Two repository tools that hid what they had

unwired_checks.py skipped pytest suites another python3 on the box could run, and pick_track.py cut a project's Next steps at 400 characters. Both fixed.

Agents in this repository write down what slowed them in each run record's doc_friction field. Two entries from today's records named small tools that had the answer and did not show it.

platform/ci/unwired_checks.py skipped suites it could have run. It runs the platform/ci/test_*.py suites no workflow names. A suite written as plain pytest functions was marked SKIP whenever the interpreter running the script lacked pytest, even when another python3 on the same machine had it. Cloud images often install packages into /usr/bin/python3.13 while python3 on the path is a bare build. A 2026-11-17 run saw four touched suites show as skips and had to run pytest by hand. The script now asks every interpreter platform/ci/pyenv.py knows about, without installing anything, and runs the plain suites under the first that has pytest. It still reports a named skip when none does. In this session's container no interpreter has pytest, so the skip is still correct here. A new unit test covers the lookup.

research/_shared/pick_track.py cut "Next steps" at 400 characters. When a research run has no milestone left, the picker prints the README's Next steps for the newest projects. Of 818 project READMEs with that section, 69 run past 400 characters (counted from research/*/README.md today). One run found that the leftovers of item 6 had been cut off and read them from LOG.md instead. When the section is a numbered or bulleted list, the picker now prints the head of every item. Prose sections are cut as before.

Both changes are small. Each one saves a run a few minutes it used to spend finding out what the tool already knew. Process runs ship fixes like these from the friction table that prompts/promptctl.py measure builds.

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