What — closed-bug-regression-audit was deferred, not skipped, during the 2026-08 repo improvement sweep. It has never run against this repo's 91 closed bug issues.
Why it matters here — 91 closed bugs is the largest regression-audit surface of the six repos swept, and the only one where the lens produces signal at all; the other five have zero closed bug issues. It was deferred because a full 3-pass parallel-subagent run does not belong in a corner of a config-and-docs sweep, not because the surface is uninteresting.
The repo has the shape that makes this productive: 695 commits, 45 contributors, 30 published releases, 151 test files against 283 source files, and a pytest.yml that runs on every PR. A bug that was fixed without a regression test can silently come back through any of those 45 contributors.
There is a specific reason to think the yield is non-zero. comfy_cli/utils.py shipped a knowingly-unfiltered tar.extractall() for the entire period cpython#107845 was open, with no test covering extraction safety either way — the regression test that proves it now only exists because of #734, filed by this same sweep. Gaps of that shape are what the lens looks for.
Evidence
$ gh issue list -R Comfy-Org/comfy-cli --state closed --label bug --limit 100 --json number -q 'length'
91
$ gh issue list -R Comfy-Org/comfy-cli --state open --label quality:missing-regression-test --json number -q 'length'
0
Fix
Run the lens: skills/closed-bug-regression-audit/SKILL.md. It uses a 3-pass parallel-subagent strategy (keyword, adversarial, fix-PR-diff, systemic-pattern) and files gap issues under quality:missing-regression-test.
Per the sweep's shared schema, apply the sweep labels alongside the native one so the results stay queryable with the rest of the sweep backlog.
Dedupe first — this repo already has sweep issues filed:
bash skills/repo-improvement-sweep/scripts/existing-findings.sh Comfy-Org/comfy-cli
bash skills/repo-improvement-sweep/scripts/in-flight-work.sh Comfy-Org/comfy-cli
Note the repo currently has 13 open PRs, several long-lived — check against those before filing anything.
Deferred from repo improvement sweep 2026-08-17. Parent: #723
What —
closed-bug-regression-auditwas deferred, not skipped, during the 2026-08 repo improvement sweep. It has never run against this repo's 91 closed bug issues.Why it matters here — 91 closed bugs is the largest regression-audit surface of the six repos swept, and the only one where the lens produces signal at all; the other five have zero closed bug issues. It was deferred because a full 3-pass parallel-subagent run does not belong in a corner of a config-and-docs sweep, not because the surface is uninteresting.
The repo has the shape that makes this productive: 695 commits, 45 contributors, 30 published releases, 151 test files against 283 source files, and a
pytest.ymlthat runs on every PR. A bug that was fixed without a regression test can silently come back through any of those 45 contributors.There is a specific reason to think the yield is non-zero.
comfy_cli/utils.pyshipped a knowingly-unfilteredtar.extractall()for the entire period cpython#107845 was open, with no test covering extraction safety either way — the regression test that proves it now only exists because of #734, filed by this same sweep. Gaps of that shape are what the lens looks for.Evidence
Fix
Run the lens:
skills/closed-bug-regression-audit/SKILL.md. It uses a 3-pass parallel-subagent strategy (keyword, adversarial, fix-PR-diff, systemic-pattern) and files gap issues underquality:missing-regression-test.Per the sweep's shared schema, apply the sweep labels alongside the native one so the results stay queryable with the rest of the sweep backlog.
Dedupe first — this repo already has sweep issues filed:
Note the repo currently has 13 open PRs, several long-lived — check against those before filing anything.
Deferred from repo improvement sweep 2026-08-17. Parent: #723