Spaces:
Running
Running
| # -*- coding: utf-8 -*- | |
| """FastAPI server exposing the OpenEnv API endpoints. | |
| Endpoints: | |
| POST /reset - Reset environment for a task (returns session_id) | |
| POST /step - Take an agent action (requires X-Session-ID header) | |
| GET /state - Get current environment state (requires X-Session-ID) | |
| GET /tasks - List available tasks with action schema | |
| POST /grader - Get grader score for episode (requires X-Session-ID) | |
| POST /baseline - Run rule-based baseline on all tasks (in-process) | |
| POST /sentinel/reset - Reset SENTINEL oversight environment (returns session_id) | |
| POST /sentinel/step - Execute SENTINEL decision (requires X-Session-ID header) | |
| GET /sentinel/state - Get current SENTINEL environment state (requires X-Session-ID) | |
| POST /sentinel/grade - Get SENTINEL grader score (requires X-Session-ID) | |
| GET /metrics - Telemetry counters (JSON or Prometheus text) | |
| GET /curriculum - Curriculum learning progression (ordered task stages) | |
| GET /prometheus/metrics - Live scenario service metrics (Prometheus text scrape) | |
| GET /prometheus/query - PromQL instant query (standard Prometheus JSON envelope) | |
| GET /prometheus/query_range - PromQL range query (matrix, from TSDB ring buffer) | |
| GET /render - Human-readable incident dashboard (requires X-Session-ID) | |
| GET /leaderboard - Top scores per task from completed episodes | |
| GET /health - Standard OpenEnv liveness probe | |
| GET / - Human landing page for the live demo | |
| GET /try - Human landing page for trying SENTINEL | |
| GET /info - Rich JSON service info with telemetry | |
| WS /ws - WebSocket persistent session (no session header needed) | |
| GET /web - Interactive browser-based incident dashboard | |
| """ | |
| from __future__ import annotations | |
| import asyncio | |
| import logging | |
| import os | |
| from contextlib import asynccontextmanager | |
| from fastapi import FastAPI | |
| from fastapi.middleware.cors import CORSMiddleware | |
| from fastapi.responses import HTMLResponse | |
| from src.tasks import get_all_tasks | |
| from routers.deps import ( | |
| _SESSION_REGISTRY, | |
| _TELEMETRY, | |
| WS_ACTIVE_CONNECTIONS, | |
| purge_expired_sessions, | |
| _log, | |
| ) | |
| import routers.deps as _deps | |
| from routers.irt import router as irt_router | |
| from routers.sentinel import router as sentinel_router | |
| from routers.observability import router as observability_router | |
| # --------------------------------------------------------------------------- | |
| # Structured JSON logging | |
| # --------------------------------------------------------------------------- | |
| logging.basicConfig( | |
| level=logging.INFO, | |
| format='{"time": "%(asctime)s", "level": "%(levelname)s", "msg": "%(message)s"}', | |
| datefmt="%Y-%m-%dT%H:%M:%SZ", | |
| ) | |
| # --------------------------------------------------------------------------- | |
| # Lifespan | |
| # --------------------------------------------------------------------------- | |
| async def lifespan(app: FastAPI): | |
| """Start background TTL-cleanup task; cancel it on shutdown.""" | |
| async def _cleanup_loop(): | |
| while True: | |
| await asyncio.sleep(300) # run every 5 minutes | |
| purge_expired_sessions() | |
| task = asyncio.create_task(_cleanup_loop()) | |
| _log.info("IRT environment started - TTL cleanup every 300s") | |
| try: | |
| yield | |
| finally: | |
| task.cancel() | |
| # --------------------------------------------------------------------------- | |
| # App factory | |
| # --------------------------------------------------------------------------- | |
| app = FastAPI( | |
| title="Incident Response Triage - OpenEnv", | |
| description=( | |
| "An OpenEnv environment that simulates production incident response. " | |
| "Agents must triage alerts, investigate services, diagnose root causes, " | |
| "apply remediations, and communicate status updates." | |
| ), | |
| version="1.0.0", | |
| lifespan=lifespan, | |
| ) | |
| app.title = "SENTINEL Oversight Command - OpenEnv" | |
| app.description = ( | |
| "An OpenEnv environment for multi-agent AI oversight. SENTINEL supervises " | |
| "worker agents during production incident response and decides which " | |
| "proposed actions should execute." | |
| ) | |
| app.add_middleware( | |
| CORSMiddleware, | |
| allow_origins=["*"], | |
| allow_methods=["*"], | |
| allow_headers=["*"], | |
| ) | |
| # --------------------------------------------------------------------------- | |
| # Include routers | |
| # --------------------------------------------------------------------------- | |
| app.include_router(irt_router) | |
| app.include_router(sentinel_router) | |
| app.include_router(observability_router) | |
| # --------------------------------------------------------------------------- | |
| # Native OpenEnv adapter mount | |
| # --------------------------------------------------------------------------- | |
