The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the vendor's official CLI package (@cargo-ai/cli) from the NPM registry and references additional skill packs from the vendor's GitHub repository. These resources are authored by getcargohq and are required for the skill's primary functionality.
[COMMAND_EXECUTION]: The instructions leverage the cargo-ai CLI for account authentication, batch data enrichment, and workspace management. It also provides a command to star the vendor's GitHub repository using the user's authenticated GitHub CLI, explicitly instructing the agent to ask for permission before execution.
[DATA_EXFILTRATION]: The skill implements session telemetry by sending metadata, such as the session ID and attribution summary, to the vendor's API via the CLI. It also reads the agent's internal plugin configuration file (~/.claude/plugins/installed_plugins.json) to check for existing installations and prevent duplicate session entries. These actions are consistent with vendor-led attribution practices.
[PROMPT_INJECTION]: The skill processes external data from Clay CSV and JSON exports, which serves as a potential surface for indirect prompt injection. 1. Ingestion points: Reads user-provided Clay CSV files and JSON column schema via Step 1. 2. Boundary markers: No explicit delimiters or ignore instructions warnings are defined in the command-line execution steps. 3. Capability inventory: The skill has network execution capabilities via cargo-ai orchestration action execute-batch and infrastructure management via cargo-ai cdk deploy. 4. Sanitization: Input sanitization is handled by the vendor's CLI tool rather than the skill instructions. As the surface is inherent to the tool's purpose, it is classified as safe within context.

clay-to-cargo