A HUD v6 environment where an agent fixes a bug in a Python web app, graded by a hidden pytest
suite. The env publishes a sandboxed ssh workspace — the agent's harness brings its own bash
and file tools and drives it over SSH; the env ships no agent tools itself.
Grading uses a 3-branch design: every task has {task}_baseline (the agent's starting state),
{task}_test (hidden tests), and {task}_golden (the reference fix). setup_task checks out the
baseline; the grader applies the hidden test patch and runs pytest — all pass → reward 1.0. The
agent never sees the tests or the solution.
Four hand-picked bugs in the coding-template-sample repo:
| Task | Difficulty | Bug |
|---|---|---|
sentry-fix |
Basic | KeyError on a missing/null user profile |
notif-bug |
Medium | Event-type delimiter _ vs . breaks routing |
settings-v2 |
Hard | CompactDict filters falsy values during iteration |
webhook-bug |
Hard | Shared-list mutation corrupts channel resolution |
uv sync
hud set HUD_API_KEY=your-key-here # CLI auth, get one at hud.ai/project/api-keys# local — clones the target repo into a per-process temp dir (macOS + Linux)
hud eval tasks.py claude --task-ids sentry-fix -y --runtime local
# deploy once, then run hosted
hud deploy .
hud eval tasks.py claude --runtime hud --fulluv run pytest tests/ -qOffline tests drive the 3-branch grading (baseline fails, golden passes) with no Docker.
See the full docs for tasks, evaluation, and scaling.