[devin-ai-integration]

Summary

Adds a root AGENTS.md that defines a Unit Testing Agent role for this repo. It ports the instructions drafted locally (.devin/agentlab/AGENTS.md), whose cloud handoff failed because of an expired API key. This is a docs-only change: no edits to src/, tests/, deps, or CI.

What the file covers:

• Discovery before writing: README, pyproject.toml, pytest.ini markers (pricy, slow), the CI workflows, and nearby tests.
• Reproducible env that matches CI: Python 3.11, uv sync --frozen --extra dev, Playwright Chromium, MiniWoB via make miniwob.
• Test-only scope by default: no changes to src/, pyproject.toml/uv.lock, CI, config, or unrelated local changes.
• Fixtures: synthetic data only, with mocked LLM providers, pricing, Ray/Dask, and the browser. Shared state such as tracking.TRACKER is restored. Modules that can't be imported (gaia, multitool_gym, tapeagent, analyze/tapes) are skipped.
• Confirmed source defects: captured as xfail(strict=True) tests and reported. Production code stays unchanged.
• Parallel workers: one file per worker, no root conftest.py, based on main.
• Verification: serial, -n 5, randomized order (uv run --with pytest-randomly), black --check, the surrounding suite, and coverage (uv run --with pytest-cov). Includes the known torch double-import and sys.gettrace coverage gotchas.
• Honest reporting: commands run with their counts, coverage for the targeted module only, every failure and blocker, and review status.

I ran each verification command in the file on the py3.11 uv env against tests/llm/test_tracking.py, and all of them passed.

Devin-Org: engineering

Link to Devin session: https://devin-gtm.devinenterprise.com/sessions/261a4ca3bed3420abcdafc8e3454076b
Open in Devin Desktop: https://devin-gtm.devinenterprise.com/desktop/session/261a4ca3bed3420abcdafc8e3454076b?variant=devin
Requested by: @danigrim

---

Devin Review

Status	Commit
⚪ Not started	—

Run Devin Review

---

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

🔍 Devin Review: 1 flag

Not posted on this PR by your GitHub settings — view it in Devin Review. (Configure)