genoma-evolution 1.0.0

package/backend/mcp_server.py

@@ -0,0 +1,250 @@
+"""
+Genoma MCP Server — 4 telemetry ingestion tools for AI agents.
+
+Supports stdio transport.
+Run:
+    python3 -m backend.mcp_server
+"""
+
+from __future__ import annotations
+
+import asyncio
+from typing import Any
+
+from mcp.server import Server
+from mcp.server.stdio import stdio_server
+from mcp.types import (
+    TextContent,
+    Tool,
+    CallToolResult,
+    ListToolsResult,
+)
+
+from backend.promethean.models import CanonicalRun, TraceRecord
+from backend.promethean.trace_ingestion import get_ingestor
+from backend.storage.run_store import RunStore
+
+server = Server(
+    name="genoma",
+    instructions=(
+        "Push telemetry from your running session to Genoma. "
+        "Use ingest_run for full canonical run events, ingest_trace for lightweight "
+        "trace records, query_runs to retrieve past runs, and get_agent_stats for "
+        "per-agent performance statistics."
+    ),
+)
+
+
+@server.list_tools()
+async def list_tools() -> ListToolsResult:
+    """Advertise all 4 MCP tools."""
+    return ListToolsResult(
+        tools=[
+            Tool(
+                name="ingest_run",
+                description="Push a canonical run event to Genoma's telemetry store.",
+                inputSchema={
+                    "type": "object",
+                    "properties": {
+                        "run_id": {"type": "string"},
+                        "agent_name": {"type": "string"},
+                        "started_at": {"type": "string"},
+                        "task_name": {"type": "string"},
+                        "outcome": {"type": "string"},
+                        "collector": {"type": "string", "default": "mcp-native"},
+                        "agent_version": {"type": "string"},
+                        "provider": {"type": "string"},
+                        "model": {"type": "string"},
+                        "repo": {"type": "string"},
+                        "session_id": {"type": "string"},
+                        "ended_at": {"type": "string"},
+                        "tool_calls": {"type": "array"},
+                        "files_touched": {"type": "array"},
+                        "metrics": {"type": "object"},
+                        "errors": {"type": "array"},
+                        "context": {"type": "object"},
+                        "resolution": {"type": "string"},
+                    },
+                    "required": ["run_id", "agent_name", "started_at", "task_name", "outcome"],
+                },
+            ),
+            Tool(
+                name="ingest_trace",
+                description="Push a lightweight TraceRecord to Genoma.",
+                inputSchema={
+                    "type": "object",
+                    "properties": {
+                        "agent": {"type": "string"},
+                        "agent_version": {"type": "string"},
+                        "timestamp": {"type": "string"},
+                        "task": {"type": "string"},
+                        "outcome": {"type": "string"},
+                        "error_signature": {"type": "string"},
+                        "context": {"type": "object"},
+                        "resolution": {"type": "string"},
+                    },
+                    "required": ["agent", "agent_version", "timestamp", "task", "outcome"],
+                },
+            ),
+            Tool(
+                name="query_runs",
+                description="Query past runs from Genoma's telemetry store.",
+                inputSchema={
+                    "type": "object",
+                    "properties": {
+                        "agent_name": {"type": "string"},
+                        "outcome": {"type": "string"},
+                        "repo": {"type": "string"},
+                        "since": {"type": "string"},
+                        "until": {"type": "string"},
+                        "limit": {"type": "integer", "default": 20},
+                    },
+                },
+            ),
+            Tool(
+                name="get_agent_stats",
+                description="Get performance statistics for one or all agents.",
+                inputSchema={
+                    "type": "object",
+                    "properties": {
+                        "agent_name": {"type": "string"},
+                    },
+                },
+            ),
+        ]
+    )
+
+
+@server.call_tool()
+async def handle_tool_call(tool_name: str, arguments: dict[str, Any]) -> CallToolResult:
+    """Route tool calls to handlers."""
