The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external contact lists (CSV files) and transmit data to enrichment and advertising providers. While this constitutes an attack surface for indirect prompt injection, the skill mitigates risk through programmatic handling and extensive sanitization.
Ingestion points: External data is loaded via ctx.csv in play files like plays/enrich-audience-waterfall.play.ts and plays/build-hash-only-audience.play.ts.
Boundary markers: Instructions are primarily implemented via TypeScript logic rather than natural language prompt interpolation, reducing the risk of accidental obedience to data-embedded instructions.
Capability inventory: The skill utilizes ctx.tools.execute to interact with authorized APIs (Google Ads, Meta, LinkedIn, Aviato, LimaData) and ctx.dataset for local file operations.
Sanitization: Robust validation is present, including email regex checks, LinkedIn URL hostname verification, and mandatory SHA-256 hashing of emails and phone numbers before they are sent to ad platforms.
[EXTERNAL_DOWNLOADS]: The skill's setup instructions include downloading the deepline CLI tool from a vendor-controlled registry (https://code.deepline.com/api/v2/npm/). This represents standard installation behavior for the vendor's primary product.
[COMMAND_EXECUTION]: The skill provides numerous CLI command templates (e.g., deepline plays run) for the agent to orchestrate the audience management workflow. These commands are restricted to the vendor's own tooling and the skill's distributed play scripts.

deepline-ads-audiences