The Agent Skills Directory

[METADATA_POISONING]: The skill metadata in SKILL.md and metadata.json attributes the project to the 'Laravel Community'. This conflicts with the provided author context 'asyrafhussin', creating a deceptive impression of official framework support.
[INDIRECT_PROMPT_INJECTION]: The skill instructs on building agents that process external data, creating a surface for indirect injection.
Ingestion points: Code in rules/agent-prompting.md and rules/files-vector-stores.md shows ingestion of data from user messages, file uploads, and external URLs.
Boundary markers: No specific instructions are provided for implementing boundary markers or system prompt instructions to handle embedded commands within untrusted data.
Capability inventory: Agents possess significant capabilities including writing to the file system (Image::store), making network requests to AI providers, and executing custom application code through tools.
Sanitization: Examples lack demonstration of content sanitization or validation of external data sources.
[EXTERNAL_DOWNLOADS]: The skill references and demonstrates fetching resources from external locations.
Evidence: Documentation in rules/files-vector-stores.md and rules/media-images.md includes examples of fetching files and images from https://example.com/doc.pdf and https://example.com/reference.jpg.
Neutral Reference: The skill references the laravel/ai package from the Laravel organization's repository.

laravel-ai-sdk