The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/gen_cover.py attempts to read the user's ~/.zshenv file to extract the OPENAI_API_KEY when the variable is not found in the environment. Accessing shell configuration files to harvest secrets is an unsafe practice for credential management.
[COMMAND_EXECUTION]: The script scripts/gen_cover.py uses the subprocess module to execute external commands, including curl for interacting with the OpenAI API and codex exec for local image generation tasks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from article or video content to generate image prompts and cover copy (SKILL.md). This represents a potential attack surface where malicious content could influence the agent's behavior or output.
Ingestion points: Reads article text and video scripts from user-specified directories or the Obsidian vault (SKILL.md Step 1).
Boundary markers: The workflow does not explicitly use delimiters or instruction-ignore markers when processing external content.
Capability inventory: The skill possesses shell execution capabilities through scripts/gen_cover.py and file-write capabilities across the workspace.
Sanitization: There is no evidence of content sanitization or filtering before the article data is interpolated into prompts.

ray-cover