The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill suite processes external website content, creating a surface area for indirect prompt injection where malicious site data could influence the agent's behavior during analysis and planning phases.
Ingestion points: External HTML structure, metadata, and scripts are fetched in website-analyzer/SKILL.md using the WebFetch tool.
Boundary markers: The instructions in website-analyzer/SKILL.md request a broad extraction of content without explicit delimiters or instructions to ignore embedded commands, which may allow malicious content to affect the downstream translation and PRD generation.
Capability inventory: The suite has capabilities for file system modification (Write tool), shell command execution (git, npm), and CI/CD workflow generation in website-builder/SKILL.md.
Sanitization: While the skill translates metrics to plain language, which acts as a summary, the implementation plan and improvement proposal are generated based on the model's interpretation of the potentially untrusted source site.
[COMMAND_EXECUTION]: The orchestrator and builder phases execute shell commands to manage the project lifecycle and build the improved website.
Evidence: SKILL.md and website-builder/SKILL.md use shell commands including git rev-parse, git fetch, git pull, npm install, npx shadcn init, and npm run build.
[EXTERNAL_DOWNLOADS]: The skill suite downloads dependencies and interacts with external repositories as part of its core building functionality.
Evidence: The website-builder/SKILL.md file downloads Node.js packages from the NPM registry and clones or syncs code from GitHub repositories.
[DYNAMIC_EXECUTION]: The builder phase dynamically generates configuration and executable script files at runtime to facilitate project setup and deployment.
Evidence: website-builder/SKILL.md creates vite.config.js and a GitHub Actions workflow file (.github/workflows/deploy-pages.yml) using internal templates to deploy the static site artifact.

website-cloner