feat(python): recognize poetry run, filtering output like uv run

[albatrossflyon-coder]

Summary

Fixes #2969.

poetry run <tool> never reached the registry — per the issue, it's the single most frequent unhandled command in rtk discover output (408 calls in 30 days, ahead of everything else unhandled), leaving ~96% noise unfiltered on a typical verbose pytest run through it.

Approach

Added poetry as a first-class command (rtk poetry), mirroring exactly how uv run already works via rtk uv — not as a transparent-prefix strip. poetry run wraps an arbitrary tool the same way uv run does, so the same generic traceback/error-block filter applies regardless of what's actually running underneath (pytest, ruff, black, whatever). poetry_cmd.rs mirrors uv_cmd.rs structurally for exactly that reason: same filtering logic, same passthrough behavior for non-run invocations, same exit code propagation. This repo's existing convention is one file per tool (checked — no shared generic wrapper module exists yet for pip/uv/etc.), so this follows that rather than introducing a new pattern.

Wiring (four places, in case any of it's useful context for review)

1. New RtkRule (^poetry\s+run → rtk poetry, prefix poetry) so rewrite_command picks it up the same way it already does for uv run.
2. Commands::Poetry variant + dispatch to poetry_cmd::run.
3. Added to is_operational_command's whitelist — this gates hook-pipeline integrity verification, and per that function's own doc comment a forgotten command here fails open (no check), so it needs the same treatment as every other command already in that list, not just enough to make the rewrite itself work.
4. Added to PASSTHROUGH in the test_every_subcommand_is_classified consistency check — caught this one by actually running the full suite before opening the PR, not just the commands I wrote.

Verified live

$ rtk rewrite "poetry run pytest tests/"
rtk poetry run pytest tests/
$ rtk rewrite "poetry run ruff check ."
rtk poetry run ruff check .
$ rtk rewrite "poetry run black ."
rtk poetry run black .

(All exit 3 — "ask", the documented default for any rewritten command without an explicit permission rule, same as e.g. git status. Not a bug — see the exit-code-protocol tests in rewrite_cmd.rs.)

Full execution path against a real poetry project:

$ rtk poetry --version                                    # passthrough (non-run)
Poetry (version 2.4.1)

$ rtk poetry run python -c "print('hello')"                # success, filtered
ok

$ rtk poetry run python -c "raise RuntimeError('boom')"     # failure, filtered
Traceback (most recent call last):
File "<string>", line 1, in <module>
raise RuntimeError('boom')
RuntimeError: boom
$ echo $?
1

Testing

• New unit tests in poetry_cmd.rs: success-noise suppression, traceback truncation, failure-summary-line preservation, failure fallback
• cargo test: 2448 passed, 8 ignored, 0 failed (initially 1 failed — the whole-repo test_every_subcommand_is_classified consistency check — fixed by point 4 above; leaving it in this summary since it's a good example of why running the full suite before opening matters)
• cargo clippy --all-targets -- -D warnings: clean

[tapheret2]

Review: feat(python): recognize poetry run, filtering output like uv run

Files: src/cmds/python/poetry_cmd.rs, src/discover/rules.rs, src/main.rs
Size: +348 / -1

Notes

• Static pass on the patch: no obvious blockers from the diff alone.

Thanks @albatrossflyon-coder — independent review on the patch.