synth-ai 0.2.6.dev1__py3-none-any.whl → 0.2.6.dev3__py3-none-any.whl

synth_ai/demos/core/cli.py

@@ -6,11 +6,23 @@ import os
 import sys
 import time
 from typing import Any, Dict, Callable
+import shutil
+import stat
 
 from synth_ai.demos.demo_task_apps import core as demo_core
 from synth_ai.demos.demo_task_apps.core import DemoEnv
 
 
+def _is_modal_public_url(u: str) -> bool:
+    try:
+        s = (u or "").strip().lower()
+        if not (s.startswith("http://") or s.startswith("https://")):
+            return False
+        return (".modal.run" in s) and ("modal.local" not in s) and ("pypi-mirror" not in s)
+    except Exception:
+        return False
+
+
 def cmd_check(_args: argparse.Namespace) -> int:
     env = demo_core.load_env()
     cwd_env_path = os.path.join(os.getcwd(), ".env")
@@ -21,6 +33,15 @@ def cmd_check(_args: argparse.Namespace) -> int:
         env = demo_core.load_env()
         local_env = demo_core.load_dotenv_file(cwd_env_path)
 
+    def _is_modal_public_url(u: str) -> bool:
+        try:
+            s = (u or "").strip().lower()
+            if not (s.startswith("http://") or s.startswith("https://")):
+                return False
+            return (".modal.run" in s) and ("modal.local" not in s) and ("pypi-mirror" not in s)
+        except Exception:
+            return False
+
     def _maybe_fix_task_url() -> None:
         if not env.task_app_name:
             return
@@ -28,9 +49,7 @@ def cmd_check(_args: argparse.Namespace) -> int:
         needs_lookup = False
         if not current:
             needs_lookup = True
-        elif not current.endswith(".run") or current.endswith(".moda") or current.count(".") < 2:
-            needs_lookup = True
-        elif not current.startswith("http://") and not current.startswith("https://"):
+        elif not _is_modal_public_url(current):
             needs_lookup = True
         if not needs_lookup:
             return
@@ -48,7 +67,7 @@ def cmd_check(_args: argparse.Namespace) -> int:
             return
         new_url = ""
         for token in out.split():
-            if token.startswith("http://") or token.startswith("https://"):
+            if _is_modal_public_url(token):
                 new_url = token.strip().rstrip("/")
                 break
         if new_url and new_url != current:
@@ -166,6 +185,45 @@ def _popen_stream(cmd: list[str], cwd: str | None = None, env: dict | None = Non
     return int(proc.returncode or 0)
 
