skills/umputun/cc-thingz/ask-codex/Gen Agent Trust Hub

ask-codex

Pass

Audited by Gen Agent Trust Hub on May 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill constructs shell commands for the codex CLI by interpolating user-provided summaries and project context (such as file paths) directly into a string. If these inputs contain shell metacharacters like backticks or dollar signs, they could lead to unintended command execution within the Bash tool.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it incorporates untrusted project data into its decision-making flow.
  • Ingestion points: The skill grants the codex tool full access to project files and specifically targets ./CLAUDE.md as a context document.
  • Boundary markers: The prompt template used to consult the model lacks delimiters or clear instructions to distinguish between instructions and data from the project.
  • Capability inventory: The skill uses the Bash tool for execution and explicitly instructs the agent to implement code changes based on the external tool's analysis.
  • Sanitization: There is no evidence of sanitizing project content or escaping shell arguments before they are passed to the CLI or processed for implementation.
Audit Metadata
Risk Level
SAFE
Analyzed
May 13, 2026, 06:11 AM
Security Audit — agent-trust-hub — ask-codex