The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of browser-based AI features, which naturally processes text data through language models.
Ingestion points: Functions like promptText in assets/language-model-service.template.ts ingest PromptInput which may contain untrusted data from the web environment.
Boundary markers: The instructions recommend using initialPrompts for system-level instructions to provide structural separation, though this is not a complete defense against adversarial prompts.
Capability inventory: The skill enables text generation, streaming, and tool execution, which define the potential impact of a successful injection.
Sanitization: The skill explicitly advocates for responseConstraint using JSON Schema to validate model outputs and recommends parsing results before use in the application.
[COMMAND_EXECUTION]: The skill includes and instructs the agent to run a local discovery script.
Evidence: SKILL.md contains the command node scripts/find-frontend-targets.mjs ..
Analysis: The script is a standard developer tool that scans the local directory for project files, correctly ignoring sensitive directories like .git and node_modules.

prompt-api