The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The bundle-artifact.sh script automatically installs several packages (parcel, @parcel/config-default, parcel-resolver-tspaths, html-inline) using pnpm add -D. These dependencies are fetched from the public NPM registry at runtime without version pinning, which introduces a supply chain risk where a compromised package version could execute malicious code during the build process.
COMMAND_EXECUTION (MEDIUM): The skill facilitates the execution of shell scripts that modify the file system and install software. Specifically, bundle-artifact.sh uses cat to overwrite .parcelrc, rm -rf to delete directories, and pnpm exec to run local binaries. There is a risk of command injection if the user-provided <project-name> or THEME_COLOR variables contain shell metacharacters (e.g., ;, &&), as they are passed directly to the shell in SKILL.md instructions.
INDIRECT_PROMPT_INJECTION (LOW): The skill has a surface for indirect injection by processing user-controlled project names and theme colors into shell execution paths.
Ingestion points: User-provided <project-name> and THEME_COLOR variables in SKILL.md Step 1.
Boundary markers: Absent; inputs are interpolated directly into shell command strings.
Capability inventory: The script bundle-artifact.sh possesses capabilities for file deletion (rm), file writing (cat), and package execution (pnpm exec).
Sanitization: No sanitization logic is present in the provided shell script to escape or validate the project name or theme color before execution.
DATA_EXPOSURE (SAFE): No patterns of sensitive file access (like SSH keys or AWS credentials) or exfiltration to external domains were found.

artifacts-builder