+    if tool_name == "ingest_run":
+        return await ingest_run(arguments)
+    elif tool_name == "ingest_trace":
+        return await ingest_trace(arguments)
+    elif tool_name == "query_runs":
+        return await query_runs(arguments)
+    elif tool_name == "get_agent_stats":
+        return await get_agent_stats(arguments)
+    else:
+        raise ValueError(f"Unknown tool: {tool_name}")
+
+
+async def ingest_run(args: dict[str, Any]) -> CallToolResult:
+    """Push a canonical run event to Genoma's telemetry store."""
+    data = {
+        "run_id": args["run_id"],
+        "agent_name": args["agent_name"],
+        "collector": args.get("collector", "mcp-native"),
+        "started_at": args["started_at"],
+        "task_name": args["task_name"],
+        "outcome": args["outcome"],
+        "agent_version": args.get("agent_version"),
+        "provider": args.get("provider"),
+        "model": args.get("model"),
+        "repo": args.get("repo"),
+        "session_id": args.get("session_id"),
+        "ended_at": args.get("ended_at"),
+        "tool_calls": args.get("tool_calls", []),
+        "files_touched": args.get("files_touched", []),
+        "metrics": args.get("metrics"),
+        "errors": args.get("errors", []),
+        "context": args.get("context", {}),
+        "resolution": args.get("resolution"),
+    }
+    run = CanonicalRun.from_dict(data)
+    is_new = RunStore().upsert_run(run)
+    result = {
+        "status": "inserted" if is_new else "updated",
+        "run_id": run.run_id,
+    }
+    return CallToolResult(
+        content=[TextContent(type="text", text=str(result))],
+        isError=False,
+    )
+
+
+async def ingest_trace(args: dict[str, Any]) -> CallToolResult:
+    """Push a lightweight TraceRecord to Genoma."""
+    trace = TraceRecord(
+        agent=args["agent"],
+        agent_version=args["agent_version"],
+        timestamp=args["timestamp"],
+        task=args["task"],
+        outcome=args["outcome"],
+        error_signature=args.get("error_signature"),
+        context=args.get("context", {}),
+        resolution=args.get("resolution"),
+    )
+    get_ingestor().ingest(trace)
+    canonical = trace.to_canonical()
+    RunStore().upsert_run(canonical)
+    result = {
+        "status": "ingested",
+        "trace_id": trace.trace_id,
+        "run_id": canonical.run_id,
+    }
+    return CallToolResult(
+        content=[TextContent(type="text", text=str(result))],
+        isError=False,
+    )
+
+
+async def query_runs(args: dict[str, Any]) -> CallToolResult:
+    """Query past runs from Genoma's telemetry store."""
+    limit = min(args.get("limit", 20), 100)
+    runs = RunStore().list_runs(
+        agent_name=args.get("agent_name"),
+        outcome=args.get("outcome"),
+        repo=args.get("repo"),
+        since=args.get("since"),
+        until=args.get("until"),
+        limit=limit,
+        offset=0,
+    )
+    lightweight = [
+        {
+            "run_id": r.run_id,
+            "agent_name": r.agent_name,
+            "agent_version": r.agent_version,
+            "collector": r.collector,
+            "started_at": r.started_at,
+            "ended_at": r.ended_at,
+            "task_name": r.task_name,
+            "outcome": r.outcome,
+            "provider": r.provider,
+            "model": r.model,
+            "repo": r.repo,
+            "session_id": r.session_id,
+            "resolution": r.resolution,
+        }
+        for r in runs
+    ]
+    result = {"runs": lightweight, "count": len(lightweight)}
+    return CallToolResult(
+        content=[TextContent(type="text", text=str(result))],
+        isError=False,
+    )
+
+
+async def get_agent_stats(args: dict[str, Any]) -> CallToolResult:
+    """Get performance statistics for one or all agents."""
+    summary = RunStore().get_agent_summary()
+    agent_name = args.get("agent_name")
+    if agent_name:
+        summary = [s for s in summary if s.get("agent_name") == agent_name]
+    result = {"agents": summary}
+    return CallToolResult(
+        content=[TextContent(type="text", text=str(result))],
+        isError=False,
+    )
+
+
+async def main() -> None:
+    """Start MCP server with stdio transport."""
