The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill workflow is designed to ingest data from external sources to ground its prototyping tasks, creating a potential surface for indirect instructions.
Ingestion points: The agent is instructed to inspect current product repositories, design systems, and components, as well as briefs from other skills such as the shape skill (SKILL.md, Step 2).
Boundary markers: The process implements a neutral brief to fix invariants and utilizes a separate judge agent to verify compliance, which acts as a check against unexpected behaviors (SKILL.md, Steps 3 and 5).
Capability inventory: The skill utilizes capabilities to build, run, and render code within the target environment to evidence directions (SKILL.md, Step 4).
Sanitization: No explicit sanitization or filtering of the input from the repository or design system is described before it is processed by the designer agents.
[DYNAMIC_EXECUTION]: The skill involves the generation and execution of temporary code for UI prototypes and logic demonstrations.
Evidence: Instructions include building, running, and rendering UI directions in SKILL.md and creating self-contained HTML files with script blocks for logic testing in logic.md.
Mitigation: The skill emphasizes the use of isolated workspaces for code exploration and explicitly defines a production boundary that forbids the reuse of prototype code for final implementation.
[PRIVILEGE_ESCALATION]: The skill requires designer agents to operate in isolated workspaces with non-overlapping territories, which helps to enforce least privilege and prevents lateral data access or contamination between design candidates.

prototype