The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires downloading a CLI tool from a remote repository.
Source: github.com/mvanhorn/printing-press-library/library/commerce/shopping/cmd/shopping-pp-cli via Go.
Source: @mvanhorn/printing-press-library via NPX.
[REMOTE_CODE_EXECUTION]: The skill uses npx and go install to download and execute code from external registries during the setup process.
[COMMAND_EXECUTION]: The skill requires the Read Bash tool to execute the shopping-pp-cli binary for retail data analysis and profitability calculations.
[DATA_EXFILTRATION]: The skill includes functionality to transmit tool output to external network endpoints.
Feature: The --deliver webhook:<url> flag allows the agent to POST command results to arbitrary external URLs.
Feature: The feedback --send feature transmits locally stored notes to a configured network endpoint.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it aggregates content from multiple external retailers.
Ingestion points: Product names, descriptions, and price history fetched from over 70 third-party retailers via the index and compare commands.
Boundary markers: None explicitly defined in the skill instructions to separate untrusted product data from agent instructions.
Capability inventory: The agent has access to the Bash tool and can execute shell commands with the results of retail queries.
Sanitization: No explicit sanitization or filtering of external product content is specified before processing.
Mitigation: Wrap external content in delimiters with explicit instructions to ignore embedded commands, and utilize the --json flag with strict schema validation.

pp-shopping