The classifier that hadn't read the script

说明与用途

分类器错误地将破坏性任务标记为只读，导致提前执行。更新后，分类器解析脚本主体，检测破坏性命令，并正确阻止提前运行。

问题

A task acceptance gate misclassified a destructive job as read‑only because it only inspected the action line, not the script body, leading to unsafe early execution.

结果

The updated classifier accurately detects destructive actions by reading the script body, preventing unsafe early task execution and ensuring proper scheduling.

工作流

• Inspect the task’s action line for a script path and arguments.
• Parse the script body to identify destructive verbs (e.g., write_text, rename, Remove-Item).
• If destructive verbs are found, label the task as destructive; otherwise label as read‑only or desired‑effect.
• Reject early execution for destructive tasks; schedule them for their first real window.
• Log the classification verdict and any corrections for audit.
• Run smoke tests only on safe (read‑only or desired‑effect) tasks.
• Update the classifier’s verb list and regex to match paths anywhere in the line and include missing verbs.
• Re‑classify affected tasks and verify correct labeling.
• Document the changes and provide receipts for each task.

证据

The classifier ran, printed its verdicts, and was wrong in the quietest possible way.

来源

The rework matches the path anywhere in the line and adds the missing verbs; the rollup and its sibling verifier immediately reclassified as desired-effect, which is what they always were — and the classifier now fails visibly when it can't find the script, instead of grading a file it never opened.

来源