maintain-verification-skill

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code and feature files during the source analysis wave (Step 2 and 3).
  • Ingestion points: Reads feature files and project source code via a swarm of reader agents.
  • Boundary markers: No explicit delimiters are specified in the instructions for separating ingested source data from agent instructions.
  • Capability inventory: The agent can execute harness scripts, modify files within the skill's directory, and create draft Pull Requests based on its analysis.
  • Sanitization: The instructions do not mandate specific sanitization for data extracted from source code before it is used to reconcile the feature map.
  • [DYNAMIC_EXECUTION]: The skill executes and potentially modifies harness scripts (Step 4 and 5) to drive the application during the live verification pass.
  • Execution method: Invokes project-local scripts as defined by the verification skill's own launch model.
  • Context: This behavior is intrinsic to the skill's purpose as a test automation and verification tool, operating within the scope of a local development environment.
  • [EXTERNAL_DOWNLOADS]: The skill's documentation (UPSTREAM.md) references the official Cursor plugins repository on GitHub for maintenance and fidelity tracking.
  • Evidence: References github.com/cursor/plugins for upstream source synchronization.
  • Context: The references target a well-known service and are documented neutrally for provenance and maintenance purposes.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 01:27 PM