rr

---
name: rr
description: "Use when running tests or editing remote test tasks with rr."
---

# Running tests with rr

This project runs its test suites on remote machines through rr. The task names and when to use them are in the rr section of `AGENTS.md`. Run `rr --help` for command syntax and `rr doctor` for setup and connectivity checks. The Claude marketplace plugin is not installed by this adapter.

- Use the named task (`rr <task>`) over an ad-hoc `rr run "..."`. Tasks run from the project root; pass test filters after `--` (`rr test -- -k name`).
- Scope runs while iterating (a file, a test name); save the full suite for the final check.
- Read the final `{"type":"result",...}` event on stderr: `details.failures` has each failing test with file and line, `details.no_tests` means nothing was collected (not a pass), and `details.log_file` has the full output. Read the log instead of rerunning.
- If rr fails before the command starts, branch on `error.code` (`LOCK_HELD`: another run holds the host, wait or `rr unlock <host>` if the holder is gone; `DEPENDENCY_MISSING`: a tool is missing on the host, see `rr provision`).
- Never nest rr inside `rr exec`, and don't pipe test output through `tail`/`grep` without pipefail: the pipe's status hides failures.