The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process prospect data, CRM records, and enrichment results from various external providers (e.g., LinkedIn, HubSpot, Clearbit). This creates a surface for indirect prompt injection if the processed data contains malicious instructions. The skill mentions 'acceptable use gates' and 'relevance' checks, but these primarily target business policy rather than prompt injection defense.
Ingestion points: cargo-ai storage query execute and segment fetch operations used to retrieve model data.
Capability inventory: The agent can execute a wide range of CLI commands via cargo-ai, including triggering batches (batch create), running agents (action execute), and syncing data to CRM systems.
Boundary markers: Mentions the use of 'acceptable use gates' to verify data origin and relevance before outreach.
Sanitization: No specific prompt sanitization or escaping techniques are detailed for external content interpolation.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the Cargo CLI and the skills bundle from official sources. This includes using npm install -g @cargo-ai/cli and npx skills add getcargohq/cargo-skills. These are vendor-owned resources and are documented neutrally.
[COMMAND_EXECUTION]: The skill heavily relies on the cargo-ai CLI to perform its primary functions, including workspace management, model querying, and workflow orchestration. It includes patterns for polling asynchronous operations and handling CLI errors.
[DATA_EXFILTRATION]: The skill includes instructions for the agent to generate and send session summaries, diagnostic reports, and 'session-share' activity to the vendor's management API (workspaceManagement report create). While these actions are documented as requiring user consent and are intended for troubleshooting and product feedback, they involve the egress of session metadata and activity traces.

cargo