The Agent Skills Directory

[DATA_EXFILTRATION]: The skill documents a --deliver webhook:<url> flag for its CLI tool, which allows the output of any command to be POSTed to an arbitrary external URL. This represents a potential exfiltration vector if the agent is manipulated into routing sensitive local data or configuration information to an attacker-controlled endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill fetches and renders content from dozens of external recipe websites, creating a surface for embedded instructions to influence the agent's behavior.
Ingestion points: External recipe content fetched from third-party URLs via recipe get, goat, and search commands.
Boundary markers: Absent. The skill instructions do not specify the use of delimiters or warnings to prevent the agent from obeying instructions embedded in the external data.
Capability inventory: The agent can execute local shell commands via the recipe-goat-pp-cli, read and write to a local SQLite database, and modify local configuration files.
Sanitization: Absent. There is no evidence of content sanitization or validation for the data retrieved from external sources before it is presented to the agent.
[EXTERNAL_DOWNLOADS]: The skill fetches its core CLI and MCP components from the vendor's official GitHub repository and NPM package registry.
Evidence: Fetches the library from @mvanhorn/printing-press-library via npx.
Evidence: Downloads Go modules from github.com/mvanhorn/printing-press-library.

pp-recipe-goat