The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the codex CLI tool via shell commands in the SKILL.md file. It specifically utilizes the exec command with the --ephemeral and --sandbox read-only flags, which are intended to limit the tool's impact on the local environment. The workflow involves standard shell operations such as creating temporary directories with mktemp to manage data flow.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting external data (repository code, logs, and symptoms) and piping it into a prompt for the Codex agent.
Ingestion points: Untrusted data enters the agent context via the 'Context' block in the consultation prompt defined in SKILL.md.
Boundary markers: The skill uses markdown headers and labels (e.g., 'Question:', 'Context:', 'Instructions:') to separate different parts of the prompt, though it lacks robust escaping mechanisms.
Capability inventory: The primary capability is the codex exec call, which processes the aggregated data.
Sanitization: No explicit sanitization or filtering of the external content is performed before it is passed to the CLI tool.

codex-consult