 
+def _popen_stream_capture(cmd: list[str], cwd: str | None = None, env: dict | None = None) -> tuple[int, str]:
+    """Stream subprocess output to stdout and also capture it into a buffer."""
+    import subprocess
+    import threading
+
+    buf_lines: list[str] = []
+    try:
+        proc = subprocess.Popen(
+            cmd,
+            cwd=cwd,
+            env=env,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.STDOUT,
+            text=True,
+            bufsize=1,
+        )
+    except Exception as exc:
+        print(f"Failed to launch {' '.join(cmd)}: {exc}")
+        return 1, ""
+
+    def _pump(stdout) -> None:
+        try:
+            for line in stdout:
+                line = line.rstrip()
+                print(line)
+                buf_lines.append(line)
+        except Exception:
+            pass
+
+    if proc.stdout is not None:
+        t = threading.Thread(target=_pump, args=(proc.stdout,), daemon=True)
+        t.start()
+        proc.wait()
+        t.join(timeout=1.0)
+    else:
+        proc.wait()
+    return int(proc.returncode or 0), "\n".join(buf_lines)
+
+
 def cmd_deploy(args: argparse.Namespace) -> int:
     env = demo_core.load_env()
     url = ""
@@ -187,9 +245,12 @@ def cmd_deploy(args: argparse.Namespace) -> int:
             # Auto-detect app path if not supplied; prompt for name and confirmation.
             app_path = os.path.abspath(args.app) if args.app else None
             if not app_path or not os.path.isfile(app_path):
+                # Prefer the synth_demo/ app seeded by `rl_demo init` over any root-level files
                 candidates = [
+                    os.path.abspath(os.path.join(os.getcwd(), "synth_demo", "task_app.py")),
+                    os.path.abspath(os.path.join(os.getcwd(), "task_app.py")),
+                    os.path.abspath(os.path.join(os.getcwd(), "app.py")),
                     os.path.abspath(os.path.join(os.getcwd(), "math_task_app.py")),
-                    "/Users/joshpurtell/Documents/GitHub/monorepo/tests/applications/math/rl/math_task_app.py",
                 ]
                 app_path = next((p for p in candidates if os.path.isfile(p)), None)
             if not app_path and args.script:
@@ -200,12 +261,14 @@ def cmd_deploy(args: argparse.Namespace) -> int:
                     app_name = args.name
             else:
                 if not app_path:
-                    entered = input("Path to Modal app.py (e.g., tests/applications/math/rl/math_task_app.py): ").strip()
+                    entered = input("Path to Modal app.py (e.g., ./task_app.py): ").strip()
                     if not entered:
                         raise FileNotFoundError("No app.py path provided and auto-detect failed")
                     app_path = os.path.abspath(entered)
                 if not os.path.isfile(app_path):
                     raise FileNotFoundError(f"App file not found: {app_path}")
+                # Surface the app path before asking for the name
+                print(f"Using task app: {app_path}")
                 suggested_name = args.name or f"synth-{os.path.splitext(os.path.basename(app_path))[0]}"
                 name_in = input(f"Modal app name [{suggested_name}]: ").strip() or suggested_name
                 app_name = name_in
@@ -217,22 +280,44 @@ def cmd_deploy(args: argparse.Namespace) -> int:
                     print("Aborted by user.")
                     return 1
                 deploy_cmd = ["uv", "run", "python", "-m", "modal", "deploy", "--name", name_in, app_path]
-                code, out = _popen_capture(deploy_cmd)
-                print(out)
+                print("\nStreaming Modal build/deploy logs (this can take several minutes on first run)…\n")
+                code, deploy_logs = _popen_stream_capture(deploy_cmd)
                 if code != 0:
                     raise RuntimeError(f"modal deploy failed (exit {code})")
+                # Try to parse URL directly from streamed logs
+                if not url:
+                    try:
+                        import re as _re
+                        m_all = _re.findall(r"https?://[^\s]+\.modal\.run", deploy_logs or "")
+                        if m_all:
+                            url = m_all[-1].strip().rstrip("/")
+                    except Exception:
+                        pass
                 url_cmd = ["uv", "run", "python", "-m", "modal", "app", "url", name_in]
                 code2, out2 = _popen_capture(url_cmd)
                 if code2 == 0:
                     for token in out2.split():
-                        if token.startswith("http://") or token.startswith("https://"):
+                        if _is_modal_public_url(token):
                             url = token.strip().rstrip("/")
                             break
+                # Fallback: try reading recent Modal logs for the app to find a URL line
                 if not url:
-                    for token in (out + "\n" + out2).split():
-                        if token.startswith("http://") or token.startswith("https://"):
-                            url = token.strip().rstrip("/")
-                            break
+                    code3, out3 = _popen_capture(["uv", "run", "python", "-m", "modal", "app", "list"])
+                    if code3 == 0 and out3:
+                        for line in out3.splitlines():
+                            if name_in in line:
+                                for token in line.split():
+                                    if _is_modal_public_url(token):
+                                        url = token.strip().rstrip("/")
+                                        break
+                            if url:
+                                break
+                # Prompt user if still no valid URL
+                if not url:
+                    print("\nCould not auto-detect a public Modal URL for the app.")
+                    entered = input("Enter the Modal public URL (must contain '.modal.run'), or press Enter to abort: ").strip()
+                    if entered and _is_modal_public_url(entered):
+                        url = entered.rstrip("/")
                 if not url:
                     raise RuntimeError("Failed to resolve public URL from modal CLI output")
         if not url:
@@ -286,9 +371,34 @@ def cmd_configure(args: argparse.Namespace) -> int:
     demo_core.persist_dotenv_values({"ENVIRONMENT_API_KEY": env_key})
 
