hyp smoke client_attach_idempotent is in the release battery (CLAUDE.md §Release checklist) and currently fails on origin/master.
Evidence
Observed repeatedly while reviewing PRs #650 and #651, each time verified against a clean tree rather than assumed:
- fails at
origin/master 7116f95
- fails at every PR branch tested off it, identically, so no open PR causes it
- passes at
1fbfa0d (v1.19.0)
Failing assertion: attach without ai-gateway prints a clear error, with the observed message the claude adapter is not enabled on this install.
Why it matters
This is a release gate, so it blocks a clean battery run. It is also a genuine behaviour change rather than a flaky test: the smoke expects one error for "the gateway is not present" and is now getting a different one about the claude adapter, which suggests the precedence or the wording of the attach precondition checks moved.
Bisect range
1fbfa0d (passing) to 7116f95 (failing). That window includes the v1.20.0 publish and the onboarding-wizard merge (#629), among others, so a bisect over the smoke should land it quickly.
Suggested first look
Compare the attach precondition ordering in the claude adapter and hyp attach between the two revisions: whichever change made the adapter-enabled check fire before the gateway-present check would produce exactly this symptom.
hyp smoke client_attach_idempotentis in the release battery (CLAUDE.md§Release checklist) and currently fails onorigin/master.Evidence
Observed repeatedly while reviewing PRs #650 and #651, each time verified against a clean tree rather than assumed:
origin/master7116f951fbfa0d(v1.19.0)Failing assertion:
attach without ai-gateway prints a clear error, with the observed messagethe claude adapter is not enabled on this install.Why it matters
This is a release gate, so it blocks a clean battery run. It is also a genuine behaviour change rather than a flaky test: the smoke expects one error for "the gateway is not present" and is now getting a different one about the claude adapter, which suggests the precedence or the wording of the attach precondition checks moved.
Bisect range
1fbfa0d(passing) to7116f95(failing). That window includes the v1.20.0 publish and the onboarding-wizard merge (#629), among others, so a bisect over the smoke should land it quickly.Suggested first look
Compare the attach precondition ordering in the claude adapter and
hyp attachbetween the two revisions: whichever change made the adapter-enabled check fire before the gateway-present check would produce exactly this symptom.