The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx snapai to download and execute code from the NPM registry. The package is maintained by the skill's author (code-with-beto).
[COMMAND_EXECUTION]: The skill performs several command-line operations using tools like npx, magick (ImageMagick), sips, brew, and jq for file and image processing.
[INDIRECT_PROMPT_INJECTION]: The skill contains a command injection vulnerability because it interpolates user-supplied text (icon prompt and style) directly into a shell command (npx snapai icon --prompt "YOUR_PROMPT_HERE"). If a user provides input containing shell escapes like backticks or subshells, it could lead to unauthorized command execution.
Ingestion points: User-specified prompt and style parameters in SKILL.md.
Boundary markers: The command uses double quotes, which can be bypassed with backticks or subshell syntax.
Capability inventory: Access to npx, brew, and local file system utilities.
Sanitization: The instructions provide no sanitization or validation steps for the user input.
[CREDENTIALS_UNSAFE]: The skill handles OpenAI API keys by instructing the user to store them in ~/.snapai/config.json. While it includes explicit warnings to prevent the agent from leaking these keys during verification, the operational flow involves interacting with and verifying these credentials in the local environment.

app-icon