The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and summarize untrusted data from external sources, including Dataverse records and user-supplied search queries. While the skill provides references for safe parsing and explicitly recommends against unsafe rendering methods (like innerHTML), the ingestion of this data into an AI summarization pipeline represents a potential attack surface.
Ingestion Points: Data enters the system via the /_api/search/v1.0/summary and /_api/summarization/data/v1.0/ endpoints, as well as via user-supplied arguments passed during skill invocation.
Boundary Markers: The skill instructions and technical references include logic for parsing inline citations and normalizing summary strings to mitigate accidental obedience, though these do not prevent sophisticated adversarial injection.
Capability Inventory: The skill has access to tools for file modification (Write, Edit) and shell execution (Bash), and it performs build operations (npm run build) and site