Failure
Claude reviews now fail before producing any review. This differs from the placeholder/publication defect fixed by #5184 and #5266. The corrected publisher correctly refuses to publish a verdict.
Examples: Snyk #5312 diagnostic run, Google #5315, and Azure #5313.
The action reports --json-schema was provided but Claude did not return structured_output. Result subtype: success. Its execution result has is_error=true, one turn, zero cost, and empty modelUsage, completing in under one second. No actual review is produced. A temporary trusted-branch diagnostic read the action's documented execution output and emitted only a fixed error category, without printing model text or credentials. It reported unclassified-error; authentication, quota, billing, model availability, and service health therefore remain unconfirmed.
Required recovery
Determine and repair the Claude service/account/CLI failure, then obtain genuine reviews on current PR heads. Preserve the structured-output contract, read-only tools, trusted publisher, and existing merge gates. Do not enable full logs, fabricate a verdict, or repeatedly rerun unchanged failures.
After recovery, regenerate these closed generated snapshots from latest main: mvn #5272, flux #5279, pulumi #5281, eksctl #5282, terraform #5283, cargo #5300, hadolint #5305, syft #5303, and az #5313. Their queued/running CI was cancelled and verified empty. Snyk additionally needs #5312 merged before regeneration. AWS generation is still running independently.
The service failure also blocks review of #5312, #5314, and #5315. Related acceptance issues: #5244, #5137, #5008, #4692, #4712, #5009, #4474, #4331, and #5180.
Diagnostic source: commit 4af90f1 on the temporary issue-5183-review-diagnostics branch; both focused Node tests and actionlint passed. This diagnostic does not change publication or merge gates.
Failure
Claude reviews now fail before producing any review. This differs from the placeholder/publication defect fixed by #5184 and #5266. The corrected publisher correctly refuses to publish a verdict.
Examples: Snyk #5312 diagnostic run, Google #5315, and Azure #5313.
The action reports
--json-schema was provided but Claude did not return structured_output. Result subtype: success. Its execution result hasis_error=true, one turn, zero cost, and emptymodelUsage, completing in under one second. No actual review is produced. A temporary trusted-branch diagnostic read the action's documented execution output and emitted only a fixed error category, without printing model text or credentials. It reportedunclassified-error; authentication, quota, billing, model availability, and service health therefore remain unconfirmed.Required recovery
Determine and repair the Claude service/account/CLI failure, then obtain genuine reviews on current PR heads. Preserve the structured-output contract, read-only tools, trusted publisher, and existing merge gates. Do not enable full logs, fabricate a verdict, or repeatedly rerun unchanged failures.
After recovery, regenerate these closed generated snapshots from latest main: mvn #5272, flux #5279, pulumi #5281, eksctl #5282, terraform #5283, cargo #5300, hadolint #5305, syft #5303, and az #5313. Their queued/running CI was cancelled and verified empty. Snyk additionally needs #5312 merged before regeneration. AWS generation is still running independently.
The service failure also blocks review of #5312, #5314, and #5315. Related acceptance issues: #5244, #5137, #5008, #4692, #4712, #5009, #4474, #4331, and #5180.
Diagnostic source: commit 4af90f1 on the temporary
issue-5183-review-diagnosticsbranch; both focused Node tests and actionlint passed. This diagnostic does not change publication or merge gates.