The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of a companion CLI binary using npx -y @mvanhorn/printing-press-library. This package is provided by the author of the skill and is required for all operations.
[COMMAND_EXECUTION]: Operational commands are executed via the printing-press-golden-pp-cli binary using the agent's Bash tool. The implementation uses exec.CommandContext with pre-tokenized argument slices, which effectively prevents shell injection. Additionally, the skill includes a sql tool that allows for read-only queries against a local SQLite database, protected by a validator that blocks multi-statement inputs and mutations.
[DATA_EXFILTRATION]: The CLI includes built-in logic to redact API keys and other sensitive credentials from error messages and logs, reducing the risk of accidental credential exposure to the agent context or local files.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by processing data from an external API and providing the agent with write capabilities. (1) Ingestion points: API response data from endpoints like /projects and /tasks in internal/mcp/tools.go. (2) Boundary markers: Not present in the standard output. (3) Capability inventory: Includes shell command execution via Bash and state-mutating API actions. (4) Sanitization: The CLI scrubs terminal control characters from output in internal/cliutil/text.go, but it does not sanitize data against adversarial prompt instructions.

pp-printing-press-golden