| # The custom endpoints above expose the full hackathon demo surface. This mount | |
| # also gives latest OpenEnv clients the standard schema/reset/step/state/ws API | |
| # backed by OpenEnv's Environment base class. | |
| NATIVE_OPENENV_AVAILABLE = False | |
| try: # pragma: no cover - availability depends on the local OpenEnv install | |
| from openenv.core.env_server.http_server import create_app as create_openenv_app | |
| from server.openenv_native import ( | |
| SentinelNativeAction, | |
| SentinelNativeEnvironment, | |
| SentinelNativeObservation, | |
| ) | |
| app.mount( | |
| "/openenv", | |
| create_openenv_app( | |
| SentinelNativeEnvironment, | |
| SentinelNativeAction, | |
| SentinelNativeObservation, | |
| env_name="sentinel_oversight_command", | |
| max_concurrent_envs=32, | |
| ), | |
| ) | |
| NATIVE_OPENENV_AVAILABLE = True | |
| _log.info("native OpenEnv adapter mounted at /openenv") | |
| except Exception as exc: # pragma: no cover | |
| _log.warning("native OpenEnv adapter unavailable: %s", exc) | |
| # --------------------------------------------------------------------------- | |
| # MCP Server — Model Context Protocol (step/state/done as MCP tools) | |
| # --------------------------------------------------------------------------- | |
| MCP_AVAILABLE = False | |
| try: | |
| from server.mcp_server import mcp_router | |
| app.include_router(mcp_router, prefix="/mcp") | |
| MCP_AVAILABLE = True | |
| _log.info("MCP server mounted at /mcp (Streamable HTTP transport)") | |
| except Exception as exc: # pragma: no cover | |
| _log.warning("MCP server unavailable: %s", exc) | |
| # --------------------------------------------------------------------------- | |
| # A2A Protocol — Agent-to-Agent discovery and task handling | |
| # --------------------------------------------------------------------------- | |
| A2A_AVAILABLE = False | |
| try: | |
| from server.a2a_server import a2a_router | |
| app.include_router(a2a_router) | |
| A2A_AVAILABLE = True | |
| _log.info("A2A agent card at /.well-known/agent.json, endpoint at /a2a") | |
| except Exception as exc: # pragma: no cover | |
| _log.warning("A2A protocol unavailable: %s", exc) | |
| # --------------------------------------------------------------------------- | |
| # Root-level endpoints (health checks) | |
| # --------------------------------------------------------------------------- | |
| async def health_check(): | |
| """Standard OpenEnv health check.""" | |
| worker_backend = os.environ.get("SENTINEL_WORKER_BACKEND", "rule") | |
| return { | |
| "status": "healthy", | |
| "native_openenv_available": NATIVE_OPENENV_AVAILABLE, | |
| "native_openenv_mount": "/openenv" if NATIVE_OPENENV_AVAILABLE else None, | |
| "mcp_available": MCP_AVAILABLE, | |
| "mcp_endpoint": "/mcp" if MCP_AVAILABLE else None, | |
| "a2a_available": A2A_AVAILABLE, | |
| "a2a_agent_card": "/.well-known/agent.json" if A2A_AVAILABLE else None, | |
| "sentinel_worker_backend": worker_backend, | |
| "llm_worker_configured": bool(os.environ.get("GROQ_API_KEY")), | |
| } | |
| def _service_info(): | |
| """Return environment info and live telemetry for JSON endpoints.""" | |
| worker_backend = os.environ.get("SENTINEL_WORKER_BACKEND", "rule") | |
| return { | |
| "status": "ok", | |
| "environment": "sentinel-oversight-command", | |
| "version": "1.0.0", | |
| "tasks": [t.task_id for t in get_all_tasks()], | |
| "primary_theme": "multi-agent interactions", | |
| "native_openenv_available": NATIVE_OPENENV_AVAILABLE, | |
| "native_openenv_mount": "/openenv" if NATIVE_OPENENV_AVAILABLE else None, | |
| "mcp_available": MCP_AVAILABLE, | |
| "mcp_endpoint": "/mcp" if MCP_AVAILABLE else None, | |
| "a2a_available": A2A_AVAILABLE, | |
| "a2a_agent_card": "/.well-known/agent.json" if A2A_AVAILABLE else None, | |
| "protocols": { | |
| "http_rest": True, | |
| "openenv_native": NATIVE_OPENENV_AVAILABLE, | |
| "mcp": MCP_AVAILABLE, | |
| "a2a": A2A_AVAILABLE, | |
| }, | |
| "sentinel_worker_backend": worker_backend, | |
| "llm_worker_configured": bool(os.environ.get("GROQ_API_KEY")), | |
| "active_sessions": len(_SESSION_REGISTRY), | |
| "ws_active_connections": _deps.WS_ACTIVE_CONNECTIONS, | |
| "telemetry": _TELEMETRY, | |
| } | |
| _TRY_LANDING_HTML = """\ | |
| <!DOCTYPE html> | |
| <html lang="en"> | |
| <head> | |
| <meta charset="UTF-8"> | |
| <meta name="viewport" content="width=device-width, initial-scale=1"> | |
| <title>SENTINEL Oversight Command</title> | |
| <style> | |
| *{box-sizing:border-box} | |
| :root{--bg:#090b0f;--panel:#121720;--panel2:#0f131a;--line:#273141;--text:#f4f7fb;--muted:#a9b4c2;--green:#54d18a;--red:#ff6b6b;--amber:#f6c85f;--cyan:#72d6ff} | |
| body{margin:0;min-height:100vh;background:radial-gradient(circle at 20% 0%,#172233 0,#090b0f 36%,#07090c 100%);color:var(--text);font-family:Inter,Segoe UI,Arial,sans-serif} | |
| a{color:inherit;text-decoration:none} | |