     task_url = env.task_app_base_url
-    if not task_url:
-        print("Task app URL missing. Run: uvx synth-ai rl_demo deploy")
-        return 1
+    if not task_url or not _is_modal_public_url(task_url):
+        # If we have an app name, offer to resolve from Modal first
+        resolved = ""
+        if env.task_app_name:
+            try:
+                choice = input(f"Resolve URL from Modal for app '{env.task_app_name}'? [Y/n]: ").strip().lower() or "y"
+                if choice.startswith("y"):
+                    code, out = _popen_capture([
+                        "uv", "run", "python", "-m", "modal", "app", "url", env.task_app_name
+                    ])
+                    if code == 0 and out:
+                        for tok in out.split():
+                            if _is_modal_public_url(tok):
+                                resolved = tok.strip().rstrip("/")
+                                break
+            except Exception:
+                resolved = ""
+        if not resolved:
+            print("Task app URL not configured or not a valid Modal public URL.")
+            print("Examples: https://<app-name>-fastapi-app.modal.run")
+            entered = input("Enter Task App base URL (must contain '.modal.run'), or press Enter to abort: ").strip()
+            if not entered or not _is_modal_public_url(entered):
+                print("Valid Task App URL is required. Run: uvx synth-ai rl_demo deploy")
+                return 1
+            task_url = entered.rstrip("/")
+        else:
+            task_url = resolved
+        demo_core.persist_task_url(task_url, name=(env.task_app_name or None))
 
     app_name = env.task_app_name.strip()
     if not app_name:
@@ -317,20 +427,56 @@ def cmd_configure(args: argparse.Namespace) -> int:
         secret_args.append(f"SYNTH_API_KEY={synth_for_secret}")
 
     create_cmd = ["uv", "run", "modal", "secret", "create", secret_name, *secret_args]
-    code, out = _popen_capture(create_cmd)
+    def _mask_args(args: list[str]) -> list[str]:
+        masked: list[str] = []
+        for a in args:
+            if "=" in a and any(a.startswith(k + "=") for k in ("ENVIRONMENT_API_KEY", "OPENAI_API_KEY", "SYNTH_API_KEY")):
+                try:
+                    k, v = a.split("=", 1)
+                    suf = v[-5:] if len(v) >= 5 else ""
+                    masked.append(f"{k}=***{suf}")
+                except Exception:
+                    masked.append("<masked>")
+            else:
+                masked.append(a)
+        return masked
+
+    print("\n[configure] Creating Modal secret (streaming logs)…")
+    print("[configure] Command:", " ".join(_mask_args(create_cmd)))
+    code = _popen_stream(create_cmd)
     if code != 0:
-        print(out)
-        print("Secret create failed; retrying with delete → create…")
-        _popen_capture(["uv", "run", "modal", "secret", "delete", secret_name])
-        code, out = _popen_capture(create_cmd)
+        print("[configure] Secret create failed; attempting delete → create")
+        delete_cmd = ["bash", "-lc", f"printf 'y\\n' | uv run modal secret delete {secret_name}"]
+        print("[configure] Command:", " ".join(delete_cmd))
+        _popen_stream(delete_cmd)
+        print("[configure] Retrying secret create…")
+        print("[configure] Command:", " ".join(_mask_args(create_cmd)))
+        code = _popen_stream(create_cmd)
         if code != 0:
-            print(out)
-            print("Failed to provision Modal secret.")
+            print("[configure] Failed to provision Modal secret.")
             return 2
 
