NYC

shell-testing-framework

Fail

Audited by Gen Agent Trust Hub on Feb 15, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection (HIGH): The skill possesses a high-risk attack surface because it ingests untrusted external content (user-provided shell scripts) and has the capability to execute that content to verify coverage and performance targets.
  • Ingestion points: User-provided shell script files and function definitions provided via prompt or filesystem (referenced in README.md).
  • Boundary markers: None identified; the skill assumes all provided code is legitimate targets for execution.
  • Capability inventory: The README.md examples demonstrate the use of subshells $(...) and source commands in test-module.sh, which allow for arbitrary command execution.
  • Sanitization: No mention of sandboxing, linter-based validation, or restricted execution environments for the shell code being tested.
  • Dynamic Execution (HIGH): The skill generates and executes shell scripts (.sh) at runtime. While this is the stated purpose of a testing framework, the lack of isolation when executing dynamically generated or user-provided scripts poses a high risk to the host environment.
  • Metadata Poisoning (MEDIUM): The README.md claims the skill is already installed at a specific path (~/Library/Application Support/Claude/skills/shell-testing-framework/). This type of authoritative metadata can be used to deceive agents into trusting the skill's origin or state without verification.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 15, 2026, 11:15 PM