opencode envs

[mikasenghaas]

Summary

• CLI agent env improvements: Refactored cli_agent_env.py with better error handling — agent failures and background job polling errors are now logged/raised properly, per-request timeout increased to 1h
• OpenCode env updates: Enhanced opencode_env.py with adjusted wording, task system prompt, and agent log collection; added new opencode_qa_env.py variant
• Sandbox mixin enhancements: Added file upload helpers (read_file, file upload to sandbox), simplified poll_job, and increased default timeouts from 30s → 1h
• Hybrid math rubric overhaul: Added sandbox-based scoring support, offline difficulty filtering, remote math verification, rubric teardown lifecycle hook, and deregister/register flow from env → rubric
• Logging improvements: Guard debug-level log string building behind isEnabledFor(DEBUG) in interception_utils to skip O(n_messages) work when debug is off; compact info logs; moved logging utils
• Rubric lifecycle: Added teardown support to rubric.py and decorator discovery utils for cleanup hooks

Type of Change

• New feature (non-breaking change which adds functionality)

Testing

• All existing tests pass when running uv run pytest locally.
• New tests have been added to cover the changes

Checklist

• My code follows the style guidelines of this project as outlined in AGENTS.md
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

Note

Medium Risk
Touches sandbox lifecycle/teardown and scoring flow (including optional deferred sandbox deletion and in-sandbox scoring), which can impact resource cleanup and evaluation correctness. Changes are localized to experimental envs/rubrics but affect long-running timeouts and error propagation.

Overview
Improves the experimental CliAgentEnv/OpenCode environments with more robust agent execution handling: better background-job start/poll error surfacing (AgentError), simplified tunnel management, longer default request/provider timeouts (to 1h), and richer per-rollout info logging (tool call counts, duration, exit code, errors).

Adds post-rollout data capture and dataset controls: OpenCodeEnv now collects and stores agent logs from the sandbox; new OpenCodeQAEnv can optionally filter datasets by a difficulty/reward column range.

Extends SandboxMixin with sandbox file I/O helpers (upload_file, upload_content, read_file, upload_bundle) and explicit register_sandbox/deregister_sandbox APIs; CliAgentEnv gains keep_sandbox_for_scoring to defer deletion while still removing the sandbox from active tracking.

Updates HybridMathRubric defaults/behavior (fixed math-verify timeouts, optional judge fallback), and introduces RemoteHybridMathRubric to run math verification inside the sandbox and delete the sandbox after scoring via a cleanup hook.

Optimizes interception logging by guarding debug log construction and centralizing string truncation in logging_utils.truncate.

^{Written by Cursor Bugbot for commit c37da47. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}