     # Verify task app can read the secret by hitting rollout health with X-API-Key.
     rollout_url = task_url.rstrip("/") + "/health/rollout"
-    rc, body = _http("GET", rollout_url, headers={"X-API-Key": env_key})
+    print("[configure] Verifying rollout health:")
+    # Prefer rollout-specific health first (auth-aware), then plain /health
+    health_base = task_url.rstrip("/")
+    health_urls = [f"{health_base}/health/rollout", f"{health_base}/health"]
+    rc = 0
+    body = ""
+    for h in health_urls:
+        print("[configure] GET", h)
+        rc, body = _http("GET", h, headers={"X-API-Key": env_key})
+        if rc == 200:
+            rollout_url = h
+            break
+    print("[configure] status:", rc)
+    try:
+        import json as _json
+        preview = _json.dumps(body)[:800] if isinstance(body, dict) else str(body)[:800]
+    except Exception:
+        preview = str(body)[:800]
+    print("[configure] body:", preview)
     if rc != 200:
         print(f"Warning: rollout health check failed ({rc}). Response: {body}")
     else:
@@ -341,22 +487,26 @@ def cmd_configure(args: argparse.Namespace) -> int:
     env.task_app_name = app_name
     env.task_app_secret_name = secret_name
 
-    # Prepare a baseline TOML (formerly `prepare`): prompt and write demo_config.toml
-    defaults = [
-        os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "demo_task_apps", "math", "config.toml")),
-    ]
-    mono = "/Users/joshpurtell/Documents/GitHub/monorepo/tests/applications/math/rl/math_online.toml"
-    if os.path.isfile(mono):
-        defaults.append(mono)
-    print("Select a baseline TOML:")
-    for i, p in enumerate(defaults, 1):
-        print(f"  [{i}] {p}")
-    choice = input(f"Enter choice [1-{len(defaults)}] (default 1): ").strip() or "1"
-    try:
-        idx = max(1, min(int(choice), len(defaults))) - 1
-    except Exception:
-        idx = 0
-    base_path = defaults[idx]
+    # Prefer the seeded CWD config if present; otherwise fall back to packaged default
+    seeded_cfg = os.path.abspath(os.path.join(os.getcwd(), "demo_config.toml"))
+    if os.path.isfile(seeded_cfg):
+        base_path = seeded_cfg
+    else:
+        defaults = [
+            os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "demo_task_apps", "math", "config.toml")),
+        ]
+        mono = "/Users/joshpurtell/Documents/GitHub/monorepo/tests/applications/math/rl/math_online.toml"
+        if os.path.isfile(mono):
+            defaults.append(mono)
+        print("Select a baseline TOML:")
+        for i, p in enumerate(defaults, 1):
+            print(f"  [{i}] {p}")
+        choice = input(f"Enter choice [1-{len(defaults)}] (default 1): ").strip() or "1"
+        try:
+            idx = max(1, min(int(choice), len(defaults))) - 1
+        except Exception:
+            idx = 0
+        base_path = defaults[idx]
     with open(base_path, "r") as fh:
         text = fh.read()
     import re
@@ -413,6 +563,131 @@ def cmd_configure(args: argparse.Namespace) -> int:
     return 0
 
