Merge branch 'main' into star-prompt

Author: reacher-z

scripts/build-aliases.py

@@ -55,11 +55,30 @@ def root_version() -> str:
         raise RuntimeError("could not find version in root pyproject.toml")
     return m.group(1)
 
-# The six alias distribution names successfully registered on PyPI.
-# Three other candidates (``openclaw-bench``, ``claw-bench-harness``,
-# ``harnessbench``) were rejected by PyPI's similarity check as too
-# close to names we had already taken — so they live nowhere and are
-# omitted here on purpose.
+# Alias distribution names successfully registered on PyPI.
+#
+# First wave (packaging PR): harness-bench, clawbench-eval, clawbench-cli,
+# openclawbench, clawbench-harness, claw-harness, nail-clawbench.
+#
+# Second wave (adjacent harness/agent names): harnessos, r2agent, claw-ai,
+# claw-agent, claw-eval.
+#
+# Third wave (research-themed squats): everyday-bench, everyday-agent,
+# life-bench, realtask-bench, web-harness, task-harness, video-mcq,
+# mcq-bench, vlm-judge, video-judge, nail-bench, nail-agent, nail-eval,
+# nail-group.
+#
+# Names we tried and couldn't take:
+#   - 403 (already owned): harness, browser-use, computer-use, claw,
+#     clawbot, openclaw, claw-ops, agent-harness
+#   - 400 similarity (too close to something we just took or pre-existing):
+#     openclaw-bench, claw-bench-harness, harnessbench, harness-os, clawai,
+#     clawagent, claweval, lifebench, agentharness, webharness, taskharness,
+#     videomcq, mcqbench
+# For hyphen/underscore variants of names we DID take (``claw-ai`` ↔
+# ``claw_ai``), PEP 503 normalization means pip resolves them to the same
+# distribution anyway — so ``pip install claw_ai`` still lands on our
+# package.
 ALIAS_NAMES = [
     "harness-bench",
     "clawbench-eval",
@@ -68,7 +87,43 @@ def root_version() -> str:
     "clawbench-harness",
     "claw-harness",
     "nail-clawbench",
+    "harnessos",
+    "r2agent",
+    "claw-ai",
+    "claw-agent",
+    "claw-eval",
+    "everyday-bench",
+    "everyday-agent",
+    "life-bench",
+    "realtask-bench",
+    "web-harness",
+    "task-harness",
+    "video-mcq",
+    "mcq-bench",
+    "vlm-judge",
+    "video-judge",
+    "nail-bench",
+    "nail-agent",
+    "nail-eval",
+    "nail-group",
+    # Fourth wave — harness-hub was the only slot the permutation sweep
+    # cracked. Wave 5 (computer/browser/harness/claw × bench/eval/agent/
+    # design, plus agentic/operator/mcp/sandbox/runtime/swarm/rollout/
+    # deep-research hot-word combos, 75 candidates) returned zero: the
+    # AI-agent generic-permutation namespace is effectively fully claimed
+    # on PyPI.
+    "harness-hub",
+    "r2-harness",
+    "scaling-law",
 ]
+# Exhaustive negative result: waves 4-8 tried ~200 additional permutations
+# (generic {computer|browser|harness|claw} × {bench|eval|agent|design},
+# agent-ecosystem hot words like mcp/operator/agentic/sandbox/runtime,
+# unusual noun-noun compounds, numeric suffixes, project-ish compounds).
+# Only ``harness-hub`` stuck. Everything else was either 403 already-owned
+# or 400 similarity-blocked. Don't waste cycles retrying generic forms —
+# the AI-agent namespace is saturated on PyPI. If a specific name matters,
+# check pypi.org first rather than adding it here speculatively.
 
 PYPROJECT_TEMPLATE = '''\
 # Auto-generated by scripts/build-aliases.py — DO NOT EDIT BY HAND.

src/clawbench/__init__.py

@@ -23,6 +23,28 @@
     "harness-bench",
     "openclawbench",
     "claw-harness",
+    "harnessos",          # second-wave defensive squats
+    "r2agent",
+    "claw-ai",
+    "claw-agent",
+    "claw-eval",
+    "everyday-bench",     # third-wave research-themed squats
+    "everyday-agent",
+    "life-bench",
+    "realtask-bench",
+    "web-harness",
+    "task-harness",
+    "video-mcq",
+    "mcq-bench",
+    "vlm-judge",
+    "video-judge",
+    "nail-bench",
+    "nail-agent",
+    "nail-eval",
+    "nail-group",
+    "harness-hub",        # fourth-wave singleton winner
+    "r2-harness",         # ninth wave
+    "scaling-law",
     "claw-bench",         # original primary (blocked; left for future)
     "clawbench",          # original alias  (blocked; left for future)
 ):

src/clawbench/tui.py

@@ -6,6 +6,7 @@
 import os
 import platform
 import shutil
+import stat
 import subprocess
 import sys
 from pathlib import Path
@@ -1515,6 +1516,7 @@ def main() -> None:
                 questionary.Choice("Batch run   (models x cases)", value="batch"),
                 questionary.Choice("Human mode  (no agent, noVNC)", value="human"),
                 questionary.Choice("Configure models", value="configure"),
+                questionary.Choice("Configure secrets", value="configure_secrets"),
                 questionary.Choice("Change theme", value="theme"),
                 questionary.Choice("Star us on GitHub", value="star"),
                 questionary.Choice("Exit", value="exit"),
@@ -1546,6 +1548,10 @@ def main() -> None:
             models = load_models()
             continue
 
+        if mode == "configure_secrets":
+            mode_configure_secrets()
+            continue
+
         # Every run mode (including Human) needs a live engine. If it
         # wasn't fixable earlier, try one more time now — the user may
         # have started Docker Desktop in another window, or we may be

test-driver/run.py

@@ -1,6 +1,7 @@
 """ClawBench single test-case driver."""
 
 import argparse
+import hashlib
 import json
 import os
 import re
@@ -722,7 +723,9 @@ def main():
     if not task_file.exists():
         print(f"ERROR: {task_file} not found")
         sys.exit(1)
-    task = json.loads(task_file.read_text())
+    task_bytes = task_file.read_bytes()
+    task = json.loads(task_bytes)
+    task_json_sha256 = hashlib.sha256(task_bytes).hexdigest()
 
     case_name = task_dir.name
     time_limit_s = task["time_limit"] * 60
@@ -835,6 +838,7 @@ def handle_sigint(sig, frame):
             meta = {
                 "test_case": case_name,
                 **(task.get("metadata") or {}),
+                "task_json_sha256": task_json_sha256,
                 "instruction": task["instruction"],
                 "model": "human",
                 "thinking_level": None,
@@ -851,6 +855,7 @@ def handle_sigint(sig, frame):
             meta = {
                 "test_case": case_name,
                 **(task.get("metadata") or {}),
+                "task_json_sha256": task_json_sha256,
                 "instruction": task["instruction"],
                 "model": model_cfg["model"],
                 "thinking_level": model_cfg.get("thinking_level"),