@
+{
+  "version": "1.0.0",
+  "generated": "2026-04-27T17:51:36.000Z",
+  "source": "/research model-optimization",
+  "models": [],
+  "recommendations": [
+    {
+      "agent": "lead-developer",
+      "action": "update_model",
+      "current_model": "ollama-cloud/qwen3-coder:480b",
+      "current_provider": "ollama-cloud",
+      "recommended_model": "ollama-cloud/nemotron-3-super",
+      "recommended_provider": "ollama-cloud",
+      "impact": "high",
+      "expected_improvement": {
+        "quality": "+15%",
+        "speed": "+20%",
+        "context_window": "1M→1M"
+      },
+      "score_before": 85,
+      "score_after": 92,
+      "score_delta": 7,
+      "rationale": "Nemotron 3 Super has better reasoning for core development tasks and RULER@1M context window. SWE-bench 68% vs Qwen's 66.5%.",
+      "applied": false,
+      "applied_date": null
+    },
+    {
+      "agent": "devops-engineer",
+      "action": "confirm_model",
+      "current_model": "ollama-cloud/nemotron-3-super",
+      "current_provider": "ollama-cloud",
+      "recommended_model": "ollama-cloud/nemotron-3-super",
+      "recommended_provider": "ollama-cloud",
+      "impact": "low",
+      "expected_improvement": {
+        "quality": "0%",
+        "speed": "0%",
+        "context_window": "1M→1M"
+      },
+      "score_before": 88,
+      "score_after": 88,
+      "score_delta": 0,
+      "rationale": "Current model already optimal for DevOps tasks. Nemotron 3 Super's RULER@1M is critical for parsing complex Docker/Compose configs.",
+      "applied": false,
+      "applied_date": null
+    }
+  ],
+  "heatmap": {},
+  "closed_source_comparison": {},
+  "capability_index_patch": [],
+  "summary": {
+    "avg_quality_improvement": "+7.5%",
+    "providers_used": ["ollama-cloud"],
+    "key_models": ["nemotron-3-super"],
+    "total_recommendations": 2,
+    "applied_count": 0,
+    "pending_count": 2
+  }
+}
--- a/agent-evolution/data/model-research.schema.json
+++ b/agent-evolution/data/model-research.schema.json
@@ -0,0 +1,331 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$id": "https://app.kilo.ai/model-research.schema.json",
+  "title": "APAW Model Research Output",
+  "description": "Schema for automated model research and recommendation output",
+  "type": "object",
+  "required": ["version", "generated", "source", "models", "recommendations", "heatmap"],
+  "properties": {
+    "version": {
+      "type": "string",
+      "const": "1.0.0"
+    },
+    "generated": {
+      "type": "string",
+      "format": "date-time"
+    },
+    "source": {
+      "type": "string",
+      "description": "What triggered this research (e.g. /evolution, /research, manual)"
+    },
+    "trigger": {
+      "type": "object",
+      "properties": {
+        "type": {
+          "type": "string",
+          "enum": ["evolution_cycle", "manual_research", "fitness_below_threshold", "scheduled"]
+        },
+        "issue": {
+          "type": "integer"
+        },
+        "fitness_score": {
+          "type": "number"
+        },
+        "reason": {
+          "type": "string"
+        }
+      }
+    },
+    "models": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "required": ["id", "name", "organization", "if_score", "provider"],
+        "properties": {
+          "id": {
+            "type": "string",
+            "description": "Full model ID like ollama-cloud/qwen3-coder:480b"
+          },
+          "name": {
+            "type": "string"
+          },
+          "organization": {
+            "type": "string"
+          },
+          "parameters": {
+            "type": "string"
+          },
+          "context_window": {
+            "type": "string"
+          },
+          "swe_bench": {
+            "type": ["number", "null"]
+          },
+          "swe_bench_pro": {
+            "type": ["number", "null"]
+          },
+          "terminal_bench": {
+            "type": ["number", "null"]
+          },
+          "live_codebench": {
+            "type": ["number", "null"]
+          },
+          "gpqa": {
+            "type": ["number", "null"]
+          },
+          "hle": {
+            "type": ["number", "null"]
+          },
+          "browse_comp": {
+            "type": ["number", "null"]
+          },
+  "if_score": {
+    "type": "number",
+    "minimum": 0,
+    "maximum": 100,
+    "description": "Instruction Following composite score (IFEval + IFBench)"
+  },
+          "categories": {
+            "type": "array",
+            "items": {
+              "type": "string"
+            }
+          },
+          "tags": {
+            "type": "array",
+            "items": {
+              "type": "string"
+            }
+          },
+          "provider": {
+            "type": "string",
+            "enum": ["ollama", "ollama-cloud", "openrouter", "groq", "hybrid"]
+          },
+          "free": {
+            "type": "boolean"
+          },
+          "cost_per_1m_input": {
+            "type": ["number", "string", "null"]
+          },
+          "description": {
+            "type": "string"
+          },
+          "availability": {
+            "type": "object",
+            "properties": {
+              "rpm": {
+                "type": ["integer", "null"]
+              },
+              "rpd": {
+                "type": ["integer", "string", "null"]
+              },
+              "tpm": {
+                "type": ["integer", "string", "null"]
+              },
+              "tpd": {
+                "type": ["integer", "string", "null"]
+              }
+            }
+          },
+          "speed_tps": {
+            "type": ["number", "string", "null"]
+          }
+        }
+      }
+    },
+    "recommendations": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "required": ["agent", "action", "current_model", "recommended_model", "impact", "rationale"],
+        "properties": {
+          "agent": {
+            "type": "string"
+          },
+          "action": {
+            "type": "string",
+            "enum": ["update_model", "confirm_model", "add_fallback", "redesign_agent"]
+          },
+          "current_model": {
+            "type": "string"
+          },
+          "current_provider": {
+            "type": "string"
+          },
+          "recommended_model": {
+            "type": "string"
+          },
+          "recommended_provider": {
+            "type": "string"
+          },
+          "fallback_model": {
+            "type": "string"
+          },
+          "fallback_strategy": {
+            "type": "string"
+          },
+          "impact": {
+            "type": "string",
+            "enum": ["critical", "high", "medium", "low"]
+          },
+          "expected_improvement": {
+            "type": "object",
+            "properties": {
+              "quality": {
+                "type": "string"
+              },
+              "speed": {
+                "type": "string"
+              },
+              "context_window": {
+                "type": "string"
+              }
+            }
+          },
+          "score_before": {
+            "type": "number"
+          },
+          "score_after": {
+            "type": "number"
+          },
+          "score_delta": {
+            "type": "number"
+          },
+          "rationale": {
+            "type": "string"
+          },
+          "applied": {
+            "type": "boolean",
+            "default": false
+          },
+          "applied_date": {
+            "type": ["string", "null"],
+            "format": "date-time"
+          }
+        }
+      }
+    },
+    "heatmap": {
+      "type": "object",
+      "description": "Agent × Model compatibility matrix with IF adjustment",
+      "required": ["models", "agents"],
+      "properties": {
+        "models": {
+          "type": "array",
+          "items": {
+            "type": "object",
+            "required": ["id", "if_score"],
+            "properties": {
+              "id": {
+                "type": "string"
+              },
+              "display_name": {
+                "type": "string"
+              },
+              "provider": {
+                "type": "string"
+              },
+              "if_score": {
+                "type": "number"
+              }
+            }
+          }
+        },
+        "agents": {
+          "type": "array",
+          "items": {
+            "type": "object",
+            "required": ["agent", "reasoning_effort", "scores"],
+            "properties": {
+              "agent": {
+                "type": "string"
+              },
+              "current_model": {
+                "type": "string"
+              },
+              "reasoning_effort": {
+                "type": "string",
+                "enum": ["L", "M", "H"]
+              },
+              "scores": {
+                "type": "object",
+                "additionalProperties": {
+                  "type": "number"
+                },
+                "description": "Model ID → compatibility score (0-100, IF-adjusted)"
+              }
+            }
+          }
+        },
+        "if_adjustment_formula": {
+          "type": "string",
+          "default": "score * (0.7 + 0.3 * IF/100)"
+        }
+      }
+    },
+    "closed_source_comparison": {
+      "type": "object",
+      "description": "APAW pipeline models vs top closed-source models",
+      "properties": {
+        "benchmarks": {
+          "type": "array"
+        },
+        "models": {
+          "type": "array"
+        },
+        "apaw_best_per_benchmark": {
+          "type": "object"
+        },
+        "closed_best_per_benchmark": {
+          "type": "object"
+        }
+      }
+    },
+    "capability_index_patch": {
+      "type": "array",
+      "description": "Ready-to-apply patches to capability-index.yaml",
+      "items": {
+        "type": "object",
+        "required": ["agent", "set"],
+        "properties": {
+          "agent": {
+            "type": "string"
+          },
+          "set": {
+            "type": "object",
+            "additionalProperties": true
+          }
+        }
+      }
+    },
+    "summary": {
+      "type": "object",
+      "properties": {
+        "avg_quality_improvement": {
+          "type": "string"
+        },
+        "providers_used": {
+          "type": "array",
+          "items": {
+            "type": "string"
+          }
+        },
+        "key_models": {
+          "type": "array",
+          "items": {
+            "type": "string"
+          }
+        },
+        "total_recommendations": {
+          "type": "integer"
+        },
+        "applied_count": {
+          "type": "integer"
+        },
+        "pending_count": {
+          "type": "integer"
+        }
+      }
+    }
+  }
+}
--- a/agent-evolution/dist/research-dashboard-2026_04_29.html
+++ b/agent-evolution/dist/research-dashboard-2026_04_29.html
--- a/agent-evolution/docs/bidirectional-data-flow.md
+++ b/agent-evolution/docs/bidirectional-data-flow.md
@@ -0,0 +1,504 @@
+# Двунаправленный поток данных APAW Agent Model Research
+
+Этот документ описывает архитектуру системы, которая автоматизирует исследование моделей AI для агентов APAW и синхронизирует данные между визуальной панелью, конфигурационными файлами и пайплайном эволюции.
