<!-- SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. --> <!-- SPDX-License-Identifier: Apache-2.0 -->

Audit Every E2E Assertion Before Rerunning

Use the existing suite as the specification for a complete source review. Find the visible failure and later source-visible failures before the next expensive run. Aim to batch the necessary repairs; source inspection cannot guarantee runtime success.

The deliverable is an itemized assertion ledger with concrete inputs, producing code, and evidence. Build the ledger yourself. Do not ask the user to enumerate assertions or supervise its completion.

Follow the shared writing and review contract for the ledger and reports.

When to Use This Skill

Use this skill whenever investigating a failing or flaky E2E test, including timeouts, setup failures, assertion failures, and cleanup failures. Apply it when E2E failures emerge during broader PR or CI work, even when the user does not name this skill. Begin the review before proposing a repair or rerun.

A log classifier can locate evidence, but its summary does not complete the assertion audit. The E2E execution skill continues to own dispatch and environment authorization. Status-only queries and requests only to start a run do not require an assertion audit.

Keep the Requested Boundary

Use the assigned branch and checkout. Preserve the user's restrictions on edits, tests, execution environments, command visibility, and publication. Reuse established authorization and context. An audit request alone does not authorize code changes, pushes, live execution, or merges.

Review the requested failing suite and the prerequisites and downstream jobs needed for its stated completion. Do not turn a scoped repair into repository-wide maintenance or add unrelated checks. For audit-only work, report proposed corrections without applying them.

Establish What Actually Failed