The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands to detect project structures and environment states. This includes using lsof to check for port conflicts, ls for directory mapping, and environment file operations (e.g., copying .env.example to .env).
[REMOTE_CODE_EXECUTION]: The execution protocol guides the agent to perform package installations (npm install, pnpm install, yarn install) and run the application's development or start commands (npm run start, npx nuxt dev) on the analyzed project. This behavior executes code provided by the analyzed project, which could include malicious post-install hooks or start scripts in package.json if the project repository is not thoroughly vetted.
[DYNAMIC_EXECUTION]: To verify features and take screenshots, the skill depends on the runtime execution of the analyzed project's own code and build processes. This constitutes dynamic execution of third-party, potentially untrusted code within the agent's operating environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests large amounts of untrusted data from the project's source code, manifest files, and configurations to generate a structured report. This creates an attack surface for indirect prompt injection where malicious instructions embedded in the analyzed code could attempt to manipulate the agent's analysis results or exfiltrate environment information.
Ingestion points: Project files located in ~/.lt-showroom/<slug>/checkout (e.g., package.json, .env, controllers, services, and UI components).
Boundary markers: The instructions do not specify explicit delimiters to separate untrusted project content from analysis instructions during interpolation.
Capability inventory: The skill has access to file system reading, shell command execution, package installation, and execution of a local utility script screenshot-session.mjs.
Sanitization: The skill relies on 'Validation Rules' such as requiring file:line references for every claim and prohibiting marketing language, which provides a manual verification layer but lacks automated sanitization of the input data.

analyzing-projects