The Agent Skills Directory

[DYNAMIC_EXECUTION]: The shell scripts scripts/generate.sh and scripts/setup.sh reach outside the skill's own directory structure to execute Python files located in the parent project's directory.
Evidence: PLUGIN_ROOT="$(cd "${SCRIPT_DIR}/../../.." && pwd)" and exec python3 "${PLUGIN_ROOT}/scripts/codex-agents/install.py".
This behavior creates a dependency on external code that is not contained within the audited skill package itself.
[COMMAND_EXECUTION]: The shell scripts use exec ... "$@" to pass all provided arguments directly to the underlying Python scripts.
Evidence: Found in scripts/generate.sh and scripts/setup.sh.
While standard for CLI tools, it allows for arbitrary argument passing to the installer scripts.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to install an agent configuration file (sigma-auth-guide.toml) which can influence the agent's behavior in subsequent sessions.
Ingestion points: The installer writes configuration files based on its execution logic and potentially user-provided arguments.
Boundary markers: None mentioned in the provided scripts to distinguish between instructions and data in the generated configuration.
Capability inventory: The skill has the ability to write to the file system (implied by 'installer' purpose) and execute Python subprocesses via scripts/setup.sh.
Sanitization: No sanitization or validation of the input arguments is visible in the shell scripts before they are passed to the Python installer.

codex-agent-setup