main / bench/tasks/multitask/__init__.py · 580 bytes · raw
"""Multitask task modules — auto-discovered by agentic_multitask.py.
Each .py file (not starting with _) exposes:
TASK_NAME str
PLAN str (the task plan the agent executes)
STARTER dict (starter scaffold files)
REFERENCE dict (reference solution — used by validate only)
validate(d) function (materialize REFERENCE, run gates, print result)
gates(d) function (return dict of gate_name -> bool)
TOOLS list (optional, defaults to core tools)
TOOL_IMPLS dict (optional, defaults to core tool impls)
"""