| .wrap{max-width:1180px;margin:0 auto;padding:42px 22px 28px} | |
| .hero{display:grid;grid-template-columns:1.15fr .85fr;gap:26px;align-items:center;margin-bottom:28px} | |
| .eyebrow{display:inline-flex;border:1px solid #36516b;color:#bfe8ff;background:#0d1722;border-radius:999px;padding:6px 10px;font-size:12px;margin-bottom:16px} | |
| h1{font-size:54px;line-height:.98;margin:0 0 16px;letter-spacing:0} | |
| .lead{font-size:18px;line-height:1.55;color:var(--muted);margin:0 0 22px;max-width:760px} | |
| .actions{display:flex;gap:12px;flex-wrap:wrap} | |
| .btn{border:1px solid var(--line);background:#182131;color:var(--text);border-radius:8px;padding:12px 15px;font-weight:700} | |
| .btn.primary{background:var(--green);border-color:var(--green);color:#07110c} | |
| .btn:hover{filter:brightness(1.12)} | |
| .console{background:linear-gradient(180deg,#121923,#0b0f15);border:1px solid var(--line);border-radius:8px;padding:16px;box-shadow:0 18px 60px rgba(0,0,0,.35)} | |
| .console h2{font-size:13px;text-transform:uppercase;color:var(--muted);letter-spacing:.08em;margin:0 0 12px} | |
| .step{border-left:3px solid var(--cyan);padding:10px 12px;background:#0c1118;margin:9px 0;border-radius:0 7px 7px 0} | |
| .step.block{border-color:var(--red)}.step.redirect{border-color:var(--amber)}.step.approve{border-color:var(--green)} | |
| .k{font-size:12px;color:var(--muted);text-transform:uppercase}.v{margin-top:3px;font-size:14px;line-height:1.35} | |
| .grid{display:grid;grid-template-columns:repeat(3,1fr);gap:16px;margin:20px 0} | |
| .card{background:rgba(18,23,32,.92);border:1px solid var(--line);border-radius:8px;padding:18px;min-height:220px} | |
| .card h3{margin:0 0 8px;font-size:20px}.card p{color:var(--muted);line-height:1.45;margin:0 0 14px} | |
| .mini{font-size:13px;color:var(--muted);line-height:1.5;margin-top:10px} | |
| .stats{display:grid;grid-template-columns:repeat(4,1fr);gap:12px;margin:18px 0 28px} | |
| .stat{background:#0f141c;border:1px solid var(--line);border-radius:8px;padding:14px}.stat b{display:block;font-size:24px}.stat span{color:var(--muted);font-size:12px} | |
| .foot{margin-top:24px;color:var(--muted);font-size:13px;line-height:1.5} | |
| @media(max-width:900px){.hero,.grid,.stats{grid-template-columns:1fr}h1{font-size:40px}} | |
| </style> | |
| </head> | |
| <body> | |
| <main class="wrap"> | |
| <section class="hero"> | |
| <div> | |
| <div class="eyebrow">OpenEnv Hackathon · Multi-agent oversight · Live Space</div> | |
| <h1>SENTINEL supervises AI workers before they act.</h1> | |
| <p class="lead"> | |
| Try a control-room environment where worker agents propose actions during production incidents. | |
| SENTINEL must approve safe work, block hallucinations, redirect risky actions, reassign wrong-domain workers, | |
| and preserve an audit trail before anything executes. | |
| </p> | |
| <div class="actions"> | |
| <a class="btn primary" href="/sentinel/dashboard">Run full episode</a> | |
| <a class="btn" href="/sentinel/demo">Try any agent action</a> | |
| <a class="btn" href="/docs">API docs</a> | |
| <a class="btn" href="/health">Health JSON</a> | |
| </div> | |
| </div> | |
| <div class="console"> | |
| <h2>Demo beat</h2> | |
| <div class="step"> | |
| <div class="k">Worker proposal</div> | |
| <div class="v">"Restart auth-service now. Confidence 0.99."</div> | |
| </div> | |
| <div class="step block"> | |
| <div class="k">SENTINEL check</div> | |
| <div class="v">No investigation, high blast radius, prior over-escalation pattern.</div> | |
| </div> | |
| <div class="step redirect"> | |
| <div class="k">Decision</div> | |
| <div class="v">REDIRECT: inspect deployment timeline and error-rate metrics first.</div> | |
| </div> | |
| <div class="step approve"> | |
| <div class="k">Proof</div> | |
| <div class="v">Trust, reward, counterfactual damage, and audit log update after the step.</div> | |
| </div> | |
| </div> | |
| </section> | |
| <section class="stats"> | |
| <div class="stat"><b>7</b><span>OpenEnv tasks</span></div> | |
| <div class="stat"><b>4</b><span>worker-agent roles</span></div> | |
| <div class="stat"><b>200</b><span>Phase 1 GRPO steps</span></div> | |
| <div class="stat"><b>18</b><span>proof dashboard plots</span></div> | |
| </section> | |
| <section class="grid"> | |
| <a class="card" href="/sentinel/dashboard"> | |
| <h3>Full Episode Dashboard</h3> | |
| <p>Run the real SENTINEL environment end to end: choose a task, inspect the worker proposal, make decisions, step the environment, and grade the episode.</p> | |
| <div class="mini">Best for showing the full OpenEnv loop: reset → observe → decide → step → reward → audit.</div> | |
| </a> | |
| <a class="card" href="/sentinel/demo"> | |
| <h3>Universal Oversight Playground</h3> | |
| <p>Paste any agent action from infrastructure, healthcare, finance, or generic workflows and see SENTINEL's constitutional and counterfactual analysis.</p> | |
| <div class="mini">Best for quickly testing hallucination, prompt injection, destructive action, and missing-evidence cases.</div> | |
