Our AI-agent scanner got it backwards. Here’s what we fixed in v1.1

Описание и применение

В версии 1.1 сканер Agent PreShip Guard исправил ложные срабатывания и добавил правило AG008, которое обнаруживает выбор модели опасных функций без подтверждения.

Задача

The Agent PreShip Guard GitHub Action incorrectly flagged a safe read‑only uptime check (false positive) and missed a destructive model‑chosen delete operation (false negative).

Результат

After updating to v1.1, the action correctly flags only the unsafe model‑chosen destructive call and no longer mislabels a safe shell command.

Рабочий процесс

• 1. Create a GitHub repository with the Agent PreShip Guard action.
• 2. Add a workflow file (.github/workflows/agent-guard.yml) that checks on pull_request and push events.
• 3. Configure the action to scan the repository path and fail on high severity findings.
• 4. Run the workflow; inline PR annotations show findings.
• 5. Update the action to v1.1 to correct false positives and add new rule AG008.

Доказательства

== Agent A ::error AG001 Model/LLM output passed to shell ::warning AG004 shell=True in tool code Agent PreShip Guard: 1 high, 1 medium findings. == Agent B Agent PreShip Guard: 0 high, 0 medium findings.

Источник

== Agent A ::warning AG004 shell=True in tool code Agent PreShip Guard: 0 high, 1 medium findings. == Agent B ::error AG008 Model output selects a destructive call with no approval gate Agent PreShip Guard: 1 high, 0 medium findings.

Источник