The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted web content from user-provided URLs to recreate websites. This creates a surface where a malicious site could embed instructions to influence the agent.
Ingestion points: Data enters via tools like capture_page, inspect_page, and audit_page (e.g., SKILL.md).
Boundary markers: The skill does not explicitly instruct the agent to use delimiters or ignore embedded instructions when analyzing external content.
Capability inventory: The agent has the ability to write files to the local system and execute shell commands via npm run build (e.g., Phase 2 and Phase 4).
Sanitization: Certain tools like analyze_network_capture redact sensitive fields, but the goal of 'faithful implementation' involves copying logic from untrusted sources.
[EXTERNAL_DOWNLOADS]: The skill documentation refers to installing the spectorjs package and performing npm install for generated projects (e.g., v2-routing.md, Phase 4).
[DYNAMIC_EXECUTION]: The workflow involves generating a project based on external references and subsequently running npm run build on the resulting code (e.g., Phase 4).

recreate