The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/validate.js utility evaluates the generated scene.js file using new Function. Since the content of scene.js is authored by the AI based on potentially untrusted input (like project source code or user-provided descriptions), this execution path could be exploited if the AI is manipulated into including malicious JavaScript within the file's constants.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data sources such as external codebases and architecture documentation. This data directly influences the generation of executable configuration files.
Ingestion points: The process described in SKILL.md (Step 1) involves exploring entire codebases and reading documentation to build an architecture brief.
Boundary markers: There are no explicit delimiters or instructions provided to the AI to prevent it from following commands found within the analyzed data.
Capability inventory: The skill possesses the ability to execute code via new Function in scripts/validate.js and perform file system writes and deletions through scripts/build.js and bin/cli.js.
Sanitization: The skill lacks sanitization or validation logic to ensure that data extracted from external sources does not contain executable patterns or malicious instructions.
[COMMAND_EXECUTION]: The bin/cli.js installer script performs active file system management, including creating, copying, and deleting files and directories, as well as managing symbolic links. While these are functional requirements for installation, they provide a broad surface for file system interaction.

architecture-canvas