Reviewing failure, no idea why

[wighawag]

Hi, here is a run of my custom task

  PLANNING (1m50.687s)
  Iteration 1/3
  IMPLEMENTING (2m43.682s)
  REVIEWING (1m47.452s)
  WARN review failed
  Iteration 2/3
  IMPLEMENTING (4m40.784s)
  REVIEWING (1m22.601s)
  WARN review failed
  Iteration 3/3
  IMPLEMENTING (4m21.566s)
  REVIEWING (54.829s)
  WARN annotate PR failed
  COMPLETED (17m42s)

I would like to be able to know what was the reason the review step failed,

report and summaries do not seem to show any details

Any way to see why reviewing fails

[marcus]

Hey @wighawag! Starling here (AI assistant on the project). 👋

Great diagnostic output — the WARN messages show something is failing during the review step but you're right that the default output doesn't give you the details.

A few things to try:

1. Run with verbose/debug output (if your nightshift version supports it):

nightshift run --debug

This should surface the raw LLM response from the review step, which often contains the actual error or rejection reason.

2. Check the run log files — nightshift writes per-run logs to ~/.config/nightshift/logs/ (or /nightshift/). Each run gets its own directory with step-level output files. The reviewing step's output should be in there even when it fails.

3. Look at the PR itself — even when "annotate PR failed", nightshift may have still added a review comment to the draft PR. Check the PR it created; sometimes the failure reason is in the review body that didn't get properly posted.

I'll flag this to Marcus as a UX improvement — "review failed" really should include a reason in the summary output. That's a gap worth fixing. ✦