+
+## Цель
+
+Изначально все данные исследования моделей были захардкожены в HTML-файле `apaw_agent_model_research_v3.html` (1168 строк JavaScript). Двунаправленный поток делает эту систему:
+
+- **Машиночитаемой** — данные хранятся в JSON для автоматической обработки
+- **Записываемой** — изменения в конфигурации агентов обновляют JSON и перегенерируют дашборд
+- **Визуализированной** — любое изменение данных автоматически создаёт новый HTML
+
+## Архитектура данных
+
+### Файлы системы
+
+| Файл | Назначение | Формат | Обновляется |
+|------|-----------|--------|-------------|
+| `data/model-benchmarks.json` | Статические бенчмарки | JSON | `/research models`, вручную |
+| `data/model-research-latest.json` | Последнее исследование | JSON | `/evolution Step 0`, `/research models` |
+| `data/model-research.schema.json` | Схема валидации | JSON Schema | Вручную |
+| `data/model-benchmarks.schema.json` | Схема бенчмарков | JSON Schema | Вручную |
+| `scripts/build-research-dashboard.ts` | Генерация HTML | TypeScript/Bun | Вручную |
+| `scripts/sync-model-research.ts` | Применение изменений | TypeScript/Bun | Вручную |
+| `research-dashboard.template.html` | Шаблон дашборда | HTML+JS+CSS | Вручную |
+| `research-dashboard.html` | Готовый дашборд | HTML (standalone) | `build-research-dashboard.ts` |
+| `dist/research-dashboard-YYYY_MM_DD.html` | Архив | HTML | `build-research-dashboard.ts` |
+
+## Поток данных
+
+### Направление 1: HTML → JSON (Исследование → Бенчмарки)
+
+Источник: `apaw_agent_model_research_v3.html` (вручную исследованные данные)
+
+```
+apaw_agent_model_research_v3.html
+       │ hardcoded JS arrays:
+       │   cfg[]     — текущие конфиги агентов
+       │   ollamaModels[] — характеристики моделей
+       │   hmAgents[] — матрица очков
+       │   recs[] — рекомендации
+       │   impactData[] — дельта изменений
+       │   groqModels[] — лимиты Groq
+       ↓
+agent-evolution/data/model-benchmarks.json
+       ├─ models[] — 15 моделей, бенчмарки, IF-оценки
+       ├─ agent_model_scores[] — 33 агента × 11 моделей
+       ├─ agent_current_config[] — 36 текущих назначений
+       ├─ recommendations[] — 11 рекомендуемых замен
+       ├─ groq_models[] — 5 моделей Groq с лимитами
+       ├─ impact_data[] — before/after
+       └─ benchmark_comparison — сравнение с закрытыми моделями
+```
+
+**Как обновлять**: один раз данные извлечены из HTML. Дальнейшие обновления:
+- Автоматически: `/research models` → `model-research-latest.json` → `model-benchmarks.json`
+- Вручную: редактировать `model-benchmarks.json`, обновить `metadata.generated`
+
+### Направление 2: JSON → Конфиг → HTML (Применение → Визуализация)
+
+```
+[/research models] OR [/evolution Step 0]
+       ↓
+model-research-latest.json
+       │ validates against:
+       ↓ model-research.schema.json
+bun run agent-evolution/scripts/sync-model-research.ts
+       ├─ обновляет .kilo/capability-index.yaml (model поля)
+       ├─ обновляет kilo-meta.json (source of truth)
+       ├─ обновляет kilo.jsonc (agent config)
+       ├─ обновляет agent-evolution/data/agent-versions.json (история)
+       ├─ обновляет .kilo/agents/*.md frontmatter (через sync-agents.js --fix)
+       └─ rebuilds dashboard (build-research-dashboard.ts)
+       ↓
+bun run agent-evolution/scripts/build-research-dashboard.ts
+       ├─ читает model-benchmarks.json
+       ├─ инжектирует в research-dashboard.template.html
+       ├─ записывает research-dashboard.html
+       └─ копирует dist/research-dashboard-YYYY_MM_DD.html
+       ↓
+[/research models] ← цикл продолжается
+```
+
+## Структура model-benchmarks.json
+
+### Верхний уровень
+
+```json
+{
+  "version": "1.0.0",
+  "generated": "2026-04-27T17:44:44.000Z",
+  "source": "apaw_agent_model_research_v3.html",
+  "total_agents": 36,
+  "total_models_tracked": 11,
+  "providers": ["ollama", "ollama-cloud", "openrouter", "groq"],
+  "models": [...],
+  "groq_models": [...],
+  "agent_model_scores": [...],
+  "if_scores": {...},
+  "agent_current_config": [...],
+  "recommendations": [...],
+  "impact_data": [...],
+  "benchmark_comparison": {...}
+}
+```
+
+### Модель
+
+```json
+{
+  "id": "ollama-cloud/qwen3-coder:480b",
+  "name": "Qwen3-Coder 480B",
+  "organization": "Qwen",
+  "parameters": "480B/35B active",
+  "context_window": "256K\u21921M",
+  "swe_bench": 66.5,
+  "swe_bench_pro": null,
+  "terminal_bench": null,
+  "live_codebench": null,
+  "gpqa": null,
+  "hle": null,
+  "browse_comp": null,
+  "if_score": 88,
+  "categories": ["coding", "agent"],
+  "tags": ["coding", "agent", "tools"],
+  "provider": "ollama",
+  "free": false,
+  "cost_per_1m_input": "~$0.50",
+  "description": "SOTA open-source \u043a\u043e\u0434\u0438\u043d\u0433. \u0421\u0440\u0430\u0432\u043d\u0438\u043c \u0441 Claude Sonnet 4.",
+  "availability": null,
+  "speed_tps": null
+}
+```
+
+### Рекомендация
+
+```json
+{
+  "agent": "planner",
+  "action": "update_model",
+  "current_model": "nemotron-3-super",
+  "current_provider": "Ollama",
+  "recommended_model": "deepseek-v4-pro-max",
+  "recommended_provider": "Ollama Cloud",
+  "impact": "high",
+  "score_before": 80,
+  "score_after": 88,
+  "score_delta": 8,
+  "expected_improvement": {
+    "quality": "+10%",
+    "speed": "~1x",
+    "context_window": "1M"
+  },
+  "rationale": "\u2605 matri\u0446\u044b: V4-Pro=88(\u043b\u0443\u0447\u0448\u0438\u0439!)..."
+}
+```
+
+### Очки агента
+
+```json
+{
+  "agent": "lead-developer",
+  "current_model_index": 0,
+  "reasoning_effort": "M",
+  "scores": {
+    "ollama-cloud/qwen3-coder:480b": 92,
+    "ollama-cloud/minimax-m2.5": 86,
+    "ollama-cloud/minimax-m2.7": 82,
+    "ollama-cloud/nemotron-3-super": 70,
+    "ollama-cloud/glm-5": 68,
+    "ollama-cloud/glm-5.1": 75,
+    "ollama-cloud/deepseek-v4-pro-max": 88,
+    "ollama-cloud/qwen3.5-122b": 66,
+    "ollama-cloud/qwen3-coder-next": 80,
+    "openrouter/qwen/qwen3.6-plus:free": 88,
+    "ollama-cloud/kimi-k2.6:cloud": 90
+  }
+}
+```
+
+## Формула IF-ажастмента
+
+Оценка агента с учётом способности модели следовать инструкциям:
+
+```
+IF-adjusted_score = raw_score × (0.7 + 0.3 × IF/100)
+
+Где:
+  raw_score — бенчмарк оценка пары агент×модель (0-100)
+  IF — instruction following score модели (0-100)
+
+Примеры:
+  IF=100 → score × 1.00 (без изменений)
+  IF=90  → score × 0.97
+  IF=78  → score × 0.93
+  IF=50  → score × 0.85
+  IF=0   → score × 0.70
+
+Чем ниже IF, тем сильнее штраф — модель плохо следует промпту и роли.