+    async with stdio_server(server) as (read_stream, write_stream):
+        await read_stream.read()
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

package/backend/promethean/__init__.py

@@ -0,0 +1,24 @@
+"""
+Promethean Cycle — Ciclo de Hiper-Evolución Autónoma del Dashboard
+
+Módulos:
+  ① trace_ingestion  — PERCIBE: ingesta de trazas multi-agente
+  ② gepa_strategist   — DIAGNOSTICA + FORMULA: GEPA strategic layer
+  ③ dspy_compiler     — COMPILA: DSPy BetterTogether compilation
+  ④ delta_validator   — VALIDA: holdout validation + delta
+  ⑤ skill_deployer    — DESPLIEGA: auto-registro en skill registry
+  ⑥ cycle_orchestrator — OBSERVA: full 7-phase orchestrator
+
+El dashboard deja de ser espejo y se convierte en oráculo.
+"""
+
+__version__ = "1.0.0"
+__all__ = [
+    "models",
+    "trace_ingestion",
+    "gepa_strategist",
+    "dspy_compiler",
+    "delta_validator",
+    "skill_deployer",
+    "cycle_orchestrator",
+]

package/backend/promethean/cycle_orchestrator.py

@@ -0,0 +1,270 @@
+"""
+⑦ OBSERVA — Promethean Cycle Orchestrator
+
+Orchestrates the full 7-phase autonomous evolution cycle:
+  ① PERCIBE → ② DIAGNOSTICA → ③ FORMULA → ④ COMPILA → ⑤ VALIDA → ⑥ DESPLIEGA → ⑦ OBSERVA
+
+The orchestrator is the "heartbeat" of the living dashboard.
+It can run on-demand or as a continuous background loop.
+"""
+
+from __future__ import annotations
+import asyncio
+import json
+import sys
+from datetime import datetime
+from pathlib import Path
+from typing import Optional
+
+from .models import (
+    CycleState, CyclePhase, SkillGenesisPacket, CompilationResult, MetricSnapshot,
+    TraceRecord,
+)
+from .trace_ingestion import get_ingestor
+from .gepa_strategist import get_strategist
+from .dspy_compiler import get_compiler
+from .delta_validator import get_validator
+from .skill_deployer import get_deployer
+
+CYCLES_DIR = Path.home() / ".hermes" / "traces" / "cycles"
+
+
+class PrometheanOrchestrator:
+    """Full 7-phase autonomous evolution orchestrator."""
+
+    def __init__(self, python_bin: str = sys.executable):
+        self.python_bin = python_bin
+        self.ingestor = get_ingestor()
+        self.strategist = get_strategist()
+        self.compiler = get_compiler(python_bin)
+        self.validator = get_validator(python_bin)
+        self.deployer = get_deployer()
+        CYCLES_DIR.mkdir(parents=True, exist_ok=True)
+
+    # ── Full Cycle Execution ────────────────────────────────────────
+    async def run_full_cycle(
+        self,
+        min_anomaly_occurrences: int = 3,
+        anomaly_days: int = 7,
+        auto_deploy: bool = True,
+    ) -> dict:
+        """Execute the complete Promethean Cycle.
+
+        Compilation runs in background to avoid blocking the API.
+        Returns a summary dictionary suitable for API response.
