The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its prompt amplification workflow.
Ingestion points: In resources/prompt-tips.md, the agent is instructed to fetch external creative briefs and prompt templates from the YouMind-OpenLab repositories on GitHub using the gh api command.
Boundary markers: The instructions do not define any delimiters or safety boundaries for the fetched external content, meaning the agent might interpret malicious instructions embedded in the prompt templates as authoritative.
Capability inventory: The skill possesses significant capabilities, including executing subprocesses for image generation (oma, codex) and interacting with the GitHub API.
Sanitization: No sanitization or validation logic is present for the data retrieved from the remote prompt libraries before it is incorporated into the agent's prompt context.
[EXTERNAL_DOWNLOADS]: The skill relies on fetching content from external sources at runtime.
Evidence: resources/prompt-tips.md contains instructions for the agent to execute gh api to download markdown files from the YouMind-OpenLab/awesome-nano-banana-pro-prompts and YouMind-OpenLab/awesome-gpt-image-2 repositories. The data is then decoded from Base64 and processed locally.
[COMMAND_EXECUTION]: The skill operates primarily by executing shell commands and local CLI tools.
Evidence: The skill invokes oma image generate, codex exec, and gh api. It also instructs the user to run oma update for maintenance, as seen in SKILL.md and resources/execution-protocol.md.

oma-image