The Agent Skills Directory

Indirect Prompt Injection: The skill ingests untrusted user inputs such as customer, industry, persona, and problem to generate a persona brief and drive a simulated Design Thinking session. This creates a potential surface where malformed or adversarial inputs could influence the agent's behavior during the coaching session or affect the content of generated artifacts.
Ingestion points: customer, industry, persona, and problem inputs are processed in SKILL.md to generate the brief.
Boundary markers: None explicitly specified for input sanitization.
Capability inventory: File writing to the local workspace (.copilot-tracking/ and prototype files).
Sanitization: No explicit validation or filtering logic is described for the user-supplied strings.
Dynamic Execution / Script Generation: The skill automatically generates a runnable prototype scaffold, which includes creating executable JavaScript files (app.js, telemetry.js) and fixtures based on the Design Thinking session's outcomes. While this is necessary for the skill's purpose of creating a hi-fi prototype, the generation of active code from session context is a pattern that warrants review in development environments.

demo-setup