 
+def cmd_init(args: argparse.Namespace) -> int:
+    """Initialize a Modal-ready Math Task App in the current directory.
+
+    Copies `examples/rl/task_app.py` and `examples/rl/deploy_task_app.sh` into CWD.
+    Creates a `.env` with placeholders if it does not exist.
+    """
+    try:
+        # Ensure `modal` is installed for deployment flows
+        def _has_modal() -> bool:
+            try:
+                import importlib.util as _iu
+                return _iu.find_spec("modal") is not None
+            except Exception:
+                return False
+
+        if not _has_modal():
+            print("modal not found; installing…")
+            # Prefer uv if available; otherwise fallback to pip
+            try:
+                if shutil.which("uv"):
+                    code, out = _popen_capture(["uv", "pip", "install", "modal>=1.1.4"])
+                else:
+                    code, out = _popen_capture([sys.executable, "-m", "pip", "install", "modal>=1.1.4"])
+                if code != 0:
+                    print(out)
+                    print("Failed to install modal; continuing may fail.")
+                else:
+                    print("modal installed successfully.")
+            except Exception as e:
+                print(f"modal install error: {e}")
+            # Re-check
+            if not _has_modal():
+                print("Warning: modal is still not importable after install attempt.")
+        else:
+            print("modal found")
+
+        here = os.getcwd()
+        demo_dir = os.path.join(here, "synth_demo")
+        os.makedirs(demo_dir, exist_ok=True)
+        # Paths inside synth_demo/
+        dst_task_py = os.path.join(demo_dir, "task_app.py")
+        dst_deploy = os.path.join(demo_dir, "deploy_task_app.sh")
+        env_path = os.path.join(demo_dir, ".env")
+        dst_cfg = os.path.join(demo_dir, "demo_config.toml")
+
+        # Copy packaged math modal task app into synth_demo/task_app.py
+        src_modal = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "demo_task_apps", "math", "modal_task_app.py"))
+        if not os.path.isfile(src_modal):
+            print("Init failed: packaged math modal task app not found.")
+            print(f"Looked for: {src_modal}")
+            return 1
+        if os.path.exists(dst_task_py) and not getattr(args, "force", False):
+            print(f"Refusing to overwrite existing file: {dst_task_py} (use --force)")
+            return 1
+        shutil.copy2(src_modal, dst_task_py)
+
+        # Create deploy script in synth_demo/
+        deploy_text = """#!/usr/bin/env bash
+set -euo pipefail
+
+HERE=$(cd "$(dirname "$0")" && pwd)
+APP="$HERE/task_app.py"
+if [ -f "$HERE/.env" ]; then
+  # shellcheck disable=SC2046
+  export $(grep -v '^#' "$HERE/.env" | xargs -I{} echo {})
+fi
+uv run modal deploy "$APP" | tee "$HERE/.last_deploy.log"
+URL=$(grep -Eo 'https://[^ ]+\.modal\.run' "$HERE/.last_deploy.log" | tail -1 || true)
+if [ -n "$URL" ]; then
+  if grep -q '^TASK_APP_BASE_URL=' "$HERE/.env" 2>/dev/null; then
+    sed -i.bak "s#^TASK_APP_BASE_URL=.*#TASK_APP_BASE_URL=$URL#" "$HERE/.env" || true
+  else
+    echo "TASK_APP_BASE_URL=$URL" >> "$HERE/.env"
+  fi
+  echo "Saved TASK_APP_BASE_URL to $HERE/.env"
+fi
+"""
+        _write_text(dst_deploy, deploy_text)
+        try:
+            st = os.stat(dst_deploy)
+            os.chmod(dst_deploy, st.st_mode | stat.S_IXUSR | stat.S_IXGRP | stat.S_IXOTH)
+        except Exception:
+            pass
+
+        # Seed .env if not present
+        if not os.path.exists(env_path):
+            _write_text(env_path, "\n".join([
+                "# Required for task app auth to environment service",
+                "ENVIRONMENT_API_KEY=",
+                "",
+                "# Optional: for CLI job submission and proxying OpenAI models",
+                "SYNTH_API_KEY=",
+                "OPENAI_API_KEY=",
+                "",
+                "# Optional: set to 'prod' to use production names",
+                "ENVIRONMENT=",
+            ]) + "\n")
+
+        # Seed demo_config.toml from packaged default if not present (or overwrite with --force)
+        packaged_cfg = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "demo_task_apps", "math", "config.toml"))
+        try:
+            if os.path.isfile(packaged_cfg):
+                if not os.path.exists(dst_cfg) or getattr(args, "force", False):
+                    shutil.copy2(packaged_cfg, dst_cfg)
+        except Exception:
+            pass
+
+        print("Initialized Math Task App in synth_demo/:")
+        print(f"  - {dst_task_py}")
+        print(f"  - {dst_deploy}")
+        print(f"  - {env_path} (created if missing)")
+        if os.path.exists(dst_cfg):
+            print(f"  - {dst_cfg} (seeded)")
+        print("")
+        print("Next steps:")
+        print("  1) cd synth_demo && put your ENVIRONMENT_API_KEY in ./.env")
+        print("  2) Deploy to Modal:")
+        print("     uvx bash ./deploy_task_app.sh")
+        print("  3) From project root, run: uvx synth-ai rl_demo configure; uvx synth-ai rl_demo run")
+        return 0
+    except Exception as e:
+        print(f"Init error: {e}")
+        return 2
+
+
 def _http(method: str, url: str, headers: Dict[str, str] | None = None, body: Dict[str, Any] | None = None) -> tuple[int, Dict[str, Any] | str]:
     import urllib.request, urllib.error, json as _json
     data = None
@@ -493,6 +768,19 @@ def cmd_run(args: argparse.Namespace) -> int:
         code = _popen_stream(cmd, env=run_env)
         if code != 0:
             print(f"Clustered runner exited with code {code}")
+            # Actionable guidance for common auth issues
+            try:
+                base_url = backend_base.rstrip("/") + "/api"
+            except Exception:
+                base_url = backend_base
+            sk = (env.synth_api_key or "").strip()
+            ek = (env.env_api_key or "").strip()
+            print("Hint: If backend responded 401, verify SYNTH_API_KEY for:", base_url)
+            if sk:
+                print(f"  SYNTH_API_KEY len={len(sk)} last5={sk[-5:]}")
+            if ek:
+                print(f"  ENVIRONMENT_API_KEY len={len(ek)} last5={ek[-5:]}")
+            print("Also ensure your Modal secret contains ENVIRONMENT_API_KEY and matches the task app.")
         return code
 
