skills/vllnt/skills/improve-dx/Gen Agent Trust Hub

improve-dx

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external repository data, such as manifests, scripts, and CI configurations, to identify optimization opportunities. This creates a surface for indirect prompt injection if the processed data contains malicious instructions. The skill mitigates this by requiring a 'Quality Validation' protocol, including self-assessment and independent review of all changes. * Ingestion points: Repository manifests (package.json), lockfiles, CI configuration files, and script contents. * Boundary markers: The skill uses a structured 'Definition of Done' and 'Quality Validation' protocol to bound the agent's actions. * Capability inventory: The skill permits modifying files, changing CI configurations, and installing packages via the existing package manager. * Sanitization: The protocol emphasizes verifying results against baselines and requires justification for any code additions.
  • [EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to install various @vllnt packages from the NPM registry as defaults for new projects or alignment tasks. These are vendor-owned resources managed through standard package managers.
  • [COMMAND_EXECUTION]: The skill performs native build, lint, and CI commands to establish baselines and verify optimization gains. These operations are restricted to the authorized workspace scope and focus on improving DX without weakening existing checks.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 08:18 PM