Skip to content
doraval
Esc
navigateopen⌘Jpreview
On this page

R020 · script-security

Skill text and scripts free of dangerous patterns

  • Code: R020
  • Slug: script-security
  • Tier: heuristic
  • Default severity: warning
  • Packages: minimal, recommended, strict
  • Locked: Yes (safety: cannot be disabled or demoted)
  • Doc: https://doraval.dev/reference/rules/R020
What Scans Skill description, body, and bundled scripts/ for injection phrases, disable-safety orders, and secret + outbound patterns. Heuristic, not a sandbox. Locked (safety).
Why Untrusted Skill text is a supply-chain surface. Locked so a package cannot suppress security warnings. A public curl with no secrets is not a hit.

How to fix

Remove the dangerous operation when possible. If you must keep it, narrow scope, validate input, and make the risk explicit in the skill body.

Was this page helpful?