The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents a workflow for processing untrusted external input via a <task> argument or piped standard input (stdin).
Ingestion points: Data enters the agent's context through the command-line task argument and the stdin piping mechanism described in the Core call section of SKILL.md.
Boundary markers: While the skill mentions that stdin content is appended as a <stdin> block, there are no specific instructions or delimiters provided to ensure the agent ignores potentially malicious instructions embedded within that data.
Capability inventory: The codex exec tool is explicitly designed for command_execution and file writing (e.g., via the workspace-write sandbox), which allows successful injections to have significant local impact.
Sanitization: No sanitization, validation, or filtering of the input prompt is mentioned or required by the instructions.
[COMMAND_EXECUTION]: The core functionality of the skill is to facilitate the execution of shell commands and scripts through the codex-cli utility.
Evidence: The skill provides multiple examples of using codex exec to drive an agent to perform tasks, including monitoring background runs where command_execution events are explicitly tracked and logged.

codex-headless