The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs dependencies from the author's official GitHub repository (github.com/mvanhorn/printing-press-library) and via a vendor-scoped NPM package (@mvanhorn/printing-press-library). These are recognized as legitimate vendor resources.
[COMMAND_EXECUTION]: The skill operates by executing the costco-pp-cli binary. It includes standard utility commands for data retrieval (receipts, orders), analytics (spend, savings), and local database management (sync, sql, search).
[DATA_EXFILTRATION]: While the CLI includes features for data routing via webhooks (--deliver webhook:<url>) and a feedback mechanism (COSTCO_FEEDBACK_ENDPOINT), these are documented functional capabilities of the tool rather than malicious exfiltration patterns. Users and agents should be aware that output can be directed to external URLs if configured.
[CREDENTIALS_UNSAFE]: Authentication is handled via short-lived bearer tokens stored in a local credentials.toml file. This follows established best practices for secret management and does not involve hardcoded credentials in the skill instructions.
[DATA_INGESTION_SURFACE]: The skill ingests external data from the Costco API.
Ingestion points: Data enters the agent context via CLI commands like receipts, orders, and raw.
Boundary markers: The tool provides an --agent flag to ensure output is formatted as structured JSON, which helps the agent distinguish between data and instructions.
Capability inventory: The agent has access to Bash and Read tools, which are used to invoke the CLI and read its output.
Sanitization: The skill does not explicitly describe sanitization logic, relying on the structured output and the agent's internal filtering.

pp-costco