The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from automotive listings and OEM build data via the Auto.dev API. This represents a potential surface for indirect prompt injection if listing descriptions or dealer comments contain malicious instructions.
Ingestion points: Data returned from /listings and /vin API endpoints.
Boundary markers: None explicitly defined to separate API content from agent instructions.
Capability inventory: File system access for CSV exports, network operations for API calls and third-party integrations (Slack/Email), and app scaffolding generation.
Sanitization: No explicit sanitization of API data is described in the prompt templates.
[EXTERNAL_DOWNLOADS]: The documentation encourages the use of the @auto.dev/sdk package and other standard libraries for Python and Node.js to build automotive applications. These are legitimate vendor resources provided by the skill author.
[DATA_EXFILTRATION]: The skill provides integration recipes for sending vehicle data to external services like Slack and SendGrid. These are documented as functional features for legitimate inventory monitoring and alerting workflows.
[CREDENTIALS_UNSAFE]: Use of the AUTODEV_API_KEY environment variable is documented for authentication. The skill correctly identifies this as a secret and provides standard instructions for managing it in shell profiles and .env files.

auto-dev