The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the Recoup API and third-party connectors such as Gmail and Google Docs.
Ingestion points: Data is pulled into the agent's context through endpoints like GET /artists and external URLs from Google Drive, Sheets, or Docs.
Capability inventory: The agent has access to curl for network requests, the ability to send emails via POST /api/emails, and can execute actions on external platforms (TikTok, Instagram, Google Workspace).
Boundary markers: The instructions lack explicit delimiters or guidance to ignore embedded instructions within the retrieved data.
Sanitization: There is no mention of sanitizing or validating content retrieved from external APIs or documents before processing.
[COMMAND_EXECUTION]: Uses the curl utility to perform REST API operations and fetch documentation.
Evidence: Instructions include curl commands to query api.recoupable.dev and download information from docs.recoupable.dev.
[DATA_EXFILTRATION]: Provides the capability to send data to external endpoints and third-party services.
Evidence: Includes functionality for sending emails and interacting with external social media and productivity connectors as part of its primary purpose.
[EXTERNAL_DOWNLOADS]: Retrieves documentation and schema files from external servers.
Evidence: Fetches llms-full.txt and OpenAPI JSON files from docs.recoupable.dev to assist in endpoint mapping.
[CREDENTIALS_UNSAFE]: Instructs the agent on how to use API keys and access tokens for authentication.
Evidence: The skill references RECOUP_API_KEY and RECOUP_ACCESS_TOKEN environment variables for use in Authorization headers.

recoup-platform-api-access