diff --git a/plugins/openclaw/slash_sleep.py b/plugins/openclaw/slash_sleep.py index c8576661..f0f84cd7 100755 --- a/plugins/openclaw/slash_sleep.py +++ b/plugins/openclaw/slash_sleep.py @@ -19,6 +19,7 @@ import json import os import shutil +import subprocess import sys from pathlib import Path from datetime import datetime @@ -104,8 +105,8 @@ def run_category(category: str, *, dry_run: bool = False) -> int: print(f"=== /sleep run {category}{' (dry-run)' if dry_run else ''} ===") print(f" cmd: {' '.join(cmd)}") - rc = os.system(" ".join(f'"{c}"' for c in cmd)) - return rc + result = subprocess.run(cmd) + return result.returncode def run_all(*, dry_run: bool = False) -> int: diff --git a/pyproject.toml b/pyproject.toml index 5d50b8fe..c43860b2 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -38,10 +38,12 @@ dependencies = [ alfworld = ["alfworld>=0.4.0", "gymnasium>=0.29.0"] # Claude model backend claude = ["claude-agent-sdk>=0.1.0", "json_repair>=0.61.0"] +# Codex model backend (via OpenAI Codex SDK) +codex = ["openai-codex-sdk>=0.1.0"] # Qwen local model backend (via vLLM) -qwen = ["vllm>=0.4.0", "json_repair>=0.61.0"] +qwen = ["vllm>=0.8.4", "json_repair>=0.61.0"] # SearchQA data materialization -searchqa = ["datasets>=2.18.0"] +searchqa = ["datasets>=3.0"] # Documentation site docs = ["mkdocs-material>=9.5.0", "mkdocstrings[python]>=0.24.0"] # WebUI dashboard diff --git a/requirements.txt b/requirements.txt index 5db9e702..d22eab1f 100644 --- a/requirements.txt +++ b/requirements.txt @@ -15,7 +15,7 @@ httpx>=0.27.0 # claude-agent-sdk>=0.1.0 # ── Optional: Qwen local model (via vLLM) ──────── -# vllm>=0.4.0 +# vllm>=0.8.4 # ── Optional: tolerant JSON repair for free-form output from non-OpenAI # backends (Claude/Qwen). Without it extract_json() falls back safely and @@ -24,7 +24,7 @@ httpx>=0.27.0 # json_repair>=0.61.0 # ── Optional: WebUI dashboard ──────────────────── -# gradio>=4.0.0 +# gradio>=5.50.0 # ── Optional: Documentation site ───────────────── # mkdocs-material>=9.5.0 diff --git a/skillopt/envs/spreadsheetbench/rollout.py b/skillopt/envs/spreadsheetbench/rollout.py index aff78938..edf8b96c 100644 --- a/skillopt/envs/spreadsheetbench/rollout.py +++ b/skillopt/envs/spreadsheetbench/rollout.py @@ -314,7 +314,8 @@ def process_one( # ── Stage 1: run ReAct agent on test case 1 ───────────────────── result["phase"] = "agent" - work_dir = tempfile.mkdtemp(prefix=f"react_{task_id}_") + safe_task_id = "".join(c if c.isalnum() or c in "-_" else "_" for c in str(task_id)) + work_dir = tempfile.mkdtemp(prefix=f"react_{safe_task_id}_") try: # Copy input so agent works in an isolated directory work_input = os.path.join(work_dir, os.path.basename(ip1)) diff --git a/skillopt_webui/app.py b/skillopt_webui/app.py index 75e0ef1d..eccc93c6 100644 --- a/skillopt_webui/app.py +++ b/skillopt_webui/app.py @@ -46,6 +46,58 @@ def load_config(path: str) -> dict: return yaml.safe_load(f) +def scan_outputs(out_dir: str) -> list: + """Digest experiment results strictly under PROJECT_ROOT. + + The Output Explorer callback. Any path that escapes PROJECT_ROOT is + rejected (empty result) at the point data is read, so a traversal arg + can never read files outside the project. + """ + rows = [] + if not out_dir: + return rows + base = (PROJECT_ROOT / out_dir).resolve() + project_resolved = PROJECT_ROOT.resolve() + try: + base.relative_to(project_resolved) + except