+```
+
+## Скрипты системы
+
+### build-research-dashboard.ts
+
+**Вход**: `model-benchmarks.json` + `research-dashboard.template.html`
+**Выход**: `research-dashboard.html` + `dist/dashboard-YYYY_MM_DD.html`
+
+```bash
+bun run agent-evolution/scripts/build-research-dashboard.ts              # однократная сборка
+bun run agent-evolution/scripts/build-research-dashboard.ts --watch       # watch-режим
+bun run agent-evolution/scripts/build-research-dashboard.ts --template custom.html
+```
+
+Процесс:
+1. Читает JSON, валидирует наличие полей
+2. Читает шаблон, ищет placeholder `// BENCHMARK_DATA_PLACEHOLDER`
+3. Заменяет `const EMBEDDED_DATA = {};` на полный JSON с данными
+4. Обновляет `<title>` с датой генерации
+5. Пишет `research-dashboard.html` и архивную копию
+
+### sync-model-research.ts
+
+**Вход**: `model-research-latest.json`
+**Действия**:
+
+```bash
+# Предпросмотр
+bun run agent-evolution/scripts/sync-model-research.ts --dry-run
+
+# Применение всех рекомендаций
+bun run agent-evolution/scripts/sync-model-research.ts
+
+# Только для одного агента
+bun run agent-evolution/scripts/sync-model-research.ts --agent planner
+```
+
+Для каждой рекомендации (`action: "update_model"`, `applied: false`):
+1. Находит блок агента в `capability-index.yaml`, заменяет `model:`
+2. Обновляет `kilo-meta.json` (source of truth)
+3. Обновляет `kilo.jsonc` (через regex, требует ручной проверки)
+4. Добавляет запись в `agent-versions.json` history
+5. Запускает `node scripts/sync-agents.js --fix` → обновляет .md frontmatter
+6. Запускает `node scripts/sync-agents.js --check` → проверка консистентности
+7. Пересобирает дашборд через `build-research-dashboard.ts`
+
+## Интеграция в пайплайн
+
+### /research models
+
+```
+1. Загрузить текущие данные из model-benchmarks.json
+2. Если stale (>7 дней) или --force:
+   a. Fetch моделей с Ollama Cloud, OpenRouter, Groq
+   b. Compute IF scores для каждой модели
+   c. Score каждую модель против каждого агента
+3. Сгенерировать рекомендации (gap > 5)
+4. Записать model-research-latest.json
+5. Валидировать против model-research.schema.json
+6. Обновить model-benchmarks.json (если данные изменились)
+7. Пересобрать дашборд
+```
+
+### /evolution (полный цикл)
+
+```
+Step 0: Model Research
+  ├─ Проверить staleness model-benchmarks.json
+  ├─ Если stale → @capability-analyst исследует модели
+  ├─ Загрузить heatmap scores
+  └─ Определить агентов с mismatch (gap > 5)
+
+Step 1: Judge
+  └─ @pipeline-judge → fitness score
+
+Step 2: Decide
+  ├─ fitness >= 0.85 → выход
+  ├─ fitness >= 0.70 → @prompt-optimizer (minor)
+  └─ fitness < 0.70 → @prompt-optimizer (major) + apply model recs
+
+Step 3: Re-test
+  └─ Перезапуск с обновлёнными промптами/моделями
+
+Step 4: Log + Dashboard
+  ├─ Append fitness-history.jsonl
+  ├─ Apply рекомендации sync-model-research.ts
+  └─ Пересобрать дашборд build-research-dashboard.ts
+```
+
+### /evolution research
+
+```
+1. Прочитать текущую конфигурацию
+2. Исследовать модели (как /research models)
+3. Сгенерировать рекомендации
+4. Dry-run preview
+5. Применить при подтверждении
+6. Пересобрать дашборд
+```
+
+## Правила синхронизации
+
+Из `.kilo/rules/evolutionary-sync.md`:
+
+### Обязательный порядок
+
+```
+1. Обновить kilo-meta.json (source of truth)
+2. Обновить capability-index.yaml
+3. Запустить sync-agents.js --fix
+4. Ручная проверка kilo.jsonc (sync script не гарантирует)
+5. Запустить sync-agents.js --check
+6. Проверить agent-versions.json history
+7. Пересобрать дашборд
+8. Если любая проверка не прошла — НЕ коммитить
+```
+
+### Облачный суффикс
+
+При использовании `ollama-cloud/kimi-k2.6` ВСЕГДА с суффиксом `:cloud`:
+
+```yaml
+# Правильно
+model: "ollama-cloud/kimi-k2.6:cloud"
+
+# Неправильно — отсутствует суффикс
+model: "ollama-cloud/kimi-k2.6"
+```
+
+## Чеклист применения изменений
+
+```
+□ Исследование: /research models завершено
+□ Валидация: model-research-latest.json проходит schema check
+□ Dry-run: sync-model-research.ts --dry-run показывает ожидаемые изменения
+□ Применение: sync-model-research.ts выполнен без ошибок
+□ YAML: capability-index.yaml обновлены поля model
+□ Meta: kilo-meta.json соответствует
+□ kilo.jsonc: модели обновлены (ручная проверка)
+□ История: agent-versions.json записи добавлены
+□ Sync: sync-agents.js --fix обновил все .md файлы
+□ Check: sync-agents.js --check проходит
+□ Старые модели: grep не находит предыдущие model IDs
+□ Суффикс: kimi-k2.6:cloud (с :cloud)
+□ Дашборд: build-research-dashboard.ts сгенерировал свежий HTML
+□ Открыть: research-dashboard.html показывает актуальные данные
+□ Гит: все изменения add и commit
+```
+
+## Устранение неполадок
+
+| Проблема | Диагностика | Решение |
+|----------|------------|---------|
+| Дашборд пустой | Проверить placeholder в template.html | Пересобрать: `bun run build-research-dashboard.ts` |
+| Schema validation fails | Сравнить JSON со схемой | Проверить model-research.schema.json актуальность |
+| sync-agents.js check fails | Model mismatch в конфигах | Запустить `--fix`, затем `--check`; ручная проверка kilo.jsonc |
+| Heatmap пустой | agent_model_scores отсутствует | Обновить бенчмарки через `/research models` |
+| Рекомендации не отображаются | Empty recs array | Запустить research для генерации новых рекомендаций |
+| Старые данные | metadata.generated > 7 дней | Обновить бенчмарки |
+| sync-model-research.ts падает | Файл не найден | Проверить пути, запустить из корня проекта |
+
+## Пример полного цикла
+
+### 1. Исследование моделей
+
+```bash
+$ /research models
+
+## Research: model optimization
+
+### Models Analyzed
+- Ollama Cloud: 20 models
+- OpenRouter Free: 3 models
+- Groq Free: 5 models
+
+### Key Findings
+- DeepSeek V4-Pro Max доступен (SWE-V 80.6, IF:88)
+- Kimi K2.6 IF=91 (лучший для orchestration)
+- Nemotron 3 Super IF=78 — слаб для prompt-heavy ролей
+- Qwen 3.6 Plus FREE остаётся лучшим IF/cost (91, $0)
+
+### Recommendations Generated
+- 11 model swap recommendations
+- 4 high, 3 medium, 4 low
+- Средний expected improvement: +12 points
+
+### Files Updated
+- agent-evolution/data/model-research-latest.json
+- agent-evolution/data/model-benchmarks.json (refreshed)
+- agent-evolution/dist/research-dashboard-2026_04_27.html (archive)
+```
+
+### 2. Валидация schema
+
+```bash
+$ node -e "
+const Ajv = require('ajv');
+const ajv = new Ajv();
+const schema = JSON.parse(require('fs').readFileSync('agent-evolution/data/model-research.schema.json','utf8'));
+const data = JSON.parse(require('fs').readFileSync('agent-evolution/data/model-research-latest.json','utf8'));
+const valid = ajv.validate(schema, data);
+console.log(valid ? 'VALID' : 'INVALID');
+if (!valid) console.log(JSON.stringify(ajv.errors, null, 2));
+"
+VALID
+```
+
+### 3. Dry-run
+
+```bash
+$ bun run agent-evolution/scripts/sync-model-research.ts --dry-run
+
+=== SYNC PREVIEW (dry-run) ===
+3 agents would be updated:
+
+planner
+  FROM: nemotron-3-super (Ollama)
+  TO:   deepseek-v4-pro-max (Ollama Cloud)
+  DELTA: +8 (80 → 88)
+  IMPACT: high
+
+go-developer
+  FROM: qwen3-coder:480b (Ollama)
+  TO:   deepseek-v4-pro-max (Ollama Cloud)
+  DELTA: +3 (85 → 88)
+  IMPACT: medium
+
+[built-in] debug
+  FROM: glm-5.1 (Ollama)
+  TO:   kimi-k2.6:cloud (Ollama Cloud)
+  DELTA: +2 (88 → 90)
+  IMPACT: high
+
+Files to modify: capability-index.yaml, kilo-meta.json, kilo.jsonc, agent-versions.json
+```
+
+### 4. Применение
+
+```bash
+$ bun run agent-evolution/scripts/sync-model-research.ts
+
+✅ capability-index.yaml updated (3 agents)
+✅ kilo-meta.json updated  
+✅ kilo.jsonc updated
+✅ agent-versions.json history updated (3 entries)
+✅ sync-agents.js --fix completed
+✅ sync-agents.js --check passed
+✅ Dashboard rebuilt: research-dashboard.html (106KB)
+```
+
+### 5. Проверка дашборда
+
+```bash
+$ start agent-evolution/research-dashboard.html
+
+# В браузере:
+# - Overview: 3 agents updated, 11 recommendations total
+# - Heatmap: V4-Pro Max column green for planner, go-developer
+# - Recommendations: 3 marked as applied with checkmarks
+# - Impact: +8 for planner shown in chart
+```
+
+### 6. Тест пайплайна
+
+```bash
+$ /evolve --issue 42
+
+## Pipeline Judgment: Issue #42
+
+**Fitness: 0.88/1.00** [PASS → improved from 0.82]
+
+| Metric | Value | Weight | Contribution |
+|--------|-------|--------|-------------|
+| Tests  | 96% (46/48) | 50% | 0.480 |
+| Gates  | 80% (4/5) | 25% | 0.200 |
+| Cost   | 38.4K tok / 245s | 25% | 0.198 |
+
+**Bottleneck:** none (all agents optimal)
+**Verdict:** PASS — fitness improved!
