refactor: collapse run_smoke.py onto the PyAutoHands runners (356 → 119 lines)

[Jammy2211]

Summary

Phase 2 of PyAutoHands#260 — the last of the ten vendored copies. This file
becomes a thin delegator over autohands/run_python.py (scripts) and
autohands/run.py (notebooks): the shape the HowTo repos have used all along,
and the shape the four *_workspace_test repos took in phase 1.

The copies were not merely duplicated, they were expensive. Each of the last
three fixes had to be swept across every copy by hand — the env-resolver fork
(PyAutoHands#185), the per-script timeout and process-group kill (#226/#227),
the jupyter guard. The HowTo tier needed none of those sweeps, precisely because
it holds no logic.

The notebook leg is preserved in full

Promoted into PyAutoHands rather than reimplemented:

• --no-write-back — executes a throwaway copy, so a smoke run never
  modifies the committed notebooks. The shared runner already pins the kernel's
  cwd to the repo root, which is what this file's staged-copy-at-root trick
  existed to achieve — so that trick is superseded, not ported.
• --retry-from scripts — keeps the regenerate-from-source recovery for a
  stale notebook. A TIMEOUT is never retried, a clean sys.exit(0) skip guard
  is already a PASS and never reaches it, and the retry's verdict replaces the
  first attempt's so one notebook yields one result.
• Missing toolchain stays one FAIL with the run continuing: the shared
  runner invokes sys.executable run_notebook.py, never a bare jupyter, so
  the abort-with-no-summary failure mode is structurally absent.

Behaviour preserved — verified, not assumed

• Env resolution is identical. All 8 listed scripts resolved under both the
  old (relative) and new (absolute) path form and diffed: 0 differences.
• No profile args, so extra-args support changes nothing.
• Every entry resolves — 8/8 scripts, 2/2 notebooks — and both listed
  notebooks' source scripts exist for the retry path.
• The exit code is the worst of the two legs, so a failing notebook cannot
  be masked by passing scripts.

no_run.yaml is deliberately not applied — and this repo shows why

It is policy for the release mega-run and notebook generation; the smoke lists
are policy for this gate. This repo has searches/mcmc.py in both its
allowlist and no_run.yaml, and it runs in smoke today — as does
searches/mcmc.ipynb, whose source script is that same file. Filtering the
allowlist by no_run would have silently dropped both. Measured and fixed in
PyAutoHands#262.

Test Plan

• Two-leg delegator exercised end-to-end on a fixture: script leg, notebook
  leg, a stale nested notebook regenerating from its source and passing, the
  committed notebooks byte-identical afterwards, and the exit code correct in
  all three pass/fail combinations
• Phase 1's four *_workspace_test collapses are merged and green
• This repo's own smoke gate on this PR — the real verification

Requires PyAutoHands#261, #262 and #263, all merged. #263 matters here in
particular: both listed notebooks are in subdirectories, and before that fix
the regenerate-and-retry looked its source up by bare filename and never found it.

Generated by the PyAutoLabs agent workflow.

---

Generated by Claude Code