The Agent Skills Directory

[COMMAND_EXECUTION]: The skill workflow in Phases 5 and 8 instructs the agent to execute shell commands such as open, start, and chrome --headless using file paths that incorporate the {brandname} variable. Since this variable is sourced from untrusted user input during the 'Brand Discovery' interview without explicit instructions for sanitization, a malicious brand name containing shell metacharacters could result in arbitrary command execution on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data provided by users which influences subsequent automated actions including file creation and command execution.
Ingestion points: The 'Brand Discovery' interview (Phase 1) is the entry point for untrusted brand briefs and names.
Boundary markers: There are no instructions or boundary markers defined to isolate user input or prevent the agent from following instructions embedded within the brief.
Capability inventory: The skill possesses capabilities to write files, open them using system utilities, and execute a Node.js script that invokes shell commands via execSync.
Sanitization: No sanitization or validation logic is provided for the brand name or brief content before they are used to generate filenames or interpolated into shell commands.
[DYNAMIC_EXECUTION]: The build-assets.mjs script is a configuration-driven pipeline that the agent is instructed to modify at runtime by populating the CONFIG block with data derived from user input. Modifying and executing a script based on unvalidated user-influenced parameters presents a risk of code injection within the script's execution context.
[EXTERNAL_DOWNLOADS]: The asset production phase depends on external Node.js packages including sharp, favicons, pwa-asset-generator, svgo, and png-to-ico, as well as system tools such as resvg, librsvg, pngquant, and oxipng. These are well-known technology tools used for SVG processing and image optimization.

logo-studio