[sriram687]

Experiment Lifecycle Callbacks

Closes #417

Summary

ExpArgs.run() now broadcasts structured events at five hook points, so researchers can attach monitoring, safety guardrails, and dashboard integrations without modifying any core loop code.

Motivation

As described in #417, the existing loop is a closed, synchronous black box. In practice this causes three problems:

Problem	Impact
No live visibility	Long multi-hour benchmark runs can't be monitored
No runtime safety guardrails	Nothing stops runaway LLM token usage
No clean integration path	Streaming to W&B, MLflow, Streamlit, etc. requires monkey-patching

Backward Compatibility

No existing behaviour changes. The callbacks field defaults to [], so CallbackList iterates over an empty list — zero overhead. All existing tests pass unchanged.

Verification

python -m black . --check                          →  95 files would be left unchanged.
pytest test_callbacks.py test_callbacks_cloud.py   →  60 passed in 2.38s

Usage Example

from browsergym.experiments import ExpArgs, EnvArgs
from browsergym.experiments.callbacks import ConsoleProgressCallback, BudgetLimitCallback
exp = ExpArgs(

agent_args=...,

env_args=EnvArgs(task_name="webarena.42", max_steps=30),

callbacks=[

ConsoleProgressCallback(),

BudgetLimitCallback(max_tokens=50_000),

],

)

exp.run()

PROOF OF WORK

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[sriram687]

@Surya-5555 kindly review the integrations.......!!!