The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs developers to install several packages from public registries, including @channel3/sdk and @channel3/cli via npm, and channel3-sdk via pip. It also references the vendor's official documentation site at docs.trychannel3.com and API endpoint at api.trychannel3.com. All identified resources belong to the author's infrastructure.
[REMOTE_CODE_EXECUTION]: The documentation suggests using npx shadcn@latest add https://ui.trychannel3.com/r/all.json to integrate UI components. This command fetches a remote registry configuration and adds executable React components and hooks directly into the user's project. While this is a remote resource loading pattern, it targets the vendor's official sub-domain.
[INDIRECT_PROMPT_INJECTION]: The skill describes an agentic search mode for POST /v1/search which is designed to process "rich natural-language briefs" and "end-user messages." This creates an ingestion point where untrusted third-party data is passed into an LLM-based planning system.
Ingestion points: The query field in POST /v1/search (SKILL.md) and the message object in POST /v1/conversations (references/conversations.md).
Boundary markers: The documentation does not provide instructions for developers to use delimiters or "ignore instructions" headers when interpolating user data into these fields.
Capability inventory: The skill enables product search, recommendation retrieval, and conversational agent interaction. It does not perform direct shell execution or file system writes on the agent's host, but the results influence the agent's shopping decisions.
Sanitization: No specific sanitization or filtering logic for the input query is mentioned in the client-side integration guides.

channel3-api