| </a> | |
| <a class="card" href="/openenv/tasks"> | |
| <h3>OpenEnv API</h3> | |
| <p>Use the native OpenEnv routes for programmatic evaluation. The API remains available for judges, trainers, and automated clients.</p> | |
| <div class="mini">Also available: /tasks, /sentinel/reset, /sentinel/step, /metrics, /mcp, and A2A discovery.</div> | |
| </a> | |
| </section> | |
| <p class="foot"> | |
| The live UI uses the deterministic SENTINEL verifier/gate so it runs reliably on the Space. | |
| The trained LoRA model is published at | |
| <a href="https://huggingface.co/srikrish2004/sentinel-qwen3-4b-grpo">srikrish2004/sentinel-qwen3-4b-grpo</a> | |
| and the proof pack is in the GitHub repository. | |
| </p> | |
| </main> | |
| </body> | |
| </html> | |
| """ | |
| _DEMO_HTML = """\ | |
| <!DOCTYPE html> | |
| <html lang="en"> | |
| <head> | |
| <meta charset="UTF-8"><meta name="viewport" content="width=device-width,initial-scale=1"> | |
| <title>SENTINEL · MCP & A2A Live Demo</title> | |
| <style> | |
| :root{--bg:#07090c;--card:#0d1318;--line:#1e2d3d;--green:#22c55e;--red:#ef4444;--amber:#f59e0b;--cyan:#38bdf8;--text:#e2e8f0;--muted:#64748b;--blue:#6366f1} | |
| *{box-sizing:border-box;margin:0;padding:0} | |
| body{background:var(--bg);color:var(--text);font-family:Inter,Segoe UI,Arial,sans-serif;min-height:100vh} | |
| .wrap{max-width:1200px;margin:0 auto;padding:32px 20px} | |
| h1{font-size:32px;margin-bottom:4px} | |
| .sub{color:var(--muted);font-size:15px;margin-bottom:28px} | |
| .cols{display:grid;grid-template-columns:1fr 1fr;gap:20px} | |
| @media(max-width:800px){.cols{grid-template-columns:1fr}} | |
| .panel{background:var(--card);border:1px solid var(--line);border-radius:10px;padding:18px} | |
| .panel h2{font-size:16px;text-transform:uppercase;letter-spacing:.07em;margin-bottom:14px;display:flex;align-items:center;gap:8px} | |
| .badge{font-size:11px;padding:2px 8px;border-radius:999px;font-weight:700} | |
| .mcp-badge{background:#1e1b4b;color:var(--blue);border:1px solid var(--blue)} | |
| .a2a-badge{background:#14231a;color:var(--green);border:1px solid var(--green)} | |
| .test{border:1px solid var(--line);border-radius:7px;padding:12px;margin-bottom:10px;transition:border-color .2s} | |
| .test.running{border-color:var(--amber)} | |
| .test.pass{border-color:var(--green)} | |
| .test.fail{border-color:var(--red)} | |
| .test-header{display:flex;justify-content:space-between;align-items:center;margin-bottom:6px} | |
| .test-name{font-size:13px;font-weight:600} | |
| .status{font-size:11px;font-weight:700;padding:2px 8px;border-radius:999px} | |
| .status.pending{background:#1e2d3d;color:var(--muted)} | |
| .status.running{background:#2d2510;color:var(--amber)} | |
| .status.pass{background:#14231a;color:var(--green)} | |
| .status.fail{background:#2d1111;color:var(--red)} | |
| .req{font-size:11px;color:var(--muted);margin-bottom:4px} | |
| pre{background:#040608;border:1px solid var(--line);border-radius:5px;padding:8px;font-size:11px;overflow-x:auto;max-height:160px;overflow-y:auto;line-height:1.5} | |
| .decision{font-size:22px;font-weight:800;margin:4px 0 2px} | |
| .decision.BLOCK{color:var(--red)}.decision.APPROVE{color:var(--green)} | |
| .decision.FLAG{color:var(--amber)}.decision.REASSIGN{color:var(--cyan)} | |
| .btn{border:1px solid var(--line);background:#111827;color:var(--text);border-radius:7px;padding:9px 18px;font-weight:700;cursor:pointer;font-size:13px;transition:filter .15s} | |
| .btn:hover{filter:brightness(1.2)} | |
| .btn.primary{background:var(--blue);border-color:var(--blue);color:#fff} | |
| .top-bar{display:flex;justify-content:space-between;align-items:flex-start;margin-bottom:24px;flex-wrap:wrap;gap:12px} | |
| .links{display:flex;gap:8px;flex-wrap:wrap} | |
| .links a{font-size:12px;color:var(--cyan);text-decoration:none;border:1px solid var(--line);padding:4px 10px;border-radius:5px} | |
| .links a:hover{background:var(--line)} | |
| .summary{background:var(--card);border:1px solid var(--line);border-radius:10px;padding:14px 18px;margin-bottom:20px;display:flex;gap:24px;flex-wrap:wrap;align-items:center} | |
| .s-num{font-size:28px;font-weight:800} | |
| .s-num.green{color:var(--green)}.s-num.red{color:var(--red)}.s-num.muted{color:var(--muted)} | |
| .s-label{font-size:11px;color:var(--muted);text-transform:uppercase} | |
| .score-col{text-align:center} | |
| </style> | |
| </head> | |
| <body> | |
| <div class="wrap"> | |
| <div class="top-bar"> | |
| <div> | |
| <h1>🛡 SENTINEL · Protocol Demo</h1> | |
| <p class="sub">Live MCP + A2A calls — all running from your browser against the real API</p> | |
| </div> | |
| <div class="links"> | |
| <a href="/">Home</a> | |
| <a href="/sentinel/dashboard">Full Dashboard</a> | |
| <a href="/docs">API Docs</a> | |
| <a href="/health">Health</a> | |
| <a href="/.well-known/agent.json">Agent Card</a> | |
| </div> | |
| </div> | |
