agentic-verify

Fail

Audited by Gen Agent Trust Hub on Nov 14, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill is configured to fetch a runner component from a remote repository at github.com/timini/agentic-verification using the uvx package runner during its execution cycle.
  • [REMOTE_CODE_EXECUTION]: The skill executes the downloaded external tool directly as a shell command. Although the download is pinned to a specific commit hash, the repository is an untrusted third-party source not associated with the official skill author or any trusted organization.
  • [COMMAND_EXECUTION]: The skill requires the agent to perform multiple shell command executions, including uvx and pytest, to facilitate mutation testing and code verification, which grants the skill broad access to the local execution environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project files and git history, creating a surface for indirect prompt injection. When combined with its capability to execute shell commands, this represents a notable vulnerability.
  • Ingestion points: The skill reads source code files and git revision metadata from the <project-root> as specified in the SKILL.md workflow.
  • Boundary markers: Absent. The skill provides no instructions to the agent to treat ingested project data as untrusted or to ignore instructions embedded within those files.
  • Capability inventory: The skill possesses shell execution capabilities for uvx and pytest based on parameters derived from project analysis.
  • Sanitization: Absent. No validation or sanitization of ingested code content or commit messages is performed prior to processing.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 14, 2026, 04:34 PM