The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The environment check script doctor.py facilitates the installation of the uv package manager by downloading its official installer from astral.sh and piping it directly to the shell. This is a standard and recommended installation method for this well-known tool.
[DYNAMIC_EXECUTION]: The generate.py script performs runtime dependency management to enable local image generation. If the mflux library is not found, the script uses os.execv to re-execute itself through the uv tool, dynamically injecting the mflux package from the Python Package Index (PyPI) for the duration of the run.
[COMMAND_EXECUTION]: The skill integrates with several local system utilities to function:
It invokes the codex CLI via subprocess.run to perform image generation tasks.
It utilizes the macOS security command-line tool in keychain.py to securely store and retrieve user API keys from the system's encrypted keychain, which is a security best practice.
It uses sysctl to verify available system RAM and the open command to display generated assets to the user on macOS.
[EXTERNAL_DOWNLOADS]: The backends.py script makes network requests to well-known cloud services including Replicate, OpenAI, and Google Gemini to generate and download image assets. These operations are performed over secure connections to official service endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests user-supplied prompts in generate.py and interpolates them into the multi-line instructions sent to the codex tool. While no explicit sanitization is performed on the prompt text itself, the downstream execution environment in the codex tool is restricted by a read-only sandbox.
Ingestion points: generate.py (prompt argument)
Boundary markers: Absent
Capability inventory: subprocess.run (to codex CLI), network access, file system writes
Sanitization: Absent

generate-image