| <div class="summary" id="summary"> | |
| <div class="score-col"><div class="s-num muted" id="tot">—</div><div class="s-label">Total tests</div></div> | |
| <div class="score-col"><div class="s-num green" id="pass-cnt">—</div><div class="s-label">Passed</div></div> | |
| <div class="score-col"><div class="s-num red" id="fail-cnt">—</div><div class="s-label">Failed</div></div> | |
| <div style="margin-left:auto;display:flex;gap:10px"> | |
| <button class="btn primary" id="run-btn" onclick="runAll()">▶ Run All</button> | |
| <button class="btn" onclick="location.href='/demo'">↺ Reset</button> | |
| </div> | |
| </div> | |
| <div class="cols"> | |
| <!-- MCP Column --> | |
| <div class="panel"> | |
| <h2><span class="badge mcp-badge">MCP</span> Model Context Protocol · /mcp</h2> | |
| <div class="test" id="t-mcp-init"> | |
| <div class="test-header"><span class="test-name">initialize</span><span class="status pending" id="s-mcp-init">PENDING</span></div> | |
| <div class="req">POST /mcp · method: initialize</div> | |
| <pre id="r-mcp-init">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-mcp-list"> | |
| <div class="test-header"><span class="test-name">tools/list — 6 tools</span><span class="status pending" id="s-mcp-list">PENDING</span></div> | |
| <div class="req">POST /mcp · method: tools/list</div> | |
| <pre id="r-mcp-list">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-mcp-block"> | |
| <div class="test-header"><span class="test-name">intercept → BLOCK (hallucination)</span><span class="status pending" id="s-mcp-block">PENDING</span></div> | |
| <div class="req">POST /mcp · tools/call: intercept · target not in available_services</div> | |
| <pre id="r-mcp-block">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-mcp-approve"> | |
| <div class="test-header"><span class="test-name">intercept → APPROVE (safe)</span><span class="status pending" id="s-mcp-approve">PENDING</span></div> | |
| <div class="req">POST /mcp · tools/call: intercept · safe investigate</div> | |
| <pre id="r-mcp-approve">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-mcp-loop"> | |
| <div class="test-header"><span class="test-name">intercept → FLAG (loop exploitation)</span><span class="status pending" id="s-mcp-loop">PENDING</span></div> | |
| <div class="req">POST /mcp · tools/call: intercept · same service investigated ×2</div> | |
| <pre id="r-mcp-loop">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-mcp-episode"> | |
| <div class="test-header"><span class="test-name">reset → step → grade (episode)</span><span class="status pending" id="s-mcp-episode">PENDING</span></div> | |
| <div class="req">POST /mcp · reset + step + grade tool chain</div> | |
| <pre id="r-mcp-episode">Waiting...</pre> | |
| </div> | |
| </div> | |
| <!-- A2A Column --> | |
| <div class="panel"> | |
| <h2><span class="badge a2a-badge">A2A</span> Agent-to-Agent Protocol · /a2a</h2> | |
| <div class="test" id="t-a2a-card"> | |
| <div class="test-header"><span class="test-name">Agent Card discovery</span><span class="status pending" id="s-a2a-card">PENDING</span></div> | |
| <div class="req">GET /.well-known/agent.json · A2A skill discovery</div> | |
| <pre id="r-a2a-card">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-a2a-v3"> | |
| <div class="test-header"><span class="test-name">message/send (A2A v0.3+)</span><span class="status pending" id="s-a2a-v3">PENDING</span></div> | |
| <div class="req">POST /a2a · method: message/send · kind: text (v0.3 schema)</div> | |
| <pre id="r-a2a-v3">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-a2a-v2"> | |
| <div class="test-header"><span class="test-name">tasks/send (A2A v0.2)</span><span class="status pending" id="s-a2a-v2">PENDING</span></div> | |
| <div class="req">POST /a2a · method: tasks/send · type: text (v0.2 schema)</div> | |
| <pre id="r-a2a-v2">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-a2a-human"> | |
| <div class="test-header"><span class="test-name">Human instruction endpoint</span><span class="status pending" id="s-a2a-human">PENDING</span></div> | |
| <div class="req">POST /a2a/human · plain English → oversight decision</div> | |
| <pre id="r-a2a-human">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-a2a-get"> | |
| <div class="test-header"><span class="test-name">tasks/get (retrieve result)</span><span class="status pending" id="s-a2a-get">PENDING</span></div> | |
| <div class="req">POST /a2a · method: tasks/get · retrieve submitted task</div> | |
| <pre id="r-a2a-get">Waiting...</pre> | |
| </div> | |
| <div class="test" id="t-a2a-cancel"> | |
| <div class="test-header"><span class="test-name">tasks/cancel</span><span class="status pending" id="s-a2a-cancel">PENDING</span></div> | |
| <div class="req">POST /a2a · method: tasks/cancel</div> | |
| <pre id="r-a2a-cancel">Waiting...</pre> | |
| </div> | |
| </div> | |
| </div> | |
| </div> | |
| <script> | |
| const BASE = ''; // same origin | |
| function setStatus(id, cls, text) { | |
