The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md file defines shell command templates, such as potrace "tmp/output/<name>.pbm" -o "src/assets/icons/<name>.svg", which incorporate user-supplied variables like <name>. Without proper sanitization by the agent, this pattern creates a risk for command injection or path traversal, potentially allowing a malicious user to execute arbitrary shell commands or overwrite sensitive files on the host system.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data and uses it to drive powerful capabilities.\n
Ingestion points: User-specified icon 'subjects' and 'names' gathered during the Phase 1 discovery process.\n
Boundary markers: The instructions lack explicit delimiters or safety warnings to prevent user-supplied strings from being interpreted as instructions or command flags.\n
Capability inventory: Extensive use of subprocess calls to local Python scripts (crop_and_trace.py, remove_chroma_key.py) and external binaries (potrace, svgo).\n
Sanitization: No input validation or character escaping is defined for user-supplied strings before they are interpolated into shell commands.\n- [EXTERNAL_DOWNLOADS]: Fetches and executes the svgo utility from the npm registry using bunx to perform SVG optimization. This is a standard workflow for asset optimization and utilizes a well-known technology service.

custom-icons