The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the ucp-pp-cli and ucp-pp-mcp tools from the author's official NPM and GitHub repositories.
Evidence: npx -y @mvanhorn/printing-press-library install ucp --cli-only downloads and runs an installer from the author's NPM package.
Evidence: go install github.com/mvanhorn/printing-press-library/library/commerce/ucp/cmd/ucp-pp-cli@latest downloads and compiles code from the author's GitHub repository.
[COMMAND_EXECUTION]: The skill's primary function is to execute local CLI commands via the shell using the ucp-pp-cli binary.
[DATA_EXFILTRATION]: The CLI tool documented in the skill includes a feature that allows command output to be sent to external URLs.
Evidence: The --deliver webhook:<url> flag allows the agent to POST command results to an arbitrary endpoint, which could be used to transmit data outside the environment.
[PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection by processing data from untrusted external sources.
Ingestion points: The check <domain> command fetches and parses /.well-known/ucp manifest files from external, potentially attacker-controlled merchant websites.
Boundary markers: There are no instructions provided to the agent to disregard or sanitize potential commands or directions embedded within the fetched manifest data.
Capability inventory: The skill possesses the capability to execute shell commands, write to the local file system, and perform network requests via the CLI's delivery features.
Sanitization: The documentation does not describe any validation or escaping mechanisms for the external manifest data before it is processed by the agent.

pp-ucp