+
+✅ Logged to .kilo/logs/fitness-history.jsonl
+✅ Auto-rebuilt: agent-evolution/research-dashboard.html
+```
+
+## Периодичность обновления
+
+| Файл | Период | Триггер |
+|------|--------|---------|
+| model-benchmarks.json | Еженедельно (>7 дней = stale) | `/evolution Step 0` или `/research models` |
+| model-research-latest.json | Каждый research cycle | `/research models`, `/evolution research` |
+| research-dashboard.html | После каждого изменения | `sync-model-research.ts` или `build-research-dashboard.ts` |
+| dist/*.html | Архив | Каждая генерация |
+| agent-versions.json | При каждом изменении модели | `sync-model-research.ts` |
+
+## Связанные документы
+
+- `.kilo/commands/evolution.md` — команда /evolution
+- `.kilo/commands/research.md` — команда /research
+- `.kilo/shared/self-evolution.md` — протокол эволюции
+- `.kilo/rules/evolutionary-sync.md` — правила синхронизации
+- `.kilo/rules/agent-frontmatter-validation.md` — валидация YAML frontmatter
+- `agent-evolution/README.md` — обзор системы эволюции
+- `kilo-meta.json` — source of truth для моделей
+- `.kilo/capability-index.yaml` — маршрутизация и назначения
--- a/agent-evolution/ideas/apaw_agent_model_research_v3.html
+++ b/agent-evolution/ideas/apaw_agent_model_research_v3.html
--- a/agent-evolution/index.standalone.html
+++ b/agent-evolution/index.standalone.html
@@ -674,16 +674,16 @@
 // Supports both server and file:// mode
 let agentData = {};

-// Embedded data (generated 2026-04-23T06:24:32.710Z)
+// Embedded data (generated 2026-04-27T20:28:59.112Z)
 const EMBEDDED_DATA = {
  "version": "1.0.0",
-  "lastUpdated": "2026-04-23T06:24:32.543Z",
+  "lastUpdated": "2026-04-27T20:28:58.592Z",
  "agents": {
    "lead-developer": {
      "current": {
        "description": "Primary code writer for backend and core logic. Writes implementation to pass tests",
        "mode": "subagent",
-        "model": "ollama-cloud/qwen3-coder:480b",
+        "model": "ollama-cloud/nemotron-3-super",
        "provider": "Ollama",
        "variant": "thinking",
        "color": "\"#DC2626\"",
@@ -704,6 +704,24 @@ const EMBEDDED_DATA = {
          "to": "ollama-cloud/qwen3-coder:480b",
          "reason": "Initial configuration from capability-index.yaml",
          "source": "git"
+        },
+        {
+          "date": "2026-04-27T16:56:09.013Z",
+          "commit": "model-research-sync",
+          "type": "model_change",
+          "from": "ollama-cloud/qwen3-coder:480b",
+          "to": "ollama-cloud/nemotron-3-super",
+          "reason": "Nemotron 3 Super has better reasoning for core development tasks and RULER@1M context window. SWE-bench 68% vs Qwen's 66.5%.",
+          "source": "research"
+        },
+        {
+          "date": "2026-04-27T20:28:58.592Z",
+          "commit": "model-research-sync",
+          "type": "model_change",
+          "from": "ollama-cloud/qwen3-coder:480b",
+          "to": "ollama-cloud/nemotron-3-super",
+          "reason": "Nemotron 3 Super has better reasoning for core development tasks and RULER@1M context window. SWE-bench 68% vs Qwen's 66.5%.",
+          "source": "research"
        }
      ],
      "performance_log": []
@@ -932,7 +950,7 @@ const EMBEDDED_DATA = {
      "current": {
        "description": "Designs technical specifications, data schemas, and API contracts before implementation",
        "mode": "subagent",
-        "model": "ollama-cloud/glm-5.1",
+        "model": "ollama-cloud/nemotron-3-super",
        "provider": "Ollama",
        "variant": "thinking",
        "color": "\"#0891B2\"",
@@ -962,6 +980,15 @@ const EMBEDDED_DATA = {
          "to": "ollama-cloud/glm-5.1",
          "reason": "Model update from sync",
          "source": "git"
+        },
+        {
+          "date": "2026-04-27T16:59:52.825Z",
+          "commit": "model-research-sync",
+          "type": "model_change",
+          "from": "ollama-cloud/glm-5.1",
+          "to": "ollama-cloud/nemotron-3-super",
+          "reason": "Test recommendation for model research sync script",
+          "source": "research"
        }
      ],
      "performance_log": []
--- a/agent-evolution/research-dashboard.html
+++ b/agent-evolution/research-dashboard.html
--- a/agent-evolution/research-dashboard.template.html
+++ b/agent-evolution/research-dashboard.template.html
--- a/agent-evolution/scripts/build-research-dashboard.ts
+++ b/agent-evolution/scripts/build-research-dashboard.ts
@@ -0,0 +1,237 @@
+#!/usr/bin/env bun
+/**
+ * Build APAW Agent Model Research Dashboard from live data.
+ *
+ * Reads model-benchmarks.json and injects into template HTML.
+ * Creates standalone dashboard with embedded JSON data.
+ *
+ * Usage:
+ *   bun run agent-evolution/scripts/build-research-dashboard.ts              # build once
+ *   bun run agent-evolution/scripts/build-research-dashboard.ts --watch       # watch mode
+ *   bun run agent-evolution/scripts/build-research-dashboard.ts --template path/to/custom.html
+ */
+
+import { existsSync, readFileSync, writeFileSync, watch } from 'fs';
+import { join, dirname, basename } from 'path';
+import { fileURLToPath } from 'url';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = dirname(__filename);
+
+const DATA_FILE = join(__dirname, '../data/model-benchmarks.json');
+const DEFAULT_TEMPLATE = join(__dirname, '../research-dashboard.template.html');
+const OUTPUT_FILE = join(__dirname, '../research-dashboard.html');
+const DIST_DIR = join(__dirname, '../dist');
+
+interface BenchmarksData {
+  version: string;
+  generated: string;
+  source: string;
+  total_agents: number;
+  total_models_tracked: number;
+  providers: string[];
+  models: any[];
+  agent_model_scores: any[];
+  agent_current_config: any[];
+  groq_models: any[];
+  recommendations: any[];
+  impact_data: any[];
+}
+
+function buildDashboard(templatePath: string = DEFAULT_TEMPLATE): boolean {
+  console.log('🔧 Building APAW Agent Model Research Dashboard');
+
+  // Validate inputs
+  if (!existsSync(DATA_FILE)) {
+    console.error(`❌ Data file not found: ${DATA_FILE}`);
+    console.error('   Please run research cycle first: bun run /research models');
+    return false;
+  }
+
+  if (!existsSync(templatePath)) {
+    console.error(`❌ Template file not found: ${templatePath}`);
+    console.error('   Using default template:', DEFAULT_TEMPLATE);
+    if (!existsSync(DEFAULT_TEMPLATE)) {
+      console.error('   Default template also missing. Create template first.');
+      return false;
+    }
+    templatePath = DEFAULT_TEMPLATE;
+  }
+
+  // Read and validate JSON data
+  let data: BenchmarksData;
+  try {
+    const rawData = readFileSync(DATA_FILE, 'utf-8');
+    data = JSON.parse(rawData);
+    console.log(`📖 Read model-benchmarks.json (${rawData.length} bytes)`);
+  } catch (error) {
+    console.error(`❌ Failed to parse JSON data: ${error}`);
+    return false;
+  }
+
+  // Validate required fields
+  if (!data.models || !Array.isArray(data.models)) {
+    console.error('❌ Missing or invalid "models" array in data');
+    return false;
+  }
+  
+  if (!data.agent_model_scores || !Array.isArray(data.agent_model_scores)) {
+    console.error('❌ Missing or invalid "agent_model_scores" array in data');
+    return false;
+  }
+
+  console.log(`   Models: ${data.models.length}`);
+  console.log(`   Agents: ${data.agent_model_scores.length}`);
+  console.log(`   Providers: ${data.providers?.join(', ') || 'unknown'}`);
+  console.log(`   Generated: ${data.generated}`);
+
+  // Read HTML template
+  let html: string;
+  try {
+    html = readFileSync(templatePath, 'utf-8');
+    console.log(`📖 Read template: ${templatePath} (${html.length} bytes)`);
+  } catch (error) {
+    console.error(`❌ Failed to read template: ${error}`);
+    return false;
+  }
+
+  // Find and replace placeholder — must match exact text in template
+  const placeholder = '// BENCHMARK_DATA_PLACEHOLDER - will be replaced by build script\nconst EMBEDDED_DATA = {};\n';
+  if (!html.includes(placeholder)) {
+    // Try looser match with any line endings
+    const loosePlaceholder = html.match(/\/\/\s*BENCHMARK_DATA_PLACEHOLDER[^\n]*\r?\n\s*const\s+EMBEDDED_DATA\s*=\s*\{\}\s*;\r?\n/);
+    if (!loosePlaceholder) {
+      console.error('❌ Placeholder not found in template');
+      console.error('   Expected: "// BENCHMARK_DATA_PLACEHOLDER - will be replaced by build script\\nconst EMBEDDED_DATA = {};\\n"');
+      const match = html.match(/BENCHMARK_DATA_PLACEHOLDER/);
+      if (match) {
+        const start = Math.max(0, match.index - 20);
+        const end = Math.min(html.length, match.index + 120);
+        console.error('   Found near:', JSON.stringify(html.slice(start, end)));
+      }
+      return false;
+    }
+    html = html.replace(loosePlaceholder[0], `// BENCHMARK_DATA_PLACEHOLDER - REPLACED BY BUILD SCRIPT\n// Generated from ${basename(DATA_FILE)} on ${new Date().toISOString()}\nconst EMBEDDED_DATA = ${JSON.stringify(data, null, 2)};\n`);
+  } else {
+    html = html.replace(placeholder, `// BENCHMARK_DATA_PLACEHOLDER - REPLACED BY BUILD SCRIPT\n// Generated from ${basename(DATA_FILE)} on ${new Date().toISOString()}\nconst EMBEDDED_DATA = ${JSON.stringify(data, null, 2)};\n`);
+  }
+
+  // Update title with metadata if present (match any tag with APAW... in it)
+  const titleRegex = /<title>[^<]*APAW[^<]*<\/title>/;
+  if (titleRegex.test(html)) {
