Reads
gstack’s /qa-only test stops failing Opus 5.5 for reading a skill file
wc -l qa-only/SKILL.md; ls qa-only …same read, inside the fixturegstack's /qa-only test stops failing Claude Opus 5.5 for looking at a skill file. A line count and a directory listing inside the test folder now pass.
Errors
A read may now discard or merge its errors
ls; ls qa/templates; cat TODOS.md 2>/dev/null | head -20… 2>&12> file · > · >> · | tee · /etcSending a read's errors to dev null, or merging them into its output, now passes in the /qa-only and docs sync tests. Every other redirect, and any read that leaves the test folder, still fails.
Browser stop
Stopping the browser now counts as cleanup
closetab"$B" stopstop before a probe · stop --all · outside the guardStopping the browser now counts as cleanup, the same as closing a tab, when the agent does it inside the guard after its last probe. A stop before a probe, or outside the guard, still fails.
Docs sync
The docs-sync test forgives one wrong path
<no such path>/gstack-docs-candidatesame command, declared pathwrong path that exists · any other exit codeThe docs sync test forgives one wrong path. An agent that calls the docs helper at a missing path, then reruns it at the declared path, now passes. A wrong path that does exist still fails.
Unchanged
Thresholds, trials and behavior checks stay the same
CHANGELOG.mdgstack release notes, every case that still failsPass thresholds, trial counts and behavior checks stay the same, and the changelog lists every case that still fails. Which model runs your /qa-only checks, Opus 5.5 or Fable 5.1?