     # Fallback: legacy jobs API flow
@@ -591,6 +879,17 @@ def cmd_run(args: argparse.Namespace) -> int:
         except Exception:
             print(str(js))
         print("Request body was:\n" + json.dumps(body, indent=2))
+        # Extra hints for auth failures
+        try:
+            sk = (env.synth_api_key or "").strip()
+            if int(code) == 401 or (isinstance(js, dict) and any(isinstance(v, str) and "Invalid API key" in v for v in js.values())):
+                base_url = env.dev_backend_url
+                print("Hint: HTTP 401 Unauthorized from backend. Verify SYNTH_API_KEY for:", base_url)
+                if sk:
+                    print(f"  SYNTH_API_KEY len={len(sk)} last5={sk[-5:]}")
+                print("Also ensure your Modal secret contains a valid ENVIRONMENT_API_KEY.")
+        except Exception:
+            pass
         return 2
     job_id = js.get("job_id") or js.get("id") or ""
     if not job_id:
@@ -639,6 +938,96 @@ def cmd_run(args: argparse.Namespace) -> int:
     return 0
 
 
+def cmd_eval(args: argparse.Namespace) -> int:
+    env = demo_core.load_env()
+    # Ensure required env
+    if not env.task_app_base_url:
+        print("Task app URL missing. Run: uvx synth-ai rl_demo deploy")
+        return 1
+    # Load config: prefer CWD demo_config.toml; else packaged default
+    cfg_path: str | None = None
+    if getattr(args, "config", None):
+        p = os.path.abspath(args.config)
+        if not os.path.isfile(p):
+            print(f"Config not found: {p}")
+            return 1
+        cfg_path = p
+    else:
+        cwd_prepared = os.path.abspath(os.path.join(os.getcwd(), "demo_config.toml"))
+        if os.path.isfile(cwd_prepared):
+            cfg_path = cwd_prepared
+        else:
+            packaged = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "demo_task_apps", "math", "config.toml"))
+            cfg_path = packaged if os.path.isfile(packaged) else None
+    if not cfg_path:
+        print("No config TOML found.")
+        return 1
+
+    import tomllib
+    with open(cfg_path, "rb") as fh:
+        inline_cfg = tomllib.load(fh)
+
+    # Model selection prompt
+    default_model = (args.model or (inline_cfg.get("model", {}) or {}).get("name") or "Qwen/Qwen3-0.6B")
+    entered = input(f"Model to evaluate [{default_model}]: ").strip()
+    model = entered or default_model
+    confirm = (input(f"Use model '{model}'? [Y/n]: ").strip().lower() or "y").startswith("y")
+    if not confirm:
+        print("Aborted by user.")
+        return 1
+
+    # Build on-board rollout request to the Task App (no backend RL job)
+    # Use Synth backend chat-completions proxy as inference URL (derive from DEV_BACKEND_URL)
+    # Ensure /api suffix for backend, then use proxy prefix for chat completions
+    backend_api = (env.dev_backend_url or "https://agent-learning.onrender.com/api").rstrip("/")
+    if not backend_api.endswith("/api"):
+        backend_api = f"{backend_api}/api"
+    inference_url = f"{backend_api}/proxy"
+    # ops: alternate agent/env for a small number of decisions (from config max_steps_per_episode if present)
+    try:
+        steps = int((inline_cfg.get("rollout", {}) or {}).get("max_steps_per_episode", 4))
+    except Exception:
+        steps = 4
+    ops: list[str] = []
+    for _ in range(max(1, steps // 2)):
+        ops.extend(["agent", "env"])
