ci(prek): quieten the CI log — drop --verbose, skip the identity hook

[potiuk]

Summary

Two sources of noise in the prek CI log, around the one thing a reader opens
that log for.

-  prek run --show-diff-on-failure --color=always --all-files
-  --verbose --skip workspace-pytest
+  prek run --show-diff-on-failure --color=always --all-files
+  --skip workspace-pytest --skip identity

--skip identity. The identity meta-hook echoes every file passed to the
static checks. Useful locally when working out why a hook did or did not fire,
but on --all-files it prints the entire repository into the log before any
result appears. It stays enabled locally and is skipped only in CI.

Drop --verbose — see below, because the reason it was there turns out not
to hold.

--verbose never bought live progress

The flag carried a comment saying that without it, prek renders a progress UI a
non-TTY log buffers to the end. The first half is true; the conclusion is not.
Measured on this repo, timestamping each line and comparing against total
runtime:

Invocation	First output	Total
plain	19s	19s
--verbose	19s	20s
--no-progress	19s	19s
-q	(no output)	18s

First output and total runtime coincide in every mode. prek emits nothing
until the run finishes, with or without the flag.

It streams only when stdout is a TTY, where it draws an in-place progress UI
— spinner, Running hooks..., cursor-control escapes redrawing lines. Against a
pipe (CI) it suppresses that renderer and prints one block at the end. Confirmed
by running it under script to allocate a pty, which does stream — and fills the
output with ESC[2K / ESC[2A redraw sequences.

So in CI the flag's only effect was to add every passing hook's stdout to that
final block: several hundred lines wrapped around nothing. Failures still print
their output, and --show-diff-on-failure still shows what a fixer hook changed.

Real streaming would mean allocating a pty in the workflow (script -qec …),
which trades a clean block for a log full of ANSI redraw escapes. Waiting for the
block is the better trade.

For completeness: PYTHONUNBUFFERED does not apply — prek is a Rust binary, and
it captures each hook's stdout and replays it after the hook completes, so a
Python hook's own buffering is not in the path either.

A trap worth documenting

My first attempt wrote --skip workspace-pytest,identity. Prek takes a single
HOOK|PROJECT per occurrence, so a comma-joined value matches no hook — and it
only warns:

warning: selector `--skip=workspace-pytest,identity` did not match any hooks

then runs identity anyway. The comma form reads as working while skipping
nothing. Caught by running the command instead of trusting the edit; the workflow
now carries a note so the next person does not repeat it.

Why workspace-pytest stays skipped (unchanged behaviour)

.github/workflows/tests.yml already runs pytest as a data-driven matrix — one
parallel job per workspace member, named after the member, so a failure points at
the culprit. The workspace-pytest hook runs the same tests bundled serially
inside the prek job, where a failure is buried. Path filtering on PRs is not a
hole: four shared build inputs (pyproject.toml, uv.lock,
tools/dev/run-workspace-check.sh, tests.yml) force the full matrix, and
non-PR events always run everything. The hook still runs locally on
git commit / prek run, where there is no matrix to defer to.

Test plan

• Ran the exact post-change CI command locally — exit 0, identity absent
• Timestamped every output line across four invocations to establish the
  buffering behaviour above, rather than asserting it
• Confirmed under a pty that streaming exists but costs ANSI redraw escapes
• Confirmed the comma form silently skips nothing

An earlier revision of this PR claimed per-hook lines "still stream" without
--verbose. That was wrong — they never streamed in CI either way — and the
commit and workflow comment have been corrected.

🤖 Generated with Claude Code

https://claude.ai/code/session_01So3JRGXrbqSGrohtZuHWKg