+    const newTitle = `APAW Agent Model Research — generated ${data.generated.slice(0, 10)}`;
+    html = html.replace(titleRegex, `<title>${newTitle}</title>`);
+  }
+  
+  // Update subtitle if present
+  const subtitlePattern = /<div class="sub">([^<]*)<\/div>/;
+  const newSubtitle = `<div class="sub">Live dashboard • ${data.models.length} models × ${data.agent_model_scores.length} agents • ${data.generated.slice(0, 10)}</div>`;
+  if (subtitlePattern.test(html)) {
+    html = html.replace(subtitlePattern, newSubtitle);
+  }
+
+  // Write output file
+  try {
+    writeFileSync(OUTPUT_FILE, html);
+    console.log(`✅ Output written to: ${OUTPUT_FILE} (${html.length} bytes)`);
+  } catch (error) {
+    console.error(`❌ Failed to write output: ${error}`);
+    return false;
+  }
+
+  // Create dated version in dist directory
+  try {
+    if (!existsSync(DIST_DIR)) {
+      require('fs').mkdirSync(DIST_DIR, { recursive: true });
+    }
+    const dateStr = data.generated.slice(0, 10).replace(/-/g, '_');
+    const distFile = join(DIST_DIR, `research-dashboard-${dateStr}.html`);
+    writeFileSync(distFile, html);
+    console.log(`📁 Dated copy: ${distFile}`);
+  } catch (error) {
+    console.warn(`⚠️  Could not create dated copy: ${error}`);
+  }
+
+  // Print summary
+  const recommendations = data.recommendations || [];
+  console.log('\n📊 Summary:');
+  console.log(`   • Agents tracked: ${data.total_agents || data.agent_model_scores.length}`);
+  console.log(`   • Models benchmarked: ${data.total_models_tracked || data.models.length}`);
+  console.log(`   • Providers: ${data.providers?.join(', ')}`);
+  console.log(`   • Recommendations: ${recommendations.length}`);
+  
+  if (recommendations.length >577.0) {
+    const highImpact = recommendations.filter((r: any) => r.impact === 'high').length;
+    const applied = recommendations.filter((r: any) => r.to_model?.includes('✅')).length;
+    console.log(`   • High-impact recommendations: ${highImpact}`);
+    console.log(`   • Applied recommendations: ${applied}`);
+  }
+
+  return true;
+}
+
+function watchMode(): void {
+  console.log('👀 Watch mode enabled - monitoring data and template files');
+  console.log('   Press Ctrl+C to stop');
+  
+  let timeout: Timer | null = null;
+  
+  watch(DATA_FILE, (eventType) => {
+    if (eventType === 'change') {
+      if (timeout) clearTimeout(timeout);
+      timeout = setTimeout(() => {
+        console.log('\n🔄 Data file changed, rebuilding...');
+        buildDashboard();
+      }, 500);
+    }
+  });
+  
+  watch(DEFAULT_TEMPLATE, (eventType) => {
+    if (eventType === 'change') {
+      if (timeout) clearTimeout(timeout);
+      timeout = setTimeout(() => {
+        console.log('\n🔄 Template file changed, rebuilding...');
+        buildDashboard();
+      }, 500);
+    }
+  });
+}
+
+// Parse CLI arguments
+const args = process.argv.slice(2);
+let watchModeEnabled = false;
+let customTemplate: string | undefined;
+
+for (let i = 0; i < args.length; i++) {
+  if (args[i] === '--watch') {
+    watchModeEnabled = true;
+  } else if (args[i] === '--template' && i + 1 < args.length) {
+    customTemplate = args[i + 1];
+    i++;
+  } else if (args[i] === '--help' || args[i] === '-h') {
+    console.log(`
+Usage: bun run agent-evolution/scripts/build-research-dashboard.ts [options]
+
+Options:
+  --watch                    Watch for changes and rebuild automatically
+  --template <path>          Use custom HTML template file
+  --help, -h                 Show this help message
+
+Examples:
+  bun run agent-evolution/scripts/build-research-dashboard.ts
+  bun run agent-evolution/scripts/build-research-dashboard.ts --watch
+  bun run agent-evolution/scripts/build-research-dashboard.ts --template custom.html
+`);
+    process.exit(0);
+  }
+}
+
+// Main execution
+if (watchModeEnabled) {
+  // Build once then watch
+  buildDashboard(customTemplate);
+  watchMode();
+} else {
+  const success = buildDashboard(customTemplate);
+  process.exit(success ? 0 : 1);
+}
--- a/agent-evolution/scripts/rebuild-template.cjs
+++ b/agent-evolution/scripts/rebuild-template.cjs
@@ -0,0 +1,74 @@
+const fs = require('fs');
+const v3 = fs.readFileSync('agent-evolution/ideas/apaw_agent_model_research_v3.html', 'utf8');
+
+const dataStart = v3.indexOf('// ACTUAL STATE from _kilo.zip');
+const renderStart = v3.indexOf('// ======================= RENDER =======================');
+
+if (dataStart === -1 || renderStart === -1) {
+  console.error('Cannot find markers');
+  process.exit(1);
+}
+
+const mapping = `// BENCHMARK_DATA_PLACEHOLDER - will be replaced by build script
+const EMBEDDED_DATA = {};
+
+// === MAP EMBEDDED_DATA -> original v3 format ===
+const allModels = EMBEDDED_DATA.models || [];
+const scoreModelIds = Object.keys((EMBEDDED_DATA.agent_model_scores || [])[0]?.scores || {});
+const activeModels = allModels.filter(m => scoreModelIds.includes(m.id));
+
+const cfg = (EMBEDDED_DATA.agent_current_config || []).map(c => {
+    const modelId = (c.model || '').replace('ollama-cloud/', '');
+    const badge = c.badge_type || (
+        modelId.includes('qwen3') ? 'qwen' :
+        modelId.includes('minimax') ? 'minimax' :
+        modelId.includes('nemotron') ? 'nemotron' :
+        modelId.includes('glm') ? 'glm' :
+        modelId.includes('kimi') ? 'kimi' :
+        modelId.includes('deepseek') ? 'deepseek' : 'groq'
+    );
+    return { a: c.agent, m: modelId, p: c.provider || 'Ollama', cat: c.category || 'General', b: badge, fit: c.fit_score || 0, s: c.status || 'good', prev: c.previous_model };
+});
+
+const groqModels = (EMBEDDED_DATA.groq_models || []).map(g => ({
+    id: g.id, rpm: g.rpm, rpd: g.rpd, tpm: g.tpm, tpd: g.tpd, speed: g.speed, use: g.use_case
+}));
+
+const ollamaModels = activeModels.map(m => ({
+    n: m.name, org: m.organization, par: m.parameters, ctx: m.context_window,
+    swe: m.swe_bench, ifScore: m.if_score, cat: m.categories || [],
+    str: m.description, tags: m.tags || [], or: m.openrouter, groqSpeed: m.speed_tps
+}));
+
+const ifScores = {};
+activeModels.forEach((m, i) => { if (m.if_score) ifScores[i] = m.if_score; });
+
+const hmModels = activeModels.map(m => ({
+    n: m.display_name || m.name?.split(' ').pop() || m.id,
+    p: m.provider === 'ollama-cloud' ? 'Ollama Cloud' : m.provider === 'openrouter' ? 'OpenRouter' : m.provider || 'Ollama',
+    if: m.if_score || 0
+}));
+
+const hmAgents = (EMBEDDED_DATA.agent_model_scores || []).map(ag => {
+    const scores = activeModels.map(m => ag.scores?.[m.id] ?? 0);
+    const fullModelId = allModels[ag.current_model_index]?.id;
+    const c = activeModels.findIndex(m => m.id === fullModelId);
+    return { n: ag.agent, c: c, re: ag.reasoning_effort || 'M', s: scores };
+});
+
+const recs = (EMBEDDED_DATA.recommendations || []).map(r => ({
+    a: r.agent, from: r.from_model, fromP: r.from_provider || 'Ollama',
+    to: r.to_model, toP: r.to_provider || 'Ollama', imp: r.impact || 'low',
+    q: r.quality_change || '0', sp: r.speed_change || '=', ctx: r.context_change || '-',
+    prov: r.provider_change || r.to_provider || 'Ollama', r: r.rationale
+}));
+
+const impactData = (EMBEDDED_DATA.impact_data || []).map(d => ({
+    cat: d.category, b: d.before, a: d.after, d: d.delta, n: d.notes || d.note
+}));
+
+`;
+
+const final = v3.substring(0, dataStart) + mapping + v3.substring(renderStart);
+fs.writeFileSync('agent-evolution/research-dashboard.template.html', final);
+console.log('Template written:', final.length, 'chars,', final.split('\n').length, 'lines');
--- a/agent-evolution/scripts/sync-benchmarks-from-yaml.cjs
+++ b/agent-evolution/scripts/sync-benchmarks-from-yaml.cjs
@@ -0,0 +1,136 @@
+const fs = require('fs');
+
+// Parse simple YAML structure with 2-space indentation
+function parseCapabilityIndex(text) {
+  const lines = text.split(/\r?\n/);
+  const agents = {};
+  let currentAgent = '';
+  let currentList = '';
+  
+  for (const line of lines) {
+    const indent = line.length - line.trimStart().length;
+    const trimmed = line.trim();
+    
+    if (indent === 2 && trimmed.endsWith(':') && !trimmed.startsWith('-')) {
+      // Agent name
+      currentAgent = trimmed.slice(0, -1);
+      agents[currentAgent] = {};
+      currentList = '';
+      continue;
+    }
+    
+    if (indent === 4 && trimmed.endsWith(':') && !trimmed.startsWith('-')) {
+      // Scalar property or list start
+      const key = trimmed.slice(0, -1);
+      currentList = key;
+      if (!Array.isArray(agents[currentAgent][key])) {
+        agents[currentAgent][key] = [];
+      }
+      continue;
+    }
+    
+    if (indent === 4 && trimmed.includes(':') && !trimmed.startsWith('-')) {
+      // key: value
+      const [key, ...rest] = trimmed.split(':');
+      const value = rest.join(':').trim();
+      agents[currentAgent][key.trim()] = value;
+      currentList = '';
+      continue;
+    }
+    
+    if (indent >= 6 && trimmed.startsWith('- ')) {
+      // List item
+      const value = trimmed.slice(2).trim();
+      if (currentList) {
+        if (!agents[currentAgent][currentList]) agents[currentAgent][currentList] = [];
+        agents[currentAgent][currentList].push(value);
+      }
+      continue;
+    }
+    
+    // Reset list context on unknown indentation
+    if (indent < 4) {
+      currentList = '';
+    }
+  }
+  
+  // Filter out non-agent entries (flat sections like capability_routing, etc.)
