how it works
Four steps, nothing executed.
Paste or upload the script
A PowerShell script from ChatGPT, a colleague, a vendor or your own library.
Say what it should do
For example: restart three servers, or stop the test VMs.
Run Preflight
The official parser reads it. Nothing is executed. Rule packs and the permission map are applied.
Read the evidence
Blast radius, the expected-object-count check, and every finding with its rule ID and line span.
- Rules before AI
- Deterministic findings come first. AI explains but can never lower a severity.
- Honest coverage
- When the engine can't identify a command, the report lists it and won't call the script clean.
- Every rewrite is re-checked
- Proposed changes are parsed and scanned again before you see them.
- Evidence you can attach
- Rule IDs, line spans and engine versions in every report.
PowerShell today. SQL, Bash and Python are next.
how it compares
Not a linter. Not a chatbot.
| Capability | Linters (PSScriptAnalyzer, ShellCheck) | AI chat | RMM approval queues | ScriptPreflight |
|---|---|---|---|---|
| What it changes and how far | No | Partly | Partly | Yes |
| Permissions needed vs granted | No | Partly | No | Microsoft Graph (preview, 25 common commands); AD, Exchange and Azure rolling out |
| Can it be undone | No | Partly | No | Yes |
| Does it match the intent | No | Partly | No | Expected object count only; the AI intent check is coming |
| Evidence an approver can sign | Partly | No | No | Yes |
| Never executes the script | Yes | Yes | No | Yes |
Use them together: linters for style, ScriptPreflight for impact.
Run your own script.
No card required. Your script is never executed.