The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing external commands via subprocess.run to perform its core functions.
In scripts/generate_icon.py, the generate_icon_codex function invokes the codex CLI with a natural language instruction constructed from user input.
In examples/icon_set.py, the script executes the generate_icon.py tool as a subprocess to create a batch of icons.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection where malicious instructions could be embedded in data processed by the agent.
Ingestion points: User-provided strings for icon descriptions (prompt) and batch items (batch) are accepted as command-line arguments in scripts/generate_icon.py.
Boundary markers: Absent. There are no delimiters or specific instructions to the underlying model to ignore embedded commands within the user-provided prompt.
Capability inventory: The skill can execute shell commands through the codex CLI and write PNG files to the local file system.
Sanitization: Absent. The script directly interpolates user input into the command string. A developer comment in scripts/generate_icon.py explicitly notes that the prompt is not sanitized because it is assumed to come from a trusted shell session.
[EXTERNAL_DOWNLOADS]: The skill downloads external assets during runtime.
Selecting the birefnet transparency method causes the rembg library to download a ~400 MB ONNX model file from a remote source to the local ~/.u2net directory.
[CREDENTIALS_UNSAFE]: The skill probes the local environment for authentication data.
It checks for the existence of a Codex authentication file at ~/.codex/auth.json and loads secrets from local and home-directory .env files to authenticate with the OpenAI API.

transparent-icons