The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches source code and container images from the vendor's official repositories and registries. Specifically, it references https://github.com/autonnel/autonnel for source checkout and ghcr.io/autonnel/autonnel for Docker deployments.
[COMMAND_EXECUTION]: The instructions guide the operator through standard deployment and management tasks using well-known tools such as wrangler for Cloudflare Workers, docker compose for container orchestration, and npm/pnpm for Node.js package management.
[DATA_EXPOSURE]: The skill implements secure credential handling by advising the use of environment variables and platform-native secret managers (e.g., wrangler secret put). It explicitly warns against hardcoding sensitive information like DATABASE_URL or encryption keys in configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill describes an MCP tool surface that allows an agent to programmatically manage funnel data. While this creates a theoretical attack surface for indirect prompt injection, it is the primary intended purpose of the skill and incorporates tenant-based authentication and write-access controls.
Ingestion points: Tools such as get_page, get_funnel, and list_pages allow the agent to retrieve external data from the self-hosted instance.
Boundary markers: The instructions do not specify explicit delimiters or "ignore embedded instructions" warnings for the data retrieved via the MCP interface.
Capability inventory: The agent has capabilities to modify content (update_page, create_page), upload media, and manage funnel configurations.
Sanitization: The skill does not provide specific instructions for sanitizing content before processing, relying on the application's internal handlers.

self-hosted-funnel-launch