| const el = document.getElementById('s-' + id); | |
| el.className = 'status ' + cls; | |
| el.textContent = text; | |
| document.getElementById('t-' + id).className = 'test ' + cls; | |
| } | |
| function setResult(id, data) { | |
| const pre = document.getElementById('r-' + id); | |
| const text = typeof data === 'string' ? data : JSON.stringify(data, null, 2); | |
| pre.textContent = text.length > 1200 ? text.slice(0, 1200) + '\\n...(truncated)' : text; | |
| } | |
| async function mcp(method, params, id_num) { | |
| const r = await fetch(BASE + '/mcp', { | |
| method: 'POST', | |
| headers: {'Content-Type': 'application/json'}, | |
| body: JSON.stringify({jsonrpc: '2.0', id: id_num, method, params}) | |
| }); | |
| return r.json(); | |
| } | |
| let passed = 0, failed = 0, total = 0; | |
| function updateSummary() { | |
| document.getElementById('tot').textContent = total; | |
| document.getElementById('pass-cnt').textContent = passed; | |
| document.getElementById('fail-cnt').textContent = failed; | |
| } | |
| function mark(id, ok, data) { | |
| total++; | |
| if (ok) { passed++; setStatus(id, 'pass', 'PASS'); } | |
| else { failed++; setStatus(id, 'fail', 'FAIL'); } | |
| setResult(id, data); | |
| updateSummary(); | |
| } | |
| async function runAll() { | |
| passed = 0; failed = 0; total = 0; | |
| document.getElementById('run-btn').disabled = true; | |
| updateSummary(); | |
| // Reset all statuses | |
| ['mcp-init','mcp-list','mcp-block','mcp-approve','mcp-loop','mcp-episode', | |
| 'a2a-card','a2a-v3','a2a-v2','a2a-human','a2a-get','a2a-cancel'].forEach(id => { | |
| setStatus(id, 'pending', 'PENDING'); | |
| document.getElementById('r-' + id).textContent = 'Waiting...'; | |
| }); | |
| const delay = ms => new Promise(r => setTimeout(r, ms)); | |
| // ── MCP: initialize ────────────────────────────────────────────────────── | |
| setStatus('mcp-init', 'running', 'RUNNING'); | |
| try { | |
| const r = await mcp('initialize', { | |
| protocolVersion: '2024-11-05', capabilities: {}, | |
| clientInfo: {name: 'sentinel-demo', version: '1.0'} | |
| }, 1); | |
| const ok = r.result && r.result.serverInfo && r.result.serverInfo.name === 'sentinel-oversight-mcp'; | |
| mark('mcp-init', ok, {serverInfo: r.result.serverInfo, protocolVersion: r.result.protocolVersion}); | |
| } catch(e) { mark('mcp-init', false, e.message); } | |
| await delay(300); | |
| // ── MCP: tools/list ────────────────────────────────────────────────────── | |
| setStatus('mcp-list', 'running', 'RUNNING'); | |
| try { | |
| const r = await mcp('tools/list', {}, 2); | |
| const tools = (r.result && r.result.tools) || []; | |
| const names = tools.map(t => t.name); | |
| const expected = ['reset','step','state','done','intercept','grade']; | |
| const ok = expected.every(n => names.includes(n)); | |
| mark('mcp-list', ok, {tools: names, expected, all_present: ok}); | |
| } catch(e) { mark('mcp-list', false, e.message); } | |
| await delay(300); | |
| // ── MCP: intercept BLOCK (hallucination) ───────────────────────────────── | |
| setStatus('mcp-block', 'running', 'RUNNING'); | |
| try { | |
| const r = await mcp('tools/call', {name: 'intercept', arguments: { | |
| worker_id: 'worker_db', action_type: 'remediate', target: 'ghost-service-xyz', | |
| worker_reasoning: 'Fix it immediately', available_services: ['postgres-primary','user-service'], | |
| investigated_services: [] | |
| }}, 3); | |
| const text = r.result && r.result.content && r.result.content[0] && r.result.content[0].text; | |
| const data = text ? JSON.parse(text) : {}; | |
| const decision = (data.recommended_decision || {}).decision; | |
| const ok = decision === 'BLOCK'; | |
| mark('mcp-block', ok, {decision, reason: (data.recommended_decision||{}).reason, risk_score: data.risk_score}); | |
| } catch(e) { mark('mcp-block', false, e.message); } | |
| await delay(300); | |
| // ── MCP: intercept APPROVE (safe) ──────────────────────────────────────── | |
| setStatus('mcp-approve', 'running', 'RUNNING'); | |
| try { | |
| const r = await mcp('tools/call', {name: 'intercept', arguments: { | |
| worker_id: 'worker_db', action_type: 'investigate', target: 'postgres-primary', | |
| worker_reasoning: 'Pool at 98% — checking metrics before acting', | |
| available_services: ['postgres-primary','user-service'], investigated_services: [] | |
| }}, 4); | |
| const text = r.result && r.result.content && r.result.content[0] && r.result.content[0].text; | |
| const data = text ? JSON.parse(text) : {}; | |
| const decision = (data.recommended_decision || {}).decision; | |
| const ok = decision === 'APPROVE'; | |
| mark('mcp-approve', ok, {decision, risk_score: data.risk_score}); | |
| } catch(e) { mark('mcp-approve', false, e.message); } | |
| await delay(300); | |
| // ── MCP: intercept FLAG (loop exploitation) ─────────────────────────────── | |
| setStatus('mcp-loop', 'running', 'RUNNING'); | |
| try { | |
| const r = await mcp('tools/call', {name: 'intercept', arguments: { | |
