fix: prevent Daytona rollout failures (HUD-2206)

[jdchawla29]

Summary

• Launch Daytona environment servers with uv run hud serve so scaffold images resolve the SDK from /app/.venv in fresh process sessions.
• Disable Daytona's inherited auto-stop interval for both image-backed and snapshot-backed ephemeral rollout sandboxes.
• Cover the injected serve command and both sandbox creation parameter shapes in provider tests.

Root cause

Daytona process sessions start a fresh shell rather than running the image CMD, so /app/.venv/bin is not added to PATH. The previous injected hud serve command therefore failed for images produced by hud init, whose Dockerfile installs the SDK with uv sync and launches it through uv run.

Ephemeral sandboxes also inherited the organization default 15-minute auto-stop interval. Running processes do not count as activity, and stopping an ephemeral sandbox deletes it, so long rollouts could disappear mid-run. The provider already owns sandbox cleanup through its finally deletion path, so rollout sandboxes now set auto_stop_interval=0.

Impact

Daytona rollouts built from standard HUD scaffold images start reliably and remain available until the rollout completes or fails, after which the provider deletes the sandbox.

Validation

• uv run --no-sync pytest hud/eval/tests/test_docker_provider.py -q — 17 passed
• uv run --no-sync ruff check hud/eval/runtime.py hud/eval/tests/test_docker_provider.py
• uv run --no-sync ruff format --check hud/eval/runtime.py hud/eval/tests/test_docker_provider.py
• git diff --check

---

Note

Medium Risk
Changes Daytona sandbox lifecycle and startup commands for all Daytona rollouts; scoped to the Daytona provider with test coverage.

Overview
Fixes Daytona rollout sandboxes failing to start or vanishing mid-run.

The default injected serve command now prepends PATH="$PWD/.venv/bin:$PATH" before hud serve, so scaffold images that install the SDK with uv sync can find hud in fresh Daytona process sessions (where /app/.venv/bin is not on PATH).

Image-backed and snapshot-backed ephemeral sandboxes now pass auto_stop_interval=0, so org default idle auto-stop does not tear down rollout boxes while the env server runs; cleanup stays on the provider’s delete path.

Provider tests assert the updated execute command, auto_stop_interval=0 on both create param shapes, and add coverage for snapshot sandboxes.

^{Reviewed by Cursor Bugbot for commit 4e8841d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5e8ad2ea2a

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".