E2E failure issue creation rejects unsupported gh --json flag #5249

During run 37340452976, IssueBodyUpdater._create() called gh issue create --json number and received unknown flag: --json. The "Report failure" step crashed before creating an occurrence issue. The reporter should create the issue through a supported interface and preserve occurrence records in its body.

Observed while validating #5232 on commit 509a520 in the user image built by the existing E2E workflow.

The traceback points to src/autoskillit/_probe_canary.py:142. _create() passes --json number to gh issue create, then assumes the returned stdout is JSON. The pinned user-image GitHub CLI rejects that option before creating an issue.

To reproduce, trigger an E2E failure for a test without an existing occurrence issue, such as the first failed headless-smoke run. The Report failure step raises:

RuntimeError: gh issue create failed: unknown flag: --json

Keep the existing updater's integer issue-number return contract, body-file writes, and occurrence-in-body behavior when repairing issue creation.

Additional occurrence

The same unsupported gh issue create --json number failure recurred in the Report failure step of run 37346469142 on commit 4146eeb. That run's separate L3 kitchen bootstrap failure is tracked in #5250.

Additional occurrence: implementation commissioning run 37854525992

The unsupported gh issue create --json failure recurred in the Report failure step of run 37854525992, testing commit 017b3f1 for PR #5255 / #5233.

The implementation scenario first failed at its planner dispatch because the bundled recipe supplies output_dir: . to the bounded make-plan skill. The reporter then raised:

RuntimeError: gh issue create failed: unknown flag: --json

This is the same reporter defect, independently of the planning failure. Fix the existing issue-creation path while preserving integer issue-number returns and occurrence records in the issue body.

Additional reproduced occurrence — 2026-10-15

Both implementation attempts 38746791327 and 38746802415, at commit96a18e72922040f4951b343195aff1188503d7d5, failed their Report failure steps with the same unsupported issue-create JSON flag. The primary scenario failures occurred earlier in make-plan and are tracked with source/probe evidence in #5283. This occurrence does not establish the reporter as the cause of those planning failures, and no duplicate reporter ticket was filed.

Additional occurrences — 2026-10-16 commissioning

The same secondary failure reporter error occurred in both implementation attempts at PR5255 head 20ee39237e8e89a8d21f403c3a825273fcac6bf7: 38960061259 and 38960069882. Their Report failure logs record e2e_failure_issue_failed and gh issue create failed: unknown flag: --json (20:34:40Z and20:42:25Z respectively). The unsupported option prevents automatic failure-issue filing. Retained logs/artifacts still contain the original failures.

This is a reporting defect after the E2E failure, not the cause of either pipeline failure; do not use its presence to diagnose or erase the separate recipe/write-evidence findings. Reproduction remains the existing failure reporter with its actual gh CLI invocation. No duplicate issue filed.

Further occurrences — 2026-10-17 commissioning

The same secondary reporter error recurred at PR5255 head28448805b in 39082343853 and 39082347591. Report failure logs contain e2e_failure_issue_failed / gh issue create failed: unknown flag: --json (00:45:33Z and01:00:05Z on18Oct respectively). The retained redacted artifacts preserve the underlying failures.

Run1 failed canonical seed-URL validation and strict extra-branch cleanup; run2 failed canonical seed-URL validation. The reporter error happened afterward and did not cause either result. One actual prep-output write-evidence mismatch was manually filed as #5289 after two adversarial reviews and parent evidence inspection. No duplicate reporter issue was filed.

Recommissioning evidence — 2026-10-18 (e529c48)

Tested feature head e529c48a8d6576ca7bbff29a51c0d7ee12caa167, integrated develop a93c20037bddef0c534af3f7da44ca4310733b33 including #5290/#5289.

The secondary failure reporter reproduced in both completed attempts: 39177297225 and 39177304001. Each Report failure log raises RuntimeError: gh issue create failed: unknown flag: --json (retained failed.log81 in each run directory). Both primary harness results had already failed audit admission; the reporter error did not cause the model/audit failures. No duplicate reporter ticket was created.