[ricorna]

Summary

• Pass the child process exit code into generic test summarization and treat it as authoritative.
• Bypass the shared never-worse raw fallback on nonzero generic-test exits so short success-looking or empty raw output cannot erase the failure banner.
• Recover failure diagnostics from the complete raw output rather than the displayed tail.
• Keep bounded head/tail context plus anchor-centered context around detected errors and recognized specialized failures, including diagnostics displaced by cleanup tails.
• Ignore zero-valued counters such as 0 failed when locating real diagnostics.
• Keep standalone FAIL: recognition scoped to rtk test so rtk err behavior is unchanged.
• Add subprocess and unit regressions for early failures, heuristic runner-name collisions, contradictory zero-exit text, success-looking nonzero summaries, zero-failure counters, recognized failures with nearby causal diagnostics, and unknown failure formats.

Safety contract

• Exit code 0 never receives a failure banner solely because output contains words such as FAILED.
• A nonzero exit always receives a failure presentation and retains bounded actionable context.
• Full raw output remains available to RTK tee handling.
• Never-worse output capping remains active.

Verification

• cargo fmt --all -- --check
• cargo clippy --locked --all-targets -- -D warnings
• cargo test --locked
• cargo test --locked --test generic_test_runner_test

The complete locked suite passed locally: 2,515 unit tests passed, 8 ignored, plus all integration targets.

Fixes #2420

This overlaps earlier work in #2427, #2683, and #2996, but adds full-output diagnostic recovery, authoritative exit-status handling, and regression coverage for success-looking nonzero output and zero-exit scary words.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[KuSh]

Closing in favour of #2683, which fixes the same generic runner fallback in extract_test_summary for the same issue and keeps the shared error pattern list untouched. Review and comments are welcome there. If part of the scope here is not covered by #2683, say so on that PR or reply here.