The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is a wrapper for the codex CLI, designed for autonomous task execution. It encourages the use of high-risk flags such as --full-auto, which allows the tool to modify files without user confirmation, and --sandbox danger-full-access, which enables system-level operations.
[DATA_EXFILTRATION]: The documentation states that all tool invocations require dangerouslyDisableSandbox: true. This parameter disables the agent's security sandbox to allow network access (primarily for the OpenAI API), which could be leveraged to exfiltrate data from the local environment.
[EXTERNAL_DOWNLOADS]: Under the danger-full-access permission level, the skill documentation explicitly mentions "Installing packages" as a capability, which entails downloading and running third-party software.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting external content and processing it through an autonomous execution engine.
Ingestion points: Untrusted data enters the agent context through prompt arguments or stdin pipes as described in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or boundary markers to isolate untrusted data from the sub-agent's core instructions.
Capability inventory: The tool allows for file system writes, system operations, package installation, and network access.
Sanitization: No sanitization or validation of the input content is implemented before it is passed to the execution CLI.

codex-exec