codex-computer-use

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill directs the agent to read a local file (~/.codex/plugins/cache/openai-bundled/computer-use/<version>/skills/computer-use/SKILL.md) and treat its content as authoritative instructions.
  • Ingestion points: The agent reads external markdown content from the local filesystem into its active context (SKILL.md).
  • Boundary markers: There are no instructions to use delimiters or ignore potentially malicious instructions embedded in the cached file.
  • Capability inventory: The skill utilizes the mcp__node_repl__js tool for code execution and provides UI interaction capabilities (clicking, typing, screenshots) through the @oai/sky runtime.
  • Sanitization: The skill does not implement validation or sanitization of the content fetched from the local cache directory.
  • [DYNAMIC_EXECUTION]: The skill employs dynamic JavaScript loading via await import("@oai/sky") within a REPL environment to initialize its core functionality.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 10:57 PM
Security Audit — agent-trust-hub — codex-computer-use