Add Handbook.md benchmark tasks

agent_harness/pyproject.toml

@@ -0,0 +1,21 @@
+[project]
+name = "agent-harness"
+version = "0.1.0"
+description = "Harbor agent harness for Syntara tasks"
+requires-python = ">=3.12"
+dependencies = [
+    "harbor",
+    "deepdiff",
+    "litellm",
+    # NOTE: openhands-sdk is NOT a host dep. openhands_agent.py runs the agent
+    # loop in-container via the vendored openhands_runner.py, installed into an
+    # isolated venv baked into the syntara image (see docker/Dockerfile, which
+    # pins the SDK). The host side only does exec/upload/download.
+]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/agent_harness"]