The Agent Skills Directory

Command Execution: The skill uses the shell tool to run an initialization script.
Evidence in SKILL.md: bash "${PLUGIN_ROOT}/scripts/plugin-init.sh"
This script is responsible for creating the project structure and initial files. While this is the intended function of the skill, it involves running shell commands that modify the local environment.
Environment Modification: The skill requests permission to write to the .codex/ directory and perform Git operations.
Evidence in SKILL.md: "The script writes .codex/ and runs git init... please approve it when prompted."
These operations are typically restricted by sandbox policies. The skill transparently informs the user that these permissions are required for the workspace setup.
Invocation Security: The skill's configuration includes a policy to prevent accidental triggering.
Evidence in agents/openai.yaml: allow_implicit_invocation: false
This setting ensures the skill only runs when explicitly requested by the user, reducing the risk of unintended modifications to the environment.
Indirect Data Processing: The skill's behavior is influenced by the terminal output of the initialization script.
Ingestion points: The agent reads the final RESULT: line from the plugin-init.sh script (documented in SKILL.md).
Boundary markers: The agent specifically looks for the predefined strings initialized and already-initialized to determine the outcome.
Capability inventory: The skill utilizes the shell tool to execute the script and perform filesystem modifications.
Sanitization: The skill performs string matching on the script's terminal output to decide which verbatim instruction to follow.

init-physics-intern