The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the @cargo-ai/cli tool (a vendor-provided resource) to interact with the Cargo platform. It executes shell commands for resource discovery, workflow creation, and data management, such as cargo-ai orchestration run create and cargo-ai system-of-record client query.
[REMOTE_CODE_EXECUTION]: The skill defines a framework for creating workflow nodes that execute arbitrary Python and JavaScript code at runtime via native nodes with actionSlug: "python" or actionSlug: "script". This feature allows for dynamic data transformation within the platform's execution environment using context passed from previous nodes.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface. It is designed to ingest untrusted data from external sources like data warehouses and segments (via segment fetch and sor client query) and interpolate this data into AI agent prompts using template expressions (e.g., {{nodes.start.domain}}). Evidence chain:
Ingestion points: Data enters the context through cargo-ai segmentation segment fetch and cargo-ai system-of-record client query.
Boundary markers: Explicit delimiters or 'ignore' instructions are absent in the provided prompt examples (e.g., "prompt": "Summarize the company {{nodes.start.domain}}" in references/nodes.md).
Capability inventory: The skill can execute subprocesses (python/script nodes), perform file downloads, and trigger network operations via third-party connector nodes.
Sanitization: There is no mention of sanitizing or escaping ingested data before it is interpolated into prompts.
[DATA_EXFILTRATION]: The skill provides tools for retrieving and downloading records from a 'System of Record' (data warehouse) and segments. While intended for business orchestration, these commands (client download, record download, batch download) allow for the bulk extraction of data from connected enterprise systems.

cargo-cli-orchestration