+    env_name = (inline_cfg.get("rollout", {}) or {}).get("env_name") or "math"
+    policy_name = (inline_cfg.get("rollout", {}) or {}).get("policy_name") or "math-react"
+    run_id = f"eval-{int(time.time())}"
+    body: Dict[str, Any] = {
+        "run_id": run_id,
+        "env": {
+            "env_name": env_name,
+            "config": inline_cfg.get("rollout", {}) or {},
+        },
+        "policy": {
+            "policy_name": policy_name,
+            "config": {"model": model, "inference_url": inference_url},
+        },
+        "ops": ops,
+        "on_done": "terminate",
+    }
+    # POST to task app rollout endpoint
+    headers = {"Content-Type": "application/json"}
+    if env.env_api_key:
+        headers["X-API-Key"] = env.env_api_key
+    rc, resp = _http("POST", env.task_app_base_url.rstrip("/") + "/rollout", headers=headers, body=body)
+    if rc not in (200, 201) or not isinstance(resp, dict):
+        print("Eval rollout failed:", rc)
+        try:
+            print(json.dumps(resp, indent=2) if isinstance(resp, dict) else str(resp))
+        except Exception:
+            print(str(resp))
+        print("Request body was:\n" + json.dumps(body, indent=2))
+        return 2
+    metrics = (resp.get("metrics") if isinstance(resp, dict) else None) or {}
+    mean = metrics.get("mean_return")
+    if mean is not None:
+        print(f"eval.reward_mean={mean}")
+    else:
+        print(json.dumps(resp, indent=2))
+    return 0
+
 def main(argv: list[str] | None = None) -> int:
     p = argparse.ArgumentParser(prog="synth-ai")
     sub = p.add_subparsers(dest="cmd")
@@ -650,6 +1039,12 @@ def main(argv: list[str] | None = None) -> int:
 
     _add_parser(["rl_demo.check", "demo.check"], configure=lambda parser: parser.set_defaults(func=cmd_check))
 
+    def _init_opts(parser):
+        parser.add_argument("--force", action="store_true", help="Overwrite existing files in CWD")
+        parser.set_defaults(func=cmd_init)
+
+    _add_parser(["rl_demo.init", "demo.init"], configure=_init_opts)
+
     # (prepare command removed)
 
     def _deploy_opts(parser):
@@ -674,6 +1069,14 @@ def main(argv: list[str] | None = None) -> int:
 
     _add_parser(["rl_demo.run", "demo.run"], configure=_run_opts)
 
+    def _eval_opts(parser):
+        parser.add_argument("--config", type=str, default=None, help="Path to TOML config (optional)")
+        parser.add_argument("--model", type=str, default=None, help="Model to evaluate (default Qwen/Qwen3-0.6B)")
+        parser.add_argument("--timeout", type=int, default=300, help="Seconds to wait for metrics")
+        parser.set_defaults(func=cmd_eval)
+
+    _add_parser(["rl_demo.eval", "demo.eval"], configure=_eval_opts)
+
     args = p.parse_args(argv)
     if not hasattr(args, "func"):
         p.print_help()

synth_ai/demos/demo_task_apps/math/_common.py

@@ -0,0 +1,17 @@
+from __future__ import annotations
+
+"""Minimal helpers for the math task app.
+
+This module provides a local fallback for install_problem_bank_into_shared so
+the modal task app can import it without requiring an external math_rl package.
+"""
+
+def install_problem_bank_into_shared() -> None:
+    """No-op placeholder for installing the Hendrycks MATH problem bank.
+
+    In production deployments, this can download or unpack the problem bank
+    into a shared directory. For the demo scaffold, it is a no-op.
+    """
+    return None
+
+