The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes complex external data including JSON workflow plans, design projects (.vectorcraft), and external assets (SVG, PNG, PDF). This ingestion of untrusted data represents an attack surface where malicious instructions could theoretically be embedded.
Ingestion points: workflow.py and commands.py ingest JSON operation plans, design projects, and external assets via the --asset parameter.
Boundary markers: The skill implements integrity checks using expectedProjectSha256 and manifest.json to ensure project consistency before performing operations.
Capability inventory: The skill possesses the ability to execute CLI commands, write files to the local filesystem (save/export), and install dependencies via npx.
Sanitization: The skill performs structural JSON validation through the check command and enforces explicit filesystem root authorization using --read-root and --write-root parameters to limit process access.

vectorcraft-cli-artboards