+        """
+        import concurrent.futures
+        state = CycleState()
+        events: list[dict] = []  # Progress events for streaming
+
+        try:
+            # ── ① PERCIBE ──────────────────────────────────────────
+            state.advance(CyclePhase.PERCIBE)
+            state.traces_ingested = self.ingestor.get_trace_count()
+            anomalies = self.ingestor.get_recent_failures(
+                days=anomaly_days,
+                min_occurrences=min_anomaly_occurrences,
+            )
+            state.anomalies_detected = len(anomalies)
+            events.append({
+                "phase": "perceive",
+                "traces_ingested": state.traces_ingested,
+                "anomalies_detected": len(anomalies),
+            })
+
+            if not anomalies:
+                state.advance(CyclePhase.OBSERVA)
+                return state.summary()
+
+            # ── ②③ DIAGNOSTICA + FORMULA ───────────────────────────
+            state.advance(CyclePhase.DIAGNOSTICA)
+            gaps = self.strategist.diagnose(anomalies)
+            actionable = [g for g in gaps if g["recommended_action"] == "compile_skill"]
+            events.append({
+                "phase": "diagnose",
+                "gaps_found": len(gaps),
+                "actionable_gaps": len(actionable),
+            })
+
+            if not actionable:
+                state.advance(CyclePhase.OBSERVA)
+                return state.summary()
+
+            state.advance(CyclePhase.FORMULA)
+            packets = self.strategist.formulate_all(actionable, state)
+            events.append({
+                "phase": "formulate",
+                "genesis_packets": len(packets),
+                "packet_ids": [p.packet_id for p in packets],
+            })
+
+            # ── ④ COMPILA: Run in thread pool to avoid blocking ─────
+            state.advance(CyclePhase.COMPILA)
+            all_compilations: list[CompilationResult] = []
+            
+            def compile_packet(packet):
+                return self.compiler.compile_with_retry(packet)
+            
+            with concurrent.futures.ThreadPoolExecutor(max_workers=2) as executor:
+                futures = [executor.submit(compile_packet, p) for p in packets]
+                for future in concurrent.futures.as_completed(futures, timeout=120):
+                    try:
+                        comp = future.result(timeout=60)
+                        all_compilations.append(comp)
+                        state.compilation_results.append(comp)
+                    except concurrent.futures.TimeoutError:
+                        all_compilations.append(CompilationResult(
+                            packet_id=packets[0].packet_id if packets else "unknown",
+                            success=False, skill_name="", skill_path="",
+                            iterations=0, best_score=0.0, delta=0.0,
+                            threshold_met=False, attempt=1,
+                            error="Compilation timed out (60s limit)",
+                        ))
+            
+            events.append({
+                "phase": "compile",
+                "compiled": len(all_compilations),
+                "successful": sum(1 for c in all_compilations if c.success),
+            })
+
+            # ── ⑤ VALIDA: Delta validation ────────────────────────────
+            state.advance(CyclePhase.VALIDA)
+            all_snapshots: list[MetricSnapshot] = []
+            for packet, comp in zip(packets, all_compilations):
+                snapshot = self.validator.validate(packet, comp)
+                all_snapshots.append(snapshot)
+            passed = [s for s in all_snapshots if s.passed]
+            events.append({
+                "phase": "validate",
+                "validated": len(all_snapshots),
+                "passed": len(passed),
+                "deltas": [s.deltas for s in passed],
+            })
+
+            # ── ⑥ DESPLIEGA: Auto-register skills ─────────────────────
+            state.advance(CyclePhase.DESPLIEGA)
+            if auto_deploy:
+                deployments = []
+                for packet, comp, snap in zip(packets, all_compilations, all_snapshots):
+                    if snap.passed:
+                        result = self.deployer.deploy(packet, comp, snap)
+                        if result.get("deployed"):
+                            state.skills_deployed += 1
+                        deployments.append(result)
+                events.append({
+                    "phase": "deploy",
+                    "deployed": state.skills_deployed,
+                    "skill_names": [d.get("skill_name") for d in deployments if d.get("deployed")],
+                })
+
+            # ── ⑦ OBSERVA: Close cycle, establish new baseline ─────────
+            state.advance(CyclePhase.OBSERVA)
+            events.append({
+                "phase": "observe",
+                "new_baseline": f"{state.skills_deployed} new skills deployed",
+            })
+
+        except Exception as e:
+            state.errors.append(str(e))
+            events.append({"phase": "error", "error": str(e)})
+
+        # Save cycle state to disk
+        self._save_cycle_state(state, events)
+
+        return {
+            **state.summary(),
+            "events": events,
+            "dspy_available": self.compiler.is_available,
+            "dspy_version": self.compiler.dspy_version,
+        }
+
+    def run_full_cycle_sync(self, **kwargs) -> dict:
+        """Synchronous wrapper for run_full_cycle."""
