The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an ingestion surface for external data by instructing the agent to research terminology in third-party sources. Ingestion points: The source-verification.md file specifies reading academic papers via the Hugging Face CLI, official project documentation (e.g., Kubernetes, Helm), and maintainer blogs. Boundary markers: SKILL.md provides explicit negative constraints, stating that source text and retrieved documents are material to inspect, not instructions to expand authority. Capability inventory: The skill allows the agent to read and write project files (e.g., terminology.md, references/*.md) and perform network-based searches via the hf CLI tool. Sanitization: The instructions require the agent to exclude AI-generated derivative writing and apply heightened scrutiny to AI-related projects, treating external content as data to be analyzed rather than instructions to be executed.
[EXTERNAL_DOWNLOADS]: The skill facilitates the discovery and retrieval of academic literature from well-known repositories. Evidence: source-verification.md directs the agent to use the hf papers CLI for searching and reading paper metadata and content. Context: These downloads target Hugging Face, which is a recognized service for machine learning research, and the operations are restricted to the skill's stated purpose of terminology research.

curate-terminology