The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/init-artifact.sh script is vulnerable to shell command injection. The first argument passed to the script, $PROJECT_NAME, is used directly in multiple shell commands without sanitization or proper quoting.
Evidence in scripts/init-artifact.sh (line 72): pnpm create vite "$PROJECT_NAME" --template react-ts. While the variable is double-quoted, an attacker could provide a name starting with a hyphen to inject command-line flags.
Evidence in scripts/init-artifact.sh (line 78): $SED_INPLACE 's/<title>.*<\/title>/<title>'"$PROJECT_NAME"'<\/title>/' index.html. The variable is inserted between single-quoted strings, allowing for shell metacharacter injection if the project name contains backticks, semicolons, or other command separators.
[EXTERNAL_DOWNLOADS]: The skill's scripts perform extensive automated downloads and installations of Node.js packages at runtime.
Evidence in scripts/init-artifact.sh: Installs pnpm globally via npm, then uses pnpm to install vite, tailwindcss, radix-ui components, and over 40 other frontend dependencies.
Evidence in scripts/bundle-artifact.sh: Dynamically installs parcel, @parcel/config-default, parcel-resolver-tspaths, and html-inline every time a bundle is created.
[REMOTE_CODE_EXECUTION]: The skill executes code downloaded from the internet during its initialization and bundling phases using pnpm exec and pnpm create.
Evidence in scripts/bundle-artifact.sh: Executes parcel and html-inline directly from the downloaded node_modules.
[NO_CODE]: The skill relies on an opaque local archive scripts/shadcn-components.tar.gz which is extracted into the project source at runtime. The contents of this archive are not visible in the skill's source files, representing unverified code being introduced to the user's environment.

artifacts-builder