The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources such as web pages, people databases, and registries to populate research claims.
Ingestion points: External data is ingested through various retrieval tools like official_web, people_database, and public_registry described in references/adaptive-research.md and jobs/finding.md.
Boundary markers: plays/shared/rerank.ts uses <untrusted_content> tags to delimit external data and provides an explicit SECURITY warning to the LLM judge regarding potential adversarial content.
Capability inventory: The skill framework enables executing provider tools, writing to durable datasets, and generating outreach copy using deeplineagent.
Sanitization: The plays/shared/rerank.ts file includes an escapeUntrusted function to escape angle brackets, and plays/shared/research-experiment.ts ensures that claim values are literally matched against raw source text.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the deepline CLI using a custom NPM registry https://code.deepline.com/api/v2/npm/. This domain is associated with the vendor and is a standard distribution channel for their tools.
[COMMAND_EXECUTION]: The instructions include usage of the deepline CLI for authentication, play execution, and inspecting runtime data using the deepline db query command.
[DYNAMIC_EXECUTION]: The skill architecture involves the creation of TypeScript .play.ts files which are executed by the platform's managed runtime; this is the primary intended functionality of the skill.

deepline-plays