cd /news/developer-tools/fix-blame-test-failures-on-windows-c… · home › topics › developer-tools › article
[ARTICLE · art-172665] src=keepplain.com ↗ pub= topic=developer-tools verified=true sentiment=· neutral

Fix blame test failures on Windows CI short temp paths

A debugging session on Claude Code (Opus 5.5, medium effort) traced Windows CI blame-test failures to 8.3 short temp path names: setting TEMP to the short name reproduced the failure locally, with 6 blame tests failing before the fix and 7/7 passing after, and the full suite passing 377/377 with the normal TEMP. The agent changed locate() in scripts/lib/blame.mjs to use relativeTo from scripts/lib/edits.mjs, because the runner's long and short temp names made path.relative return ../ paths and blame reported the file as outside the repository. The fix remains uncommitted in scripts/lib/blame.mjs, and an earlier change setting core.autocrlf=false in makeRepo in test/helpers.mjs was committed as 73bc408, taking the git tests to 12/12 with no LF/CRLF warnings.

recommendation

When blame tests pass locally but fail on Windows CI, set TEMP to the runner's 8.3 short name to reproduce, then resolve paths with relativeTo instead of path.relative.

more in playbook: Working with AI coding agents
by read3 min views16 publishedNov 7, 2026
Fix blame test failures on Windows CI short temp paths
Image: Keepplain (auto-discovered)
[Claude Code](https://keepplain.com/agent/claude-code)· Opus 5.5 · medium effort

[debug](https://keepplain.com/task/debug)CLI & tooling3.0M tokens

[keepplain/keepplain-plugin ↗](https://github.com/keepplain/keepplain-plugin)edited

Click a segment to jump to that moment

41msession

1interventions

GOAL

Find out why the blame tests pass locally on Windows but fail in CI on Windows, and fix the cause. The session also started with removing the LF to CRLF warnings from the git test output.

Moments proposed by a model from the session, reviewed and published by the author.

  1. +0Prompt · +0ok 107 - the shared fixtures are the site's, as synced # warning: in the working copy of 'app.txt', LF will be replaced by CRLF the next time Git touches it # warning: in the working copy of 'docs/store/listing-000.md', LF will be replaced by CRLF the next time Git touches it
  2. +0mAgent did · +0mFound the test in test/git.test.mjs that writes 105 listing files and runs git add with stderr not hidden. Set core.autocrlf=false in makeRepo in test/helpers.mjs. The git tests pass 12/12 with no warnings; committed as 73bc408.Code · 1 file · +2 −0 · not shown
  3. +6mPrompt · +6mnot ok 18 - a trailer whose session the site does not hold is found on this computer; signed out, nothing is asked --- error: Cannot read properties of undefined (reading 'b0c72d7750acca2da468b70e7070ba057bdb5a7e')
  4. +6mAgent did · +6mRan the blame tests locally, where they passed 7/7. Read the CI log and found the 'not in the repository' error on a runner whose temp folder uses an 8.3 short name. Reproduced it by setting TEMP to the short name: 6 blame tests failed before the fix and 7/7 passed after. Changed locate() in scripts/lib/blame.mjs to use relativeTo from scripts/lib/edits.mjs.Code · 1 file · +5 −3 · not shown
  5. +14mFail · +14mWith TEMP on the short name, the full test suite failed two tests outside blame (117 and 167).
  6. +14mAgent did · +14mRan the full suite again with the normal TEMP and got 377/377. Then started a control run without the fix in the background, which stashed blame.mjs while it ran.
  7. +30mIntervention · +30mEdited scripts/lib/blame.mjs by hand (+5 −3) while the agent's control run was still going, so the fix did not have to wait for it.Why: I did not want to wait for the stashed baseline run to finish before the fix was back in the file.Code · 1 file · +5 −3 · not shown
  8. +30mAgent did · +30mRead the control output and checked the git state. Root cause: the runner's long and short temp names made path.relative return ../ paths, so blame reported the file as outside the repository. Left the blame.mjs change uncommitted for the developer to decide.
  9. +41mOutcome · +41mFull suite 377/377 with the normal TEMP; blame tests 7/7 with the short TEMP path (6 failed before the fix). The fix is still uncommitted in scripts/lib/blame.mjs; the temp test folder the agent created was removed.

VERDICT

Next time I would say up front that the blame tests pass on my machine but fail only on the Windows CI runners, and paste the CI log with the 'not in the repository' error. The step that did the work was reproducing it locally by pointing TEMP at the 8.3 short name, which made six blame tests fail on my machine too.

I tried this · 0 #

Same starting prompt, different run

Similar sessions #

- [Diagnose git rebase --continue freeze in agent shellPi · Git, Bash · 2m · 0 int.](https://keepplain.com/b/diagnose-git-rebase-continue-freeze-in-agent-shell)
- [Diagnose chub CLI ESM launch failure with nvm/NodeCursor · Node.js, nvm, npm · 13s · 0 int.](https://keepplain.com/b/diagnose-chub-cli-esm-launch-failure-with-nvmnode)
── more in #developer-tools 4 stories · sorted by recency
── more on @claude code 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/fix-blame-test-failu…] indexed:0 read:3min 2026-11-07 · —