The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands to invoke image generation scripts. This includes spawning processes with bun or npx to run a TypeScript wrapper. Evidence is found in references/codex-imagegen.md, which details invocation methods like bun <WRAPPER>/main.ts.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes untrusted user data and includes it in prompts for image generation tools. * Ingestion points: User input is saved to source.md and then structured into structured-content.md. * Boundary markers: The prompt template in references/base-prompt.md uses simple markdown headers for variables like {{CONTENT}} but does not include explicit delimiters or instructions for the model to ignore malicious content within those variables. * Capability inventory: The skill performs file operations, tool calls for image generation, and shell execution. * Sanitization: No specific measures are mentioned for sanitizing user-provided text to prevent prompt injection attacks.
[DYNAMIC_EXECUTION]: The skill performs dynamic execution by discovering and running scripts at runtime. As described in references/codex-imagegen.md, the agent is instructed to find the path to a wrapper script and execute it using a script runner.

baoyu-infographic