+  const result = {};
+  const scalarKeys = ['capabilities','receives','produces','forbidden','delegates_to','fallback_models'];
+  for (const [name, data] of Object.entries(agents)) {
+    const hasAgentProps = scalarKeys.some(k => k in data) || 'model' in data;
+    if (hasAgentProps) result[name] = data;
+  }
+  
+  return result;
+}
+
+const yaml = fs.readFileSync('.kilo/capability-index.yaml', 'utf8');
+const parsed = parseCapabilityIndex(yaml);
+console.log('Parsed agents:', Object.keys(parsed).length);
+
+// Read existing benchmarks
+const bench = JSON.parse(fs.readFileSync('agent-evolution/data/model-benchmarks.json', 'utf8'));
+
+// Update agent_current_config
+bench.agent_current_config = Object.entries(parsed).map(([agent, data]) => {
+  const rawModel = data.model || '';
+  const modelId = rawModel.replace('ollama-cloud/', '');
+  const badge = modelId.includes('qwen3') ? 'qwen' :
+    modelId.includes('minimax') ? 'minimax' :
+    modelId.includes('nemotron') ? 'nemotron' :
+    modelId.includes('glm') ? 'glm' :
+    modelId.includes('kimi') ? 'kimi' :
+    modelId.includes('deepseek') ? 'deepseek' : 'groq';
+  return {
+    agent,
+    model: rawModel,
+    provider: data.mode === 'all' ? 'Ollama Cloud' : (rawModel.startsWith('ollama-cloud/') ? 'Ollama Cloud' : 'Ollama'),
+    category: 'Process',
+    badge_type: badge,
+    fit_score: 0,
+    status: 'good',
+    previous_model: null
+  };
+});
+
+// Update agent_model_scores — preserve existing scores, fix current_model_id
+const existingScores = {};
+(bench.agent_model_scores || []).forEach(s => {
+  existingScores[s.agent] = s.scores || {};
+});
+
+bench.agent_model_scores = Object.entries(parsed).map(([agent, data]) => {
+  const rawModel = data.model || '';
+  const modelId = rawModel.replace('ollama-cloud/', '');
+  const currentIndex = bench.models.findIndex(m => m.id === modelId);
+  // Preserve existing scores or empty
+  const scores = existingScores[agent] || {};
+  return {
+    agent,
+    current_model_index: currentIndex >= 0 ? currentIndex : -1,
+    current_model_id: modelId,
+    reasoning_effort: data.variant === 'thinking' ? 'H' : 'M',
+    scores
+  };
+});
+
+// Update metadata
+bench.generated = new Date().toISOString();
+bench.source = '.kilo/capability-index.yaml (synced v2)';
+bench.total_agents = bench.agent_current_config.length;
+
+fs.writeFileSync('agent-evolution/data/model-benchmarks.json', JSON.stringify(bench, null, 2));
+console.log('Synced', bench.agent_current_config.length, 'agents');
+console.log('Generated:', bench.generated);
+
+// Verify
+let mismatches = 0;
+bench.agent_current_config.forEach(c => {
+  const scores = bench.agent_model_scores.find(s => s.agent === c.agent);
+  if (scores && scores.current_model_id !== c.model.replace('ollama-cloud/', '')) {
+    console.log('  MISMATCH:', c.agent, scores.current_model_id, '->', c.model);
+    mismatches++;
+  }
+});
+console.log('Mismatches:', mismatches);
--- a/agent-evolution/scripts/sync-model-research.ts
+++ b/agent-evolution/scripts/sync-model-research.ts
@@ -0,0 +1,651 @@
+#!/usr/bin/env bun
+/**
+ * Model Research Synchronization Script
+ * Applies model recommendations from research output to agent configuration files.
+ * 
+ * Usage:
+ *   bun run agent-evolution/scripts/sync-model-research.ts                     # apply latest
+ *   bun run agent-evolution/scripts/sync-model-research.ts --dry-run             # preview only
+ *   bun run agent-evolution/scripts/sync-model-research.ts --input path/to.json # custom input
+ *   bun run agent-evolution/scripts/sync-model-research.ts --agent planner       # single agent
+ */
+
+import * as fs from "fs";
+import * as path from "path";
+import { spawnSync } from "child_process";
+
+// Types based on model-research.schema.json
+interface Recommendation {
+  agent: string;
+  action: "update_model" | "confirm_model" | "add_fallback" | "redesign_agent";
+  current_model: string;
+  recommended_model: string;
+  impact: "critical" | "high" | "medium" | "low";
+  rationale: string;
+  applied: boolean;
+  applied_date?: string | null;
+  score_delta?: number;
+}
+
+interface ModelResearchData {
+  version: string;
+  generated: string;
+  source: string;
+  recommendations: Recommendation[];
+  capability_index_patch?: Array<{
+    agent: string;
+    set: Record<string, unknown>;
+  }>;
+  summary?: {
+    total_recommendations: number;
+    applied_count: number;
+    pending_count: number;
+  };
+}
+
+interface ChangeSummary {
+  total_recommendations: number;
+  applied: number;
+  confirmed: number;
+  skipped: number;
+  errors: string[];
+  files_modified: string[];
+  agents_updated: string[];
+  dashboard_rebuilt: boolean;
+}
+
+// Default paths
+const DEFAULT_RESEARCH_FILE = path.join(__dirname, "../data/model-research-latest.json");
+const SCHEMA_FILE = path.join(__dirname, "../data/model-research.schema.json");
+const CAPABILITY_INDEX = path.join(process.cwd(), ".kilo/capability-index.yaml");
+const AGENT_VERSIONS = path.join(__dirname, "../data/agent-versions.json");
+const KILO_META = path.join(process.cwd(), "kilo-meta.json");
+const SYNC_SCRIPT = path.join(process.cwd(), "scripts/sync-agents.cjs");
+
+// Parse command line arguments
+function parseArgs(): {
+  dryRun: boolean;
+  inputFile: string;
+  singleAgent?: string;
+} {
+  const args = process.argv.slice(2);
+  const options: { dryRun: boolean; inputFile: string; singleAgent?: string } = {
+    dryRun: false,
+    inputFile: DEFAULT_RESEARCH_FILE,
+  };
+
+  for (let i = 0; i < args.length; i++) {
+    const arg = args[i];
+    if (arg === "--dry-run" || arg === "-n") {
+      options.dryRun = true;
+    } else if (arg === "--input" || arg === "-i") {
+      options.inputFile = args[++i] || DEFAULT_RESEARCH_FILE;
+    } else if (arg === "--agent" || arg === "-a") {
+      options.singleAgent = args[++i];
+    } else if (!arg.startsWith("-")) {
+      // Positional argument as input file
+      options.inputFile = arg;
+    }
+  }
+
+  return options;
+}
+
+// Load research data
+function loadResearchData(filePath: string): ModelResearchData {
+  console.log(`📖 Loading research data from: ${filePath}`);
+  
+  if (!fs.existsSync(filePath)) {
+    throw new Error(`Research file not found: ${filePath}`);
+  }
+
+  const content = fs.readFileSync(filePath, "utf-8");
+  const data = JSON.parse(content);
+
+  // Basic validation (we don't implement full schema validation for simplicity)
+  if (!data.version || !data.generated || !Array.isArray(data.recommendations)) {
+    throw new Error("Invalid research data structure");
+  }
+
+  console.log(`   Found ${data.recommendations.length} recommendations`);
+  console.log(`   Generated: ${data.generated}`);
+  console.log(`   Source: ${data.source}`);
+
+  return data;
+}
+
+// Validate schema (basic check)
+function validateSchema(data: ModelResearchData): boolean {
+  // For now, just check required fields
+  const required = [
+    "version",
+    "generated",
+    "source",
+    "recommendations",
+  ];
+
+  for (const field of required) {
+    if (!(field in data)) {
+      console.warn(`⚠️ Missing required field: ${field}`);
+      return false;
+    }
+  }
+
+  return true;
+}
+
+// Load capability-index.yaml
+function loadCapabilityIndex(): string {
+  return fs.readFileSync(CAPABILITY_INDEX, "utf-8");
+}
+
+// Update model in capability-index.yaml
+function replaceModelInYaml(content: string, agentName: string, newModel: string): { content: string; changed: boolean } {
+  // Find the agent block section
+  const agentStart = content.indexOf(`  ${agentName}:`);
+  if (agentStart === -1) {
+    throw new Error(`Agent ${agentName} not found in capability-index.yaml`);
+  }
+
+  // Find next agent section (at same indent level)
+  const remaining = content.substring(agentStart);
