manage-defects
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill maintains an attack surface for indirect prompt injection because it is designed to ingest and process untrusted external data, such as project source code, CI/CD logs, and browser automation artifacts.
- Ingestion points: Untrusted data enters the agent context through the reading of source files (
references/vstack/capabilities/code-architecture/REFERENCE.md), check configurations and CI workflows (references/vstack/capabilities/code-tests/REFERENCE.md), and browser/server logs (references/vstack/capabilities/nextjs-tests/REFERENCE.md). - Boundary markers: The skill utilizes structured markdown documentation and clear procedural boundaries to isolate instructions from the data being analyzed.
- Capability inventory: The agent is authorized to perform file system writes for 'authorized repairs' (
SKILL.md) and execute shell commands for running tests and performance benchmarks (references/vstack/capabilities/code-tests/REFERENCE.md,references/vstack/capabilities/runtime-performance/REFERENCE.md). - Sanitization: The skill contains explicit mitigation instructions, such as 'Sanitize sensitive evidence' (
references/vstack/capabilities/runtime-performance/REFERENCE.md) and 'without retaining secrets' during log collection (references/vstack/capabilities/nextjs-tests/REFERENCE.md). - [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of npm packages from the
@vllntscope via the npm registry. This is a standard vendor-specific dependency management process. The instructions include security best practices, such as verifying that packages are not deprecated, checking compatibility, and applying a 7-day age gate for non-exempt packages (references/vstack/capabilities/project-alignment/references/default-packages.md).
Audit Metadata