Finding severity

How Skillstore describes confirmed finding severity

A precise guide to finding severity and why findings, capability context, audit status, and install advice remain separate.

Published Jul 14, 2026 Updated Jul 15, 2026 Methodology 2026-07-14 Snapshot Jul 14, 2026

Severity belongs to findings

Finding severity Confirmed evidence Typical reviewer response
Low Confirmed, limited behavior with low impact or narrow exposure. Read the evidence and keep normal safeguards.
Medium A meaningful concern or permission boundary needs review. Review before use in a sensitive environment.
High Evidence supports a serious concern such as dangerous execution, exposure, or overreach. Require explicit confirmation and remediation planning.
Critical Evidence supports severe impact or an immediate compromise path. Automated policy should block or require the strongest exception process.

There is no Safe finding severity. When an audit records no confirmed findings, the report states that observation directly. It does not decide that the Skill is safe. A Skill may still read files, call a network service, or run a tool when that capability is declared, necessary, and bounded.

Finding severity is not capability

Capability labels describe what a Skill can do. Risk findings describe why the observed implementation may be unsafe. Treating every shell command or network request as a vulnerability creates noise; ignoring an undeclared or unbounded capability hides real exposure.

Skillstore therefore keeps capability-review items beside confirmed findings instead of turning capabilities into an overall Skill judgment.

Finding severity is not audit status

An audit can be pending, failed, completed, or missing. Those lifecycle states say whether evidence is available. They do not describe the severity of a finding or conclude that a Skill is safe.

A report is not an installation gate

Skillstore publishes install advisories for consumers:

  • allowed means the available evidence does not require additional confirmation from the shared policy.
  • confirmation_required means an Agent or operator should stop and obtain an explicit decision.
  • blocked is a policy recommendation for automated consumers, not a platform-wide 403 for a public page or manual download.
  • allowed_with_warning keeps manual installation available while placing the relevant warning before the action.

Organizations can adopt stricter rules. The public report remains the evidence source; the organization owns its execution policy.

Findings can change

Every finding applies to an identified version and evidence location. A new version can add, remove, or change behavior. Compare the old and new evidence instead of treating the absence of a finding in an earlier report as a permanent seal.