The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to download the shipsgo-pp-cli tool from the npm registry using npx -y @mvanhorn/printing-press-library and from a GitHub repository using go install github.com/mvanhorn/printing-press-library/....
[REMOTE_CODE_EXECUTION]: Both installation methods (npx and go install) involve downloading and executing code from external repositories at runtime to set up the necessary environment.
[DATA_EXFILTRATION]: The CLI tool supports a --deliver webhook:<url> flag. This allows the output of any command—which may contain shipment IDs, carrier details, or transit history—to be POSTed to an arbitrary external URL. This bypasses standard agent constraints by delegating the network transmission to the underlying binary.
[COMMAND_EXECUTION]: The skill operates by executing the shipsgo-pp-cli binary with arguments constructed from user queries. It includes a which command that takes natural language input to find other commands, which could lead to unexpected execution if the input is malicious.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through the shipsgo-pp-cli which "<query>" command.
Ingestion points: User-provided natural language queries are passed directly as strings to the which subcommand in SKILL.md.
Boundary markers: No delimiters or safety instructions are provided to the agent to treat the user input as data rather than instructions when interpolating it into the command line.
Capability inventory: The shipsgo-pp-cli has the capability to perform network requests (API calls, webhooks), read local state (SQLite), and write configuration (profiles).
Sanitization: There is no evidence of sanitization or escaping of the user-provided string before it is passed to the shell.

pp-shipsgo