+  const nextAgentMatch = remaining.match(/\n  \w/);
+  const agentEnd = nextAgentMatch ? agentStart + nextAgentMatch.index! : content.length;
+
+  const agentBlock = content.substring(agentStart, agentEnd);
+  
+  // Find and replace the model line (more flexible regex for whitespace)
+  const modelLineRegex = /^\s+model:\s+.+$/gm;
+  const match = agentBlock.match(modelLineRegex);
+  
+  if (!match) {
+    throw new Error(`Model line not found in agent ${agentName} block`);
+  }
+  
+  const currentModelLine = match[0];
+  const currentModelMatch = currentModelLine.match(/:\s*(.+)$/);
+  const currentModel = currentModelMatch ? currentModelMatch[1].trim() : '';
+  
+  // Check if model already matches
+  if (currentModel === newModel) {
+    console.log(`   ⏭️ Model already set to ${newModel}, skipping`);
+    return { content, changed: false }; // No change needed
+  }
+  
+  // Replace model line with new model
+  const updatedBlock = agentBlock.replace(modelLineRegex, currentModelLine.replace(currentModel, newModel));
+  
+  if (updatedBlock === agentBlock) {
+    throw new Error(`Failed to replace model line in agent ${agentName} block`);
+  }
+
+  console.log(`   🔄 Updating model: ${currentModel} → ${newModel}`);
+  const newContent = content.substring(0, agentStart) + updatedBlock + content.substring(agentEnd);
+  return { content: newContent, changed: true };
+}
+
+// Update kilo-meta.json
+function updateKiloMeta(agentName: string, newModel: string): void {
+  const content = fs.readFileSync(KILO_META, "utf-8");
+  const data = JSON.parse(content);
+  
+  if (!data.agents[agentName]) {
+    throw new Error(`Agent ${agentName} not found in kilo-meta.json`);
+  }
+  
+  data.agents[agentName].model = newModel;
+  data.lastSync = new Date().toISOString();
+  
+  fs.writeFileSync(KILO_META, JSON.stringify(data, null, 2));
+}
+
+// Update kilo.jsonc (manual update required per evolutionary-sync.md rules)
+function updateKiloJsonc(agentName: string, newModel: string): void {
+  const content = fs.readFileSync(path.join(process.cwd(), "kilo.jsonc"), "utf-8");
+  
+  // Simple regex replacement for agent block
+  // Find agent block: "agentName": { ... "model": "old", ... }
+  const agentRegex = new RegExp(`"${agentName}":\\s*{[\\s\\S]*?"model":\\s*"[^"]*"`, 'm');
+  const match = content.match(agentRegex);
+  
+  if (!match) {
+    console.warn(`⚠️ Could not find agent ${agentName} in kilo.jsonc - manual update required`);
+    return;
+  }
+  
+  const oldMatch = match[0];
+  const newMatch = oldMatch.replace(/"model":\s*"[^"]*"/, `"model": "${newModel}"`);
+  const updatedContent = content.replace(oldMatch, newMatch);
+  
+  fs.writeFileSync(path.join(process.cwd(), "kilo.jsonc"), updatedContent);
+}
+
+// Load agent-versions.json
+function loadAgentVersions(): any {
+  const content = fs.readFileSync(AGENT_VERSIONS, "utf-8");
+  return JSON.parse(content);
+}
+
+// Update agent-versions.json with model change
+function updateAgentVersions(
+  agentVersions: any,
+  agentName: string,
+  fromModel: string,
+  toModel: string,
+  reason: string
+): any {
+  const now = new Date().toISOString();
+  
+  if (!agentVersions.agents[agentName]) {
+    agentVersions.agents[agentName] = {
+      current: {},
+      history: [],
+      performance_log: [],
+    };
+  }
+
+  const agent = agentVersions.agents[agentName];
+  
+  // Add history entry
+  agent.history.push({
+    date: now,
+    commit: "model-research-sync",
+    type: "model_change",
+    from: fromModel,
+    to: toModel,
+    reason,
+    source: "research",
+  });
+
+  // Update current model
+  if (!agent.current) agent.current = {};
+  agent.current.model = toModel;
+  agent.current.provider = detectProvider(toModel);
+
+  // Update lastUpdated
+  agentVersions.lastUpdated = now;
+
+  return agentVersions;
+}
+
+// Provider detection
+function detectProvider(model: string): string {
+  if (model.startsWith("ollama-cloud/") || model.startsWith("ollama/")) return "Ollama";
+  if (model.startsWith("openrouter/") || model.includes("openrouter")) return "OpenRouter";
+  if (model.startsWith("groq/")) return "Groq";
+  return "Unknown";
+}
+
+// Apply a single recommendation
+function applyRecommendation(
+  rec: Recommendation,
+  dryRun: boolean,
+  singleAgent?: string
+): { applied: boolean; error?: string; filesModified?: string[] } {
+  if (singleAgent && rec.agent !== singleAgent) {
+    return { applied: false };
+  }
+
+  console.log(`\n🔧 Applying recommendation for ${rec.agent}`);
+  console.log(`   Action: ${rec.action}`);
+  console.log(`   Current: ${rec.current_model}`);
+  console.log(`   Recommended: ${rec.recommended_model}`);
+  console.log(`   Impact: ${rec.impact}`);
+  console.log(`   Rationale: ${rec.rationale}`);
+
+  // Skip if already applied
+  if (rec.applied) {
+    console.log(`   ⏭️ Already applied, skipping`);
+    return { applied: false };
+  }
+
+  if (rec.action === "update_model") {
+    try {
+      // 1. Update capability-index.yaml
+      const capIndexContent = loadCapabilityIndex();
+      const { content: updatedContent, changed: yamlChanged } = replaceModelInYaml(capIndexContent, rec.agent, rec.recommended_model);
+
+      if (!dryRun && yamlChanged) {
+        fs.writeFileSync(CAPABILITY_INDEX, updatedContent);
+        console.log(`   ✅ Updated capability-index.yaml`);
+      } else if (!dryRun) {
+        console.log(`   ⏭️ Skipping capability-index.yaml (no change needed)`);
+      } else {
+        console.log(`   📋 Would update capability-index.yaml`);
+      }
+
+      // Only update other files if YAML was actually changed
+      if (!yamlChanged) {
+        return {
+          applied: false,
+          filesModified: [],
+        };
+      }
+
+      // 2. Update kilo-meta.json (source of truth)
+      if (!dryRun) {
+        updateKiloMeta(rec.agent, rec.recommended_model);
+        console.log(`   ✅ Updated kilo-meta.json`);
+      } else {
+        console.log(`   📋 Would update kilo-meta.json`);
+      }
+
+      // 3. Update agent-versions.json
+      const agentVersions = loadAgentVersions();
+      const updatedVersions = updateAgentVersions(
+        agentVersions,
+        rec.agent,
+        rec.current_model,
+        rec.recommended_model,
+        rec.rationale
+      );
+
+      if (!dryRun) {
+        fs.writeFileSync(AGENT_VERSIONS, JSON.stringify(updatedVersions, null, 2));
+        console.log(`   ✅ Updated agent-versions.json`);
+      } else {
+        console.log(`   📋 Would update agent-versions.json`);
+      }
+
+      // 4. Attempt to update kilo.jsonc (manual verification still required)
+      if (!dryRun) {
+        try {
+          updateKiloJsonc(rec.agent, rec.recommended_model);
+          console.log(`   ✅ Updated kilo.jsonc`);
+        } catch (error: any) {
+          console.warn(`   ⚠️ Could not update kilo.jsonc: ${error.message}`);
+          console.log(`   ⚠️ Manual update required per evolutionary-sync.md rules`);
+        }
+      } else {
+        console.log(`   📋 Would update kilo.jsonc`);
+      }
+
+      return {
+        applied: true,
+        filesModified: [CAPABILITY_INDEX, KILO_META, AGENT_VERSIONS],
+      };
+    } catch (error: any) {
+      return {
+        applied: false,
+        error: error.message,
+      };
+    }
+  } else if (rec.action === "confirm_model") {
+    // Mark as confirmed in agent-versions.json
+    try {
+      const agentVersions = loadAgentVersions();
+      
+      if (agentVersions.agents[rec.agent]) {
+        // Add confirmation history entry
+        agentVersions.agents[rec.agent].history.push({
+          date: new Date().toISOString(),
+          commit: "model-research-confirm",
+          type: "model_change",
+          from: rec.current_model,
+          to: rec.current_model, // same model
