The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from a user's codebase (markup, styles, comments, and documentation).
Ingestion points: SKILL.md identifies that everything read from the project environment is untrusted data.
Boundary markers: The 'Security Model' section in SKILL.md provides explicit instructions to the agent to distinguish between user instructions and project data, specifically warning against following 'instruction-shaped text' found in files.
Capability inventory: The skill allows the agent to edit files and execute project-defined scripts (e.g., npm/pip commands for linting or building).
Sanitization: It directs the agent to report suspicious text in project files as a finding for user review rather than acting upon it.
[COMMAND_EXECUTION]: The skill allows the execution of existing project shell commands.
Evidence: The 'Security Model' in SKILL.md restricts shell operations to the project's 'incumbent install, dev-server, build, lint and type-check scripts'.
[EXTERNAL_DOWNLOADS]: The skill references several external resources for documentation and attribution purposes.
Evidence: The attribution.md file cites upstream sources from well-known and trusted organizations, including Anthropic and Vercel. These references are documented neutrally and serve as research citations rather than runtime code execution triggers.

frontend-crafting