CliAgentEnv: add SandboxMixin, refactor InterceptionServer

[rasdani]

Description

Necessary error handling and retrying to make CliAgentEnv work at scale.

Includes updates from #811

Extract sandbox lifecycle management into a reusable mixin for CliAgentEnv.
Can be used in SandboxEnv too.

Changes

• Add SandboxMixin class with ThreadedAsyncSandboxClient, retry logic, and bulk cleanup
• CliAgentEnv now inherits SandboxMixin alongside MultiTurnEnv
• Add active_sandboxes tracking to prevent orphaned sandboxes on Ctrl+C
• Add @vf.teardown hooks for bulk sandbox cleanup and client teardown
• Add retry with exponential backoff for sandbox API calls
• Use connection-pooled ThreadedAsyncSandboxClient instead of per-call AsyncSandboxClient

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Test improvement

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

Additional Notes

---

Note

Medium Risk
Touches core sandbox lifecycle/interception paths (CliAgentEnv/HarborEnv) and changes teardown/retry behavior, which could affect stability and cleanup semantics despite added tests.

Overview
Refactors sandbox-backed agent environments to be more reliable at scale by introducing a reusable SandboxMixin that centralizes sandbox creation/deletion, retries with exponential backoff, connection-pooled ThreadedAsyncSandboxClient, and teardown-time bulk cleanup to prevent orphaned sandboxes.

CliAgentEnv now uses SandboxMixin and a new shared InterceptionServer helper (moved into verifiers.utils.interception_utils) to simplify interception/streaming response handling and reduce per-call client creation; HarborEnv/TerminusHarborEnv are updated to the new post_sandbox_setup(state) signature and to run Harbor test/reward collection via background jobs with retry and better error handling (including skipping tests on prior infra errors).

Adds/updates the opencode_harbor environment to support an uploaded system prompt and disabled tool configuration via generated opencode.json, includes a bundled prompt.txt, and adds tests to validate the generated config renders as valid JSON after shell expansion; docs are updated to mention SandboxMixin tuning knobs, and core deps now require prime-tunnel>=0.1.0.

^{Written by Cursor Bugbot for commit ae3cd0c. This will update automatically on new commits. Configure here.}

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
3 out of 4 committers have signed the CLA.

✅ rasdani
✅ kcoopermiller
✅ willccbb
❌ cursoragent
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[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.}