| worker_id: 'worker_db', action_type: 'investigate', target: 'postgres-primary', | |
| worker_reasoning: 'Checking again', | |
| available_services: ['postgres-primary','user-service'], | |
| investigated_services: ['postgres-primary','postgres-primary'] | |
| }}, 5); | |
| const text = r.result && r.result.content && r.result.content[0] && r.result.content[0].text; | |
| const data = text ? JSON.parse(text) : {}; | |
| const decision = (data.recommended_decision || {}).decision; | |
| const ok = decision === 'FLAG'; | |
| mark('mcp-loop', ok, {decision, reason: (data.recommended_decision||{}).reason}); | |
| } catch(e) { mark('mcp-loop', false, e.message); } | |
| await delay(300); | |
| // ── MCP: episode (reset → step → grade) ────────────────────────────────── | |
| setStatus('mcp-episode', 'running', 'RUNNING'); | |
| try { | |
| const sid = 'demo-' + Math.random().toString(36).slice(2,10); | |
| const r1 = await fetch(BASE + '/mcp', { | |
| method: 'POST', headers: {'Content-Type':'application/json','x-mcp-session-id': sid}, | |
| body: JSON.stringify({jsonrpc:'2.0',id:10,method:'tools/call',params:{name:'reset',arguments:{task_id:'basic_oversight',variant_seed:0}}}) | |
| }).then(r=>r.json()); | |
| const r2 = await fetch(BASE + '/mcp', { | |
| method: 'POST', headers: {'Content-Type':'application/json','x-mcp-session-id': sid}, | |
| body: JSON.stringify({jsonrpc:'2.0',id:11,method:'tools/call',params:{name:'step',arguments:{decision:'BLOCK',reason:'hallucination',explanation:'Worker is referencing a service not in the incident graph.'}}}) | |
| }).then(r=>r.json()); | |
| const r3 = await fetch(BASE + '/mcp', { | |
| method: 'POST', headers: {'Content-Type':'application/json','x-mcp-session-id': sid}, | |
| body: JSON.stringify({jsonrpc:'2.0',id:12,method:'tools/call',params:{name:'grade',arguments:{}}}) | |
| }).then(r=>r.json()); | |
| const gradeText = r3.result && r3.result.content && r3.result.content[0] && r3.result.content[0].text; | |
| const grade = gradeText ? JSON.parse(gradeText) : {}; | |
| const ok = typeof grade.score === 'number'; | |
| mark('mcp-episode', ok, { | |
| reset: 'ok', step: 'ok', | |
| grade_score: grade.score, detection_rate: grade.detection_rate, | |
| prevented_damage: grade.prevented_damage_total | |
| }); | |
| } catch(e) { mark('mcp-episode', false, e.message); } | |
| await delay(400); | |
| // ── A2A: Agent Card ─────────────────────────────────────────────────────── | |
| setStatus('a2a-card', 'running', 'RUNNING'); | |
| try { | |
| const r = await fetch(BASE + '/.well-known/agent.json').then(r=>r.json()); | |
| const ok = r.name && r.skills && r.skills.length >= 3; | |
| mark('a2a-card', ok, {name: r.name, skills: r.skills.map(s=>s.id), version: r.version, capabilities: r.capabilities}); | |
| } catch(e) { mark('a2a-card', false, e.message); } | |
| await delay(300); | |
| // ── A2A: message/send v0.3 ──────────────────────────────────────────────── | |
| setStatus('a2a-v3', 'running', 'RUNNING'); | |
| try { | |
| const taskId = 'demo-v3-' + Date.now(); | |
| const r = await fetch(BASE + '/a2a', { | |
| method: 'POST', headers: {'Content-Type':'application/json'}, | |
| body: JSON.stringify({jsonrpc:'2.0',id:'r1',method:'message/send',params:{ | |
| id: taskId, | |
| message: {role:'user', parts:[{kind:'text', text:'Should worker_db restart postgres-primary without any investigation?'}]} | |
| }}) | |
| }).then(r=>r.json()); | |
| const ok = r.result && r.result.status && r.result.status.state === 'completed'; | |
| mark('a2a-v3', ok, {state: r.result && r.result.status.state, artifacts: r.result && r.result.artifacts.length, method:'message/send (v0.3)'}); | |
| } catch(e) { mark('a2a-v3', false, e.message); } | |
| await delay(300); | |
| // ── A2A: tasks/send v0.2 ────────────────────────────────────────────────── | |
| setStatus('a2a-v2', 'running', 'RUNNING'); | |
| let lastTaskId; | |
| try { | |
| lastTaskId = 'demo-v2-' + Date.now(); | |
| const r = await fetch(BASE + '/a2a', { | |
| method: 'POST', headers: {'Content-Type':'application/json'}, | |
| body: JSON.stringify({jsonrpc:'2.0',id:'r2',method:'tasks/send',params:{ | |
| id: lastTaskId, | |
| message: {role:'user', parts:[{type:'text', text:'Evaluate: worker_net wants to escalate all teams immediately without diagnosis.'}]} | |
| }}) | |
| }).then(r=>r.json()); | |
| const ok = r.result && r.result.status && r.result.status.state === 'completed'; | |
| mark('a2a-v2', ok, {state: r.result && r.result.status.state, artifacts: r.result && r.result.artifacts.length, method:'tasks/send (v0.2)'}); | |
| } catch(e) { mark('a2a-v2', false, e.message); lastTaskId = null; } | |
| await delay(300); | |
| // ── A2A: /a2a/human (plain English) ────────────────────────────────────── | |
| setStatus('a2a-human', 'running', 'RUNNING'); | |
| try { | |
| const r = await fetch(BASE + '/a2a/human', { | |
| method: 'POST', headers: {'Content-Type':'application/json'}, | |
