What changed after I stopped trusting exit code 0: ELY Agent Input Preflight v0.7.0

Описание и применение

Новый слой Admission проверяет генерируемый код на синтаксис, структуру, область применения и безопасность, прежде чем он будет принят в репозиторий, уменьшая зависимость от кода выхода 0.

Задача

AI coding agents often report success via exit code 0, but this does not guarantee that the generated code is correct, safe, or meets project requirements.

Результат

The new admission layer ensures that only code that passes explicit safety and policy checks is promoted to a verified final result, reducing reliance on agent exit codes and improving code quality assurance.

Рабочий процесс

• 1. AI agent generates code and returns exit code 0.
• 2. ELY Agent Input Preflight receives the code and runs the Generated-Code Admission layer.
• 3. Admission layer performs deterministic checks (syntax, parsing, scope, policy, dangerous constructs).
• 4. If all checks pass, code is admitted and can be used in the final result; otherwise, status is NOT_PROVEN and escalated to NEEDS_REVIEW.
• 5. Final verification decides whether the repository satisfies acceptance criteria.

Доказательства

The biggest change is a new layer called Generated-Code Admission.

Источник

For Python source, the Community release currently performs bounded, deterministic checks including: syntax validation; structural parsing; source-scope checks; project-local policy checks; baseline dangerous-construct detection.

Источник

If ELY cannot prove a result, it does not silently convert uncertainty into safety.

Источник

The Community version keeps the full core verification and safety functionality.

Источник

Before publishing the release, I ran the full regression suite plus adversarial admission cases.

Источник