Severity guide

Severity classifies evidence. Policy chooses the action.

Reports lead with CRITICAL through INFORMATIONAL so a reviewer can triage consistently. Deterministic analyzers create findings; coverage limits what can be concluded; versioned policy produces the secondary operational action.

Inspect the rules

Primary classification

Five severity levels with precise meaning

CRITICAL

Corroborated evidence with urgent potential impact. Treat as an immediate security decision; a block still requires the scanner policy and evidence class to support it.

Read the exact evidence, affected locations, and coverage before acting.
HIGH

Prioritize for human review. High severity is not, on its own, a public vulnerability claim.

Read the exact evidence, affected locations, and coverage before acting.
MEDIUM

Meaningful behavior or exposure that needs purpose and implementation context.

Read the exact evidence, affected locations, and coverage before acting.
LOW

Limited impact or weakly correlated behavior retained for review context.

Read the exact evidence, affected locations, and coverage before acting.
INFORMATIONAL

Observed context that does not independently drive an operational action.

Read the exact evidence, affected locations, and coverage before acting.
Artifact+Evidence+Coverage+Policy=Decision

Secondary operational action

Four decisions with operational meaning

BLOCK

Authoritative malicious intelligence matched, or the selected policy explicitly rejects high-confidence abuse evidence.

Do not install; remove and investigate existing exposure.
REVIEW

Sensitive capability, correlated behavior, vulnerable dependencies, or provenance concerns need human context.

Inspect the cited source, intended behavior, and remediation before approval.
INCOMPLETE

Required providers, declared entrypoints, archives, or artifacts could not be analyzed sufficiently.

Restore coverage or independently inspect unsupported content. Never interpret this as clean.
ALLOW

Required analysis completed and no evidence crossed the active block or review policy.

Approval applies to this exact artifact hash and policy version, not every future release.

Diagnostic index

Risk score · 0-100

Prioritizes sensitive capability, supply-chain exposure, provenance, weak indicators, and correlated abuse potential. It is for review ordering, not a probability.

Diagnostic index

Malware score · 0-100

Summarizes confirmed and strongly correlated malicious evidence. It is not statistically calibrated and should always be read with evidence class.

Compatibility field

Grade · A-F

A legacy human-readable rollup retained in report bundles. It is secondary to decision, completion state, and exact evidence.

GUARDRAILS

What the scanner refuses to claim

No findings ≠ safe

Unsupported files, missing providers, parse failures, and uninspected entrypoints must remain visible in coverage.

Capability ≠ malware

Shell, network, filesystem, and agent APIs are common in legitimate developer tools. Context and correlation matter.

Reputation ≠ code truth

Install count, age, publisher status, and repository health inform provenance but cannot prove artifact behavior.

AI ≠ detection authority

AI may summarize evidence for a reviewer, but it must not invent findings or silently change the deterministic decision.