ValueError: + return rows + if not base.exists() or not base.is_dir(): + return rows + for bench_dir in sorted(base.iterdir()): + if not bench_dir.is_dir(): + continue + for run_dir in sorted(bench_dir.iterdir()): + if not run_dir.is_dir(): + continue + cfg_file = run_dir / "config.yaml" + score = "—" + steps = "—" + if cfg_file.exists(): + try: + c = yaml.safe_load(cfg_file.read_text()) + steps = str(c.get("train", {}).get("num_steps", "—")) + except Exception: + pass + # Try to find best score from logs + for log_f in run_dir.glob("**/*.jsonl"): + try: + with open(log_f) as f: + for line in f: + d = json.loads(line) + if "score" in d: + score = f"{d['score']:.4f}" + except Exception: + pass + rows.append([ + run_dir.name, + bench_dir.name, + score, + steps, + ]) + return rows + + def config_to_display(cfg: dict) -> str: """Pretty-print config for display.""" return yaml.dump(cfg, default_flow_style=False, sort_keys=False) @@ -602,46 +654,6 @@ def on_refresh(): label="Experiments", ) - def scan_outputs(out_dir): - rows = [] - if not out_dir: - return rows - base = PROJECT_ROOT / out_dir - if not base.exists(): - return rows - for bench_dir in sorted(base.iterdir()): - if not bench_dir.is_dir(): - continue - for run_dir in sorted(bench_dir.iterdir()): - if not run_dir.is_dir(): - continue - cfg_file = run_dir / "config.yaml" - score = "—" - steps = "—" - if cfg_file.exists(): - try: - c = yaml.safe_load(cfg_file.read_text()) - steps = str(c.get("train", {}).get("num_steps", "—")) - except Exception: - pass - # Try to find best score from logs - for log_f in run_dir.glob("**/*.jsonl"): - try: - with open(log_f) as f: - for line in f: - d = json.loads(line) - if "score" in d: - score = f"{d['score']:.4f}" - except Exception: - pass - rows.append([ - run_dir.name, - bench_dir.name, - score, - steps, - ]) - return rows - scan_btn.click(scan_outputs, output_dir, results_table) return app @@ -668,6 +680,10 @@ def main(): parser.add_argument("--host", type=str, default="127.0.0.1", help="Server host. Default is localhost; use 0.0.0.0 " "to expose publicly (no auth, use with care).") + parser.add_argument("--auth-user", type=str, default=None, + help="Username for basic auth (or set SKILLOPT_WEBUI_USER).") + parser.add_argument("--auth-pass", type=str, default=None, + help="Password for basic auth (or set SKILLOPT_WEBUI_PASS).") args = parser.parse_args() if args.host and args.host not in ("127.0.0.1", "localhost", "::1"): @@ -679,8 +695,35 @@ def main(): file=sys.stderr, ) + if args.share: + print( + "⚠ warning: --share creates a public tunnel (gradio.live) with no " + "authentication by default. Anyone with the URL can start/stop " + "training and browse the filesystem via Output Explorer. " + "Use --auth-user / --auth-pass (or SKILLOPT_WEBUI_USER / " + "SKILLOPT_WEBUI_PASS) to require login.", + file=sys.stderr, + ) + + auth_user = args.auth_user or os.environ.get("SKILLOPT_WEBUI_USER") + auth_pass = args.auth_pass or os.environ.get("SKILLOPT_WEBUI_PASS") + # Fail-closed: authentication requires BOTH credentials. Supplying only a + # username or only a password must not silently launch the UI unauthenticated + # (a deployment could expose the training controls without login). + if bool(auth_user) != bool(auth_pass): + print( + "SKILLOPT_WEBUI authentication requires BOTH --auth-user and " + "--auth-pass (or SKILLOPT_WEBUI_USER and SKILLOPT_WEBUI_PASS). " + "Refusing