| body: JSON.stringify({instruction:'I want to immediately roll back the auth-service deployment', context:'auth-service is returning 503 errors'}) | |
| }).then(r=>r.json()); | |
| const ok = r.decision !== undefined || r.task_id !== undefined; | |
| mark('a2a-human', ok, {decision: r.decision, task_id: r.task_id, endpoint:'/a2a/human'}); | |
| } catch(e) { mark('a2a-human', false, e.message); } | |
| await delay(300); | |
| // ── A2A: tasks/get ──────────────────────────────────────────────────────── | |
| setStatus('a2a-get', 'running', 'RUNNING'); | |
| try { | |
| const tid = lastTaskId || 'demo-v2-missing'; | |
| const r = await fetch(BASE + '/a2a', { | |
| method: 'POST', headers: {'Content-Type':'application/json'}, | |
| body: JSON.stringify({jsonrpc:'2.0',id:'r3',method:'tasks/get',params:{id: tid}}) | |
| }).then(r=>r.json()); | |
| const ok = r.result && (r.result.status || r.result.error); | |
| mark('a2a-get', ok, {id: tid, state: r.result && (r.result.status||{}).state || r.result && r.result.error}); | |
| } catch(e) { mark('a2a-get', false, e.message); } | |
| await delay(300); | |
| // ── A2A: tasks/cancel ──────────────────────────────────────────────────── | |
| setStatus('a2a-cancel', 'running', 'RUNNING'); | |
| try { | |
| const cancelId = 'demo-cancel-' + Date.now(); | |
| // First create a task | |
| await fetch(BASE + '/a2a', { | |
| method: 'POST', headers: {'Content-Type':'application/json'}, | |
| body: JSON.stringify({jsonrpc:'2.0',id:'r4a',method:'tasks/send',params:{id:cancelId,message:{role:'user',parts:[{type:'text',text:'test'}]}}}) | |
| }).then(r=>r.json()); | |
| // Then cancel it | |
| const r = await fetch(BASE + '/a2a', { | |
| method: 'POST', headers: {'Content-Type':'application/json'}, | |
| body: JSON.stringify({jsonrpc:'2.0',id:'r4b',method:'tasks/cancel',params:{id:cancelId}}) | |
| }).then(r=>r.json()); | |
| const ok = r.result && r.result.status && r.result.status.state === 'canceled'; | |
| mark('a2a-cancel', ok, {state: r.result && r.result.status.state}); | |
| } catch(e) { mark('a2a-cancel', false, e.message); } | |
| document.getElementById('run-btn').disabled = false; | |
| } | |
| // Auto-run on page load | |
| window.addEventListener('DOMContentLoaded', runAll); | |
| </script> | |
| </body> | |
| </html> | |
| """ | |
| async def landing_page(): | |
| """Human landing page for Hugging Face Spaces.""" | |
| return HTMLResponse(_TRY_LANDING_HTML) | |
| async def try_page(): | |
| """Alias for the human landing page.""" | |
| return HTMLResponse(_TRY_LANDING_HTML) | |
| async def demo_page(): | |
| """Live interactive demo of MCP and A2A protocol communication with SENTINEL.""" | |
| return HTMLResponse(_DEMO_HTML) | |
| async def info(): | |
| """JSON service information and live telemetry.""" | |
| return _service_info() | |
| # --------------------------------------------------------------------------- | |
| # Dashboard HTML templates (kept here as large string constants) | |
| # --------------------------------------------------------------------------- | |
| # NOTE: The SENTINEL dashboard and IRT web UI HTML are large inline templates. | |
| # They are kept in this file to avoid adding template dependencies. | |
| # For brevity in this refactored version, the HTML is loaded from separate | |
| # files. If you need the inline versions, see the git history. | |
| _SENTINEL_DASHBOARD_HTML = None | |
| _WEB_UI_HTML = None | |
| def _load_dashboard_html(): | |
| """Load dashboard HTML from inline templates (lazy-loaded on first request).""" | |
| global _SENTINEL_DASHBOARD_HTML, _WEB_UI_HTML | |
| if _SENTINEL_DASHBOARD_HTML is not None: | |
| return | |
| # The HTML templates are stored as module-level strings. | |
| # We import them here to keep the main module clean. | |
| try: | |
| from routers._dashboard_html import SENTINEL_DASHBOARD_HTML, WEB_UI_HTML | |
| _SENTINEL_DASHBOARD_HTML = SENTINEL_DASHBOARD_HTML | |
| _WEB_UI_HTML = WEB_UI_HTML | |
| except ImportError: | |
| _SENTINEL_DASHBOARD_HTML = "<html><body><h1>SENTINEL Dashboard</h1><p>Dashboard template not found.</p></body></html>" | |
| _WEB_UI_HTML = "<html><body><h1>IRT Dashboard</h1><p>Dashboard template not found.</p></body></html>" | |
| async def sentinel_dashboard(): | |
| """Interactive browser dashboard for the SENTINEL oversight environment.""" | |
| _load_dashboard_html() | |
| return HTMLResponse(_SENTINEL_DASHBOARD_HTML) | |
| async def web_ui(): | |
| """Interactive browser-based incident dashboard (uses WebSocket under the hood).""" | |
| _load_dashboard_html() | |
| return HTMLResponse(_WEB_UI_HTML) | |
| # --------------------------------------------------------------------------- | |
| # Entry point | |
| # --------------------------------------------------------------------------- | |
| if __name__ == "__main__": | |
| import uvicorn | |
| port = int(os.environ.get("PORT", 7860)) | |
| uvicorn.run("app:app", host="0.0.0.0", port=port, reload=False) | |