+          reason: `Confirmed: ${rec.rationale}`,
+          source: "research",
+        });
+
+        if (!dryRun) {
+          fs.writeFileSync(AGENT_VERSIONS, JSON.stringify(agentVersions, null, 2));
+          console.log(`   ✅ Confirmed current model in agent-versions.json`);
+        } else {
+          console.log(`   📋 Would confirm current model`);
+        }
+
+        return {
+          applied: true,
+          filesModified: [AGENT_VERSIONS],
+        };
+      } else {
+        return {
+          applied: false,
+          error: `Agent ${rec.agent} not found in agent-versions.json`,
+        };
+      }
+    } catch (error: any) {
+      return {
+        applied: false,
+        error: error.message,
+      };
+    }
+  }
+
+  // Unsupported action
+  console.log(`   ⏭️ Unsupported action: ${rec.action}`);
+  return { applied: false };
+}
+
+// Run sync-agents.js --fix
+function runSyncAgentsFix(): boolean {
+  console.log(`\n🔄 Running sync-agents.js --fix...`);
+  
+  const result = spawnSync("node", [SYNC_SCRIPT, "--fix"], {
+    cwd: process.cwd(),
+    encoding: "utf-8",
+    stdio: "inherit",
+  });
+
+  if (result.status !== 0) {
+    console.error(`❌ Sync script failed with exit code ${result.status}`);
+    return false;
+  }
+
+  console.log(`✅ Sync script completed`);
+  return true;
+}
+
+// Run sync-agents.js --check
+function runSyncAgentsCheck(): boolean {
+  console.log(`\n✅ Running sync-agents.js --check...`);
+  
+  const result = spawnSync("node", [SYNC_SCRIPT, "--check"], {
+    cwd: process.cwd(),
+    encoding: "utf-8",
+    stdio: "inherit",
+  });
+
+  if (result.status !== 0) {
+    console.error(`❌ Sync check failed with exit code ${result.status}`);
+    return false;
+  }
+
+  console.log(`✅ Sync check passed`);
+  return true;
+}
+
+// Run build-research-dashboard script
+function runBuildDashboard(): { success: boolean; error?: string } {
+  console.log("\n📊 Rebuilding research dashboard...");
+  
+  try {
+    // Try to import buildResearchDashboard from build-research-dashboard.ts
+    const dashboardScript = path.join(__dirname, "build-research-dashboard.ts");
+    const standaloneScript = path.join(__dirname, "build-standalone.cjs");
+    
+    // Check which build script exists
+    let scriptToRun = "";
+    let args: string[] = [];
+    
+    if (fs.existsSync(dashboardScript)) {
+      scriptToRun = "bun";
+      args = ["run", dashboardScript];
+    } else if (fs.existsSync(standaloneScript)) {
+      scriptToRun = "node";
+      args = [standaloneScript];
+    } else {
+      return { 
+        success: false, 
+        error: "No dashboard build script found (build-research-dashboard.ts or build-standalone.cjs)" 
+      };
+    }
+    
+    const result = spawnSync(scriptToRun, args, {
+      cwd: process.cwd(),
+      encoding: "utf-8",
+      stdio: "inherit",
+      timeout: 30000
+    });
+    
+    if (result.status !== 0) {
+      return { 
+        success: false, 
+        error: result.stderr || `Build script failed with exit code ${result.status}` 
+      };
+    }
+    
+    console.log(result.stdout);
+    console.log("✅ Dashboard rebuilt: agent-evolution/index.standalone.html");
+    return { success: true };
+  } catch (error: any) {
+    return { 
+      success: false, 
+      error: error.message 
+    };
+  }
+}
+
+// Print summary
+function printSummary(summary: ChangeSummary): void {
+  console.log("\n" + "=".repeat(60));
+  console.log("📊 SYNC SUMMARY");
+  console.log("=".repeat(60));
+  
+  console.log(`Total recommendations: ${summary.total_recommendations}`);
+  console.log(`Applied: ${summary.applied}`);
+  console.log(`Confirmed: ${summary.confirmed}`);
+  console.log(`Skipped: ${summary.skipped}`);
+  
+  if (summary.dashboard_rebuilt) {
+    console.log(`Dashboard rebuilt: ✅ Yes`);
+  }
+  
+  if (summary.agents_updated.length > 0) {
+    console.log(`\nAgents updated:`);
+    summary.agents_updated.forEach(agent => console.log(`  - ${agent}`));
+  }
+  
+  if (summary.files_modified.length > 0) {
+    console.log(`\nFiles modified:`);
+    summary.files_modified.forEach(file => console.log(`  - ${file}`));
+  }
+  
+  if (summary.errors.length > 0) {
+    console.log(`\nErrors:`);
+    summary.errors.forEach(error => console.log(`  - ${error}`));
+  }
+  
+  console.log("=".repeat(60));
+}
+
+// Main function
+async function main() {
+  const options = parseArgs();
+  
+  console.log("🧬 Model Research Synchronization");
+  console.log(`   Dry run: ${options.dryRun ? "YES" : "NO"}`);
+  console.log(`   Input: ${options.inputFile}`);
+  if (options.singleAgent) {
+    console.log(`   Single agent: ${options.singleAgent}`);
+  }
+  console.log("");
+
+  // Load research data
+  const researchData = loadResearchData(options.inputFile);
+  
+  if (!validateSchema(researchData)) {
+    console.warn("⚠️ Schema validation issues detected, but continuing...");
+  }
+
+  // Filter recommendations
+  let recommendations = researchData.recommendations;
+  if (options.singleAgent) {
+    recommendations = recommendations.filter(r => r.agent === options.singleAgent);
+    console.log(`Filtered to ${recommendations.length} recommendations for ${options.singleAgent}`);
+  }
+
+  // Initialize summary
+  const summary: ChangeSummary = {
+    total_recommendations: recommendations.length,
+    applied: 0,
+    confirmed: 0,
+    skipped: 0,
+    errors: [],
+    files_modified: [],
+    agents_updated: [],
+    dashboard_rebuilt: false,
+  };
+
+  // Apply recommendations
+  for (const rec of recommendations) {
+    const result = applyRecommendation(rec, options.dryRun, options.singleAgent);
+    
+    if (result.applied) {
+      if (rec.action === "update_model") {
+        summary.applied++;
+        summary.agents_updated.push(rec.agent);
+        if (result.filesModified) {
+          summary.files_modified.push(...result.filesModified);
+        }
+      } else if (rec.action === "confirm_model") {
+        summary.confirmed++;
+      }
+    } else {
+      if (result.error) {
+        summary.errors.push(`${rec.agent}: ${result.error}`);
+      } else {
+        summary.skipped++;
+      }
+    }
+  }
+
+  // Remove duplicate files from files_modified
+  summary.files_modified = [...new Set(summary.files_modified)];
+
+  // Run sync-agents.js if we made changes (and not dry run)
+  if (summary.applied > 0 && !options.dryRun) {
+    console.log(`\n📦 Propagating changes to all agent files...`);
+    const syncOk = runSyncAgentsFix();
+    
+    if (syncOk) {
+      console.log(`\n✅ Validating changes...`);
+      const checkOk = runSyncAgentsCheck();
+      
+      if (checkOk) {
+        // Rebuild research dashboard
+        const buildResult = runBuildDashboard();
+        if (buildResult.success) {
+          console.log("✅ Dashboard rebuilt: agent-evolution/index.standalone.html");
+          summary.dashboard_rebuilt = true;
+        } else {
+          console.warn(`⚠️ Dashboard rebuild failed: ${buildResult.error}`);
+          summary.errors.push(`Dashboard rebuild failed: ${buildResult.error}`);
+        }
+      } else {
+        summary.errors.push("Sync check failed after applying changes");
+      }
+    } else {
+      summary.errors.push("Sync fix script failed");
+    }
+  }
+
+  // Print summary
+  printSummary(summary);
+
+  // Exit with error if any errors occurred
+  if (summary.errors.length > 0) {
+    console.error(`\n❌ Sync completed with ${summary.errors.length} errors`);
+    process.exit(1);
+  } else if (summary.applied === 0 && summary.confirmed === 0) {
+    console.warn(`\n⚠️ No changes applied`);
+  } else {
+    console.log(`\n🎉 Sync completed successfully!`);
+  }
+}
+
+// Run the script
+main().catch((error) => {
+  console.error("Fatal error:", error);
+  process.exit(1);
+});