Methodology

Analysis Pipeline

• 1

  Artifact Intake & Hash Binding

  The submitted skill package is received and an SHA-256 digest is computed immediately on arrival. All subsequent findings are cryptographically bound to this digest. Any modification to the package — even a single byte — produces a different hash and invalidates prior scan records for that artifact.

• 2

  Static Analysis

  The source code and dependency manifest are analysed without execution. Checks include: declared permission and capability scope, known-vulnerable dependency versions, obfuscated or encoded code blocks, hardcoded credentials patterns, and unexpected network or filesystem declarations. Results are summarised into structured findings.

• 3

  Sandbox Execution

  The skill is executed in an isolated, network-monitored environment. Observed behaviors — outbound connections, filesystem writes, process spawning, and environment variable access — are recorded. The sandbox session is time-limited and resource-constrained. Behaviors inconsistent with the skill's declared purpose are flagged as indicators.

• 4

  Risk Scoring

  Findings from both phases are combined into an integer risk score (0–100). Lower scores indicate fewer observed indicators. Thresholds:

  • 0–30 — Verified: No significant indicators detected. Verification is specific to the scanned build hash.
  • 31–70 — Under Review: Some indicators present; further review may be warranted.
  • 71–100 — Risk Indicators: Multiple automated indicators detected. Detailed findings available in paid report.
• 5

  Ledger Entry & Receipt

  A permanent receipt is written to the public ledger, containing the artifact SHA-256, scan timestamp (UTC), status, risk score, and redacted finding summary. Full unredacted findings are available only in paid reports. Ledger entries cannot be deleted or modified after creation.