The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands for directory management (mkdir), environment isolation via Git (git worktree), and the operation of the codex CLI tool.\n- [REMOTE_CODE_EXECUTION]: By utilizing codex exec, the skill facilitates the execution of code and actions generated by a remote language model. While the skill suggests using a sandbox (workspace-write or read-only), it explicitly notes that these do not provide full security isolation.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it consumes a user-defined or externally-sourced prompt.md file to drive the subagent's actions.\n
Ingestion points: The $run_dir/prompt.md file is read as stdin by the codex exec command.\n
Boundary markers: None specified; the skill relies on the user to define completion criteria within the prompt.\n
Capability inventory: The skill allows file system modifications (edits), web search access (tools.web_search=true), and Git repository manipulation.\n
Sanitization: No validation or sanitization of the prompt content is performed before execution.

delegate-to-codex