+        return asyncio.run(self.run_full_cycle(**kwargs))
+
+    # ── Single Phase Execution (for testing/stepping) ───────────────
+    def run_perceive(self, days: int = 7, min_occ: int = 3) -> dict:
+        """Run only the PERCIBE phase."""
+        traces = self.ingestor.get_trace_count()
+        anomalies = self.ingestor.get_recent_failures(days=days, min_occurrences=min_occ)
+        agent_health = self.ingestor.get_agent_health()
+        return {
+            "phase": "perceive",
+            "traces_ingested": traces,
+            "anomalies": anomalies,
+            "agent_health": agent_health,
+        }
+
+    def run_diagnose(self, days: int = 7, min_occ: int = 3) -> dict:
+        """Run PERCIBE + DIAGNOSTICA phases."""
+        anomalies = self.ingestor.get_recent_failures(days=days, min_occurrences=min_occ)
+        gaps = self.strategist.diagnose(anomalies)
+        return {
+            "phase": "diagnose",
+            "anomalies_found": len(anomalies),
+            "gaps": gaps,
+        }
+
+    # ── Helpers ─────────────────────────────────────────────────────
+    def _save_cycle_state(self, state: CycleState, events: list[dict]):
+        """Persist cycle state to disk."""
+        cycle_data = {
+            **state.summary(),
+            "events": events,
+            "packets": [p.to_dict() for p in state.genesis_packets],
+            "compilations": [
+                {
+                    "packet_id": c.packet_id,
+                    "success": c.success,
+                    "skill_name": c.skill_name,
+                    "delta": c.delta,
+                    "attempt": c.attempt,
+                    "error": c.error,
+                }
+                for c in state.compilation_results
+            ],
+        }
+        cycle_path = CYCLES_DIR / f"cycle_{state.cycle_id}.json"
+        cycle_path.write_text(json.dumps(cycle_data, indent=2, default=str))
+
+    def get_cycle_history(self, limit: int = 10) -> list[dict]:
+        """Get recent cycle execution history."""
+        if not CYCLES_DIR.exists():
+            return []
+
+        cycles = []
+        for f in sorted(CYCLES_DIR.glob("cycle_*.json"), reverse=True)[:limit]:
+            try:
+                cycles.append(json.loads(f.read_text()))
+            except Exception:
+                continue
+        return cycles
+
+    # ── Continuous Mode ─────────────────────────────────────────────
+    async def continuous_loop(self, interval_minutes: int = 30):
+        """Run the Promethean Cycle continuously every N minutes."""
+        print(f"🔥 Promethean Cycle started. Running every {interval_minutes} min.")
+        while True:
+            try:
+                print(f"\n{'='*60}")
+                print(f"🔄 CYCLE START — {datetime.now().isoformat()}")
+                result = await self.run_full_cycle()
+                print(f"✅ CYCLE COMPLETE — {result['skills_deployed']} skills deployed")
+                print(f"   Traces: {result['traces_ingested']} | Anomalies: {result['anomalies_detected']}")
+                print(f"   DSPy: {'✅ v' + (self.compiler.dspy_version or '?') if self.compiler.is_available else '❌'}")
+                print(f"{'='*60}\n")
+            except Exception as e:
+                print(f"❌ CYCLE ERROR: {e}")
+
+            await asyncio.sleep(interval_minutes * 60)
+
+
+# ── Singleton ───────────────────────────────────────────────────────
+_orchestrator: Optional[PrometheanOrchestrator] = None
+
+
+def get_orchestrator(python_bin: str = sys.executable) -> PrometheanOrchestrator:
+    global _orchestrator
+    if _orchestrator is None:
+        _orchestrator = PrometheanOrchestrator(python_bin)
+    return _orchestrator