cali-product-testing-ai-code

Warn

Audited by Gen Agent Trust Hub on Jul 3, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references and provides installation instructions for several third-party tools and plugins from non-trusted GitHub repositories and NPM registries. Examples include github.com/cursor/plugins, github.com/PriNova/pi-agent-codebase-workflows, and various packages from users such as @juicesharp, nicobailon, and tintinweb (e.g., pi-intercom, pi-subagents, pi-supervisor).
  • [COMMAND_EXECUTION]: The skill provides various shell command templates for the agent to use during the testing process. These include file system traversal, content searching with rg, and test runner execution (e.g., find . -path ./node_modules -prune -o -name "*.test.*" -print | xargs rg -l "module_name" > affected_tests.txt).
  • [PROMPT_INJECTION]: The skill contains an indirect prompt injection surface. It ingests data from untrusted external sources, including spec-product.md, spec-tech.md, and project files like package.json or requirements.txt.
  • Ingestion points: SKILL.md reads configuration and tech stack details from these files to determine testing breadth and tools.
  • Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the ingested data.
  • Capability inventory: The skill can generate and suggest the execution of shell commands, subagent tasks, and browser interactions based on the ingested data.
  • Sanitization: No explicit sanitization or validation of the content extracted from project files is mentioned before it is used to formulate strategy and commands.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 3, 2026, 12:53 PM
Security Audit — agent-trust-hub — cali-product-testing-ai-code