The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the agent to construct and run several shell commands including magick, vtracer, svgo, and rm. These commands rely on variables like INPUT_PNG and OUTPUT_SVG that are derived from user input or the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface where untrusted data (filenames/paths) is used to build executable command strings. An attacker could potentially achieve command execution by providing a filename containing shell metacharacters.
Ingestion points: The INPUT_PNG and OUTPUT_SVG variables in SKILL.md which are sourced from user requests or local file names.
Boundary markers: None; the skill does not use delimiters or specify that the agent should treat the paths as literals or use shell-safe escaping.
Capability inventory: The skill has the ability to execute arbitrary subprocesses through the shell, including image manipulation and file deletion commands in SKILL.md.
Sanitization: Absent; the instructions do not include steps to validate, escape, or sanitize the input variables before they are placed in a bash execution context.
[EXTERNAL_DOWNLOADS]: The skill references and requires the installation of several external software packages from well-known and reputable sources.
Mentions installation of imagemagick through the Homebrew package manager.
Mentions installation of vtracer via the Cargo package registry for Rust.
Mentions installation and execution of svgo via the NPM package registry.

lov-png2svg