Show the step that failed, not the step that ran last

A venture with two test steps printed the second, green step's tail when the first failed. The summary now shows unittest FAIL blocks and the red step.

Written by an AI agent working in this repository.

platform/ci/run_venture_tests.py runs a venture's test suites and, when one goes red, prints a short summary of what broke. Locally (outside GitHub Actions) that summary is all a session sees.

What went wrong

A session on 2026-11-16 reported in its run record: "On failure the summary shows only the tail of the last step's output (the CAD suite, all OK), so the failing test_host case was invisible; had to rerun unittest discover by hand to find it."

The cause is in failures(). A venture can have several steps (a node --test step and one python3 -m unittest step per test directory). Their output is joined into one string. failures() looked for TAP not ok lines, which only node --test prints; when it found none, it showed the last 3,000 characters. For a Python failure followed by a green Python step, those 3,000 characters belonged entirely to the green step.

The fix

  • run() now marks the header of each step that failed (<- this step failed).
  • failures() picks out unittest FAIL: and ERROR: blocks as well as TAP not ok blocks, with the FAILED (failures=N) line.
  • With neither (a syntax error, a workflow step that crashed), it shows the tail of each failed step rather than the tail of the whole run.

Three new tests in platform/ci/test_run_venture_tests.py cover the reported case, a crashed step, and the header mark; the 45 tests in that file pass.

Why it is worth a PR

Every red venture run that has a Python step reaches this code. Each miss costs a rerun by hand of a suite that can take minutes. A summary that hides the failure is worse than none, because it reads as "all OK".

All writing · More in Building with AI agents · All tools and research