The Agent Skills Directory

[SAFE]: The skill is instructional in nature and promotes defensive coding practices. It explicitly warns against connecting prototypes to real databases or production systems and instructs the agent to use scratch data for experiments. It also includes instructions to skip production scaffolding and to ensure prototype code is never promoted to production.
[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that involves reading existing project code and documentation to generate prototypes. While this creates a data ingestion surface, the skill includes process-based boundaries by requiring explicit questions and manual browser reviews of the output.
Ingestion points: The agent is instructed to "Read existing code and docs first" in SKILL.md to identify unknowns.
Boundary markers: Prompt-level delimiters are absent, but the skill requires an explicit statement of the question, observations, and manual browser review before claiming acceptance (SKILL.md, references/logic.md).
Capability inventory: The skill allows for the creation and execution of local scripts, HTML files, and React components (SKILL.md, references/ui.md).
Sanitization: No explicit sanitization or escaping of ingested code/docs is mentioned, but the focus is on generating throwaway, isolated artifacts for manual review.

to-prototype