to start with incomplete credentials.", + file=sys.stderr, + ) + sys.exit(1) + auth = (auth_user, auth_pass) if auth_user else None + app = build_ui() launch_kwargs = build_launch_kwargs(args.host, args.port, args.share) + if auth: + launch_kwargs["auth"] = auth app.launch(**launch_kwargs) diff --git a/tests/test_webui_security.py b/tests/test_webui_security.py index 5886da65..a5a28fac 100644 --- a/tests/test_webui_security.py +++ b/tests/test_webui_security.py @@ -55,3 +55,156 @@ def test_main_warns_on_public_host(webui, monkeypatch, capsys): assert "warning" in captured.err.lower() _args, kwargs = launcher.call_args assert kwargs["server_name"] == "0.0.0.0" + + +def test_main_warns_on_share(webui, monkeypatch, capsys): + """--share must emit a public-tunnel warning.""" + webui_mod = webui + launcher = mock.MagicMock() + app_mock = mock.MagicMock() + app_mock.launch = launcher + monkeypatch.setattr(webui_mod, "build_ui", lambda: app_mock) + monkeypatch.setattr(sys, "argv", ["app.py", "--share"]) + + webui_mod.main() + + captured = capsys.readouterr() + assert "share" in captured.err.lower() + assert "public" in captured.err.lower() or "tunnel" in captured.err.lower() + + +def test_main_auth_via_cli_args(webui, monkeypatch): + """--auth-user and --auth-pass must enable Gradio basic auth.""" + webui_mod = webui + launcher = mock.MagicMock() + app_mock = mock.MagicMock() + app_mock.launch = launcher + monkeypatch.setattr(webui_mod, "build_ui", lambda: app_mock) + monkeypatch.setattr(sys, "argv", ["app.py", "--auth-user", "admin", "--auth-pass", "s3cret"]) + + webui_mod.main() + + _args, kwargs = launcher.call_args + assert kwargs.get("auth") == ("admin", "s3cret") + + +def test_main_auth_via_env(webui, monkeypatch): + """SKILLOPT_WEBUI_USER / SKILLOPT_WEBUI_PASS must enable auth without CLI args.""" + webui_mod = webui + launcher = mock.MagicMock() + app_mock = mock.MagicMock() + app_mock.launch = launcher + monkeypatch.setattr(webui_mod, "build_ui", lambda: app_mock) + monkeypatch.setattr(sys, "argv", ["app.py"]) + monkeypatch.setenv("SKILLOPT_WEBUI_USER", "envuser") + monkeypatch.setenv("SKILLOPT_WEBUI_PASS", "envpass") + + webui_mod.main() + + _args, kwargs = launcher.call_args + assert kwargs.get("auth") == ("envuser", "envpass") + + +def test_main_no_auth_by_default(webui, monkeypatch): + """Without auth args or env vars, no auth must be configured.""" + webui_mod = webui + launcher = mock.MagicMock() + app_mock = mock.MagicMock() + app_mock.launch = launcher + monkeypatch.setattr(webui_mod, "build_ui", lambda: app_mock) + monkeypatch.setattr(sys, "argv", ["app.py"]) + monkeypatch.delenv("SKILLOPT_WEBUI_USER", raising=False) + monkeypatch.delenv("SKILLOPT_WEBUI_PASS", raising=False) + + webui_mod.main() + + _args, kwargs = launcher.call_args + assert "auth" not in kwargs or kwargs["auth"] is None + + +def test_scan_outputs_rejects_path_traversal(webui, tmp_path, monkeypatch): + """The scan_outputs callback must not enumerate directories outside PROJECT_ROOT.""" + monkeypatch.setattr(webui, "PROJECT_ROOT", tmp_path) + (tmp_path / "outputs").mkdir() + # Every traversal / escape form is denied at consumption: no rows, no reads. + for bad in ("/../../etc/passwd", "../outside", "outputs/../../../etc", "C:\\Windows"): + assert webui.scan_outputs(bad) == [], f"traversal {bad!r} must be denied" + + +def test_scan_outputs_allows_valid_subdir(webui, tmp_path, monkeypatch): + """scan_outputs must accept directories within PROJECT_ROOT.""" + monkeypatch.setattr(webui, "PROJECT_ROOT", tmp_path) + (tmp_path / "outputs" / "bench1" / "run1").mkdir(parents=True) + (tmp_path / "outputs" / "bench1" / "run1" / "config.yaml").write_text("a: 1\n", encoding="utf-8") + rows = webui.scan_outputs("outputs") + assert rows, "valid in-tree output area must be digested" + + +def test_scan_outputs_callback_consumes_within_project(webui, tmp_path, monkeypatch): + """The registered scan_outputs callback must digest data only inside PROJECT_ROOT + at the point data is actually read (traversal denied, in-tree consumed).""" + monkeypatch.setattr(webui, "PROJECT_ROOT", tmp_path) + # A traversal arg must be denied at consumption: no rows, no data read. + assert webui.scan_outputs("/../../etc/passwd") == [] + assert webui.scan_outputs("../outside") == [] + # A valid in-tree output area is digested (config.yaml read per run dir). + (tmp_path / "outputs/bench1/run1").mkdir(parents=True) + (tmp_path / "outputs/bench1/run1/config.yaml").write_text("alpha: 1\n", encoding="utf-8") + rows = webui.scan_outputs("outputs") + assert rows, f"expected rows from a valid in-tree output area, got {rows!r}" + + +def test_main_rejects_incomplete_cli_auth_user_only(webui, monkeypatch): + """--auth-user without --auth-pass must fail closed (never launch).""" + webui_mod = webui + launcher = mock.MagicMock() + app_mock = mock.MagicMock() + app_mock.launch = launcher + monkeypatch.setattr(webui_mod, "build_ui", lambda: app_mock) + monkeypatch.setattr(sys, "argv", ["app.py", "--host", "0.0.0.0", "--auth-user", "admin"]) + with pytest.raises(SystemExit): + webui_mod.main() + launcher.assert_not_called() + + +def test_main_rejects_incomplete_cli_auth_pass_only(webui, monkeypatch): + """--auth-pass without --auth-user must fail closed (never launch).""" + webui_mod = webui + launcher = mock.MagicMock() + app_mock = mock.MagicMock() + app_mock.launch = launcher + monkeypatch.setattr(webui_mod, "build_ui", lambda: app_mock) + monkeypatch.setattr(sys, "argv", ["app.py", "--host", "0.0.0.0", "--auth-pass", "s3cret"]) + with pytest.raises(SystemExit): + webui_mod.main() + launcher.assert_not_called() + + +def test_main_rejects_incomplete_env_auth_user_only(webui, monkeypatch): + """Only SKILLOPT_WEBUI_USER set must fail closed (never launch).""" + webui_mod = webui + launcher = mock.MagicMock() + app_mock = mock.MagicMock() + app_mock.launch = launcher + monkeypatch.setattr(webui_mod, "build_ui", lambda: app_mock) + monkeypatch.setattr(sys, "argv", ["app.py", "--host", "0.0.0.0"]) + monkeypatch.setenv("SKILLOPT_WEBUI_USER", "envuser") + monkeypatch.delenv("SKILLOPT_WEBUI_PASS", raising=False) + with pytest.raises(SystemExit): + webui_mod.main() + launcher.assert_not_called() + + +def test_main_rejects_incomplete_env_auth_pass_only(webui, monkeypatch): + """Only SKILLOPT_WEBUI_PASS set must fail closed (never launch).""" + webui_mod = webui + launcher = mock.MagicMock() + app_mock = mock.MagicMock() + app_mock.launch = launcher + monkeypatch.setattr(webui_mod, "build_ui", lambda: app_mock) + monkeypatch.setattr(sys, "argv", ["app.py", "--host", "0.0.0.0"]) + monkeypatch.setenv("SKILLOPT_WEBUI_PASS", "envpass") + monkeypatch.delenv("SKILLOPT_WEBUI_USER", raising=False) + with pytest.raises(SystemExit): + webui_mod.main() + launcher.assert_not_called()