The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill guides the user to clone an external repository from GitHub (https://github.com/StageDog/tavern_helper_template.git) and execute installation and build commands (pnpm install, pnpm build). These operations download and execute code from sources outside the immediate skill package.
[DYNAMIC_EXECUTION]: The instructions involve using $('body').load() to pull HTML and JavaScript content from remote CDN mirrors (such as testingcf.jsdelivr.net) or local development servers. This allows for the dynamic injection of external code into the operating environment.
[COMMAND_EXECUTION]: The skill provides numerous shell commands for the agent to execute, including git clone, pnpm, npx vue-tsc, and file system operations using cat and output redirection to write configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data, including schemas and templates, that could potentially contain malicious instructions intended for the agent or the SillyTavern environment.
Ingestion points: Reads schema.ts, App.vue, and index.html from the user's project directory.
Boundary markers: Uses triple backticks for code blocks and specific file path markers.
Capability inventory: Includes shell command execution, file system writes, and access to SillyTavern's chat and variable APIs.
Sanitization: Utilizes Zod for schema validation and static type checking to mitigate the processing of malformed data.
[OBFUSCATION]: The skill documentation mentions a mechanism (// @obfuscate) to trigger code obfuscation via WebpackObfuscator during the build process. While intended for user output, it demonstrates a capability for managing non-human-readable code.

tavern-ui