The Agent Skills Directory

[COMMAND_EXECUTION]: Potential for shell command injection through user-controlled arguments. The skill takes a preset-slug ($ARGUMENTS[0]) and an optional --out path from the user and interpolates them directly into several Bash commands, including npx svgo -i output/<slug>.svg ..., node lib/render-cli.js <slug>, and cp output/<slug>.min.svg <path>. An attacker could provide a malicious string (e.g., using ; or &) as a slug to execute arbitrary shell commands within the environment.
[DATA_EXFILTRATION]: Risk of path traversal and arbitrary file manipulation. The slug argument is used to construct a source file path (output/<slug>.svg) without validation, and the --out argument allows users to specify an arbitrary destination path for the cp command. This behavior could be exploited to read files from or write files to unauthorized locations on the system, especially as the skill explicitly attempts to create missing parent directories.
[PROMPT_INJECTION]: Presence of an indirect prompt injection surface through the processing of external SVG data.
Ingestion points: The skill reads and parses content from output/<slug>.svg to verify structure and metadata.
Boundary markers: None. The skill expects valid XML/SVG but does not implement delimiters or warnings to ignore instructions embedded in the file's text or metadata fields.
Capability inventory: The skill has access to shell execution (Bash), file system reads/writes, and local script execution (node lib/render-cli.js).
Sanitization: There is no evidence of sanitization or escaping of the SVG content (e.g., <title> or <desc> tags) before it is processed or included in the final report.
[EXTERNAL_DOWNLOADS]: The skill downloads the SVGO optimization tool from the official NPM registry using npx svgo at runtime.

svg-export