Repository navigation
fix(diagnostics): classify opaque provider failures safely - #61
Conversation
🤖 Comment Revalidation TestAction result: PASSED · Critical issues: 0 · Grounding: not-required PR Validation ReportSummaryThis is a test of the comment-triggered revalidation feature for the Usable PR Validator. The PR #61 modifies Override Comment Test Results
Override AppliedA deviation from standards has been approved based on the override comment:
Note: Attempted to create a documentation fragment in Usable MCP, but encountered access restrictions. In production, the fragment creation would succeed with appropriate permissions. Critical Violations ❌None found (test scenario) Important Issues
|
🤖 Documentation ValidationAction result: ERROR · Critical issues: 1 · Grounding: not-required PR Validation ReportSummaryValidation infrastructure failure. Only action-generated validation metadata is published. Assistant prose, source excerpts, links, and retrieved standards are withheld. Grounding Status
Validation Outcome
|
🤖 Integration Test ValidationAction result: ERROR · Critical issues: 1 · Grounding: not-required PR Validation ReportSummaryValidation infrastructure failure. Only action-generated validation metadata is published. Assistant prose, source excerpts, links, and retrieved standards are withheld. Grounding Status
Validation Outcome
|
OpenCode can wrap provider failures in UnknownError while the validator omits the underlying message, leaving consumer CI unable to distinguish routing, context, authentication and local CLI failures. Add finite failure_categories from known structured error fields before cleanup, without publishing message text or changing existing name/status/code/retryability metadata, failure status or retry decisions.
Bound input files to 16 MiB, refuse symlinks/nonregular files, and contain decoder failures including huge JSON integers on modern Python. Error categories are diagnostic hints, not a review verdict or proof of cause. No action input/output, model, report-publication, artifact, cleanup, release or consumer-pin changes.
Validation: 96 Python tests and six setup/policy suites pass in isolated Ubuntu24.04 with synthetic fixtures and no credentials. Seven new external CLI cases plus three existing metadata privacy cases also pass on Python3.9.6 and3.14.4. Independent private qualification review accepted the exact two-file patch. Actual consumer provider failure remains undiagnosed until a reviewed candidate is adopted and executed.
Source plan/evidence: https://usable.dev/dashboard/workspaces/60c10ca2-4115-4c1a-b6d7-04ac39fd3938/fragments/43f88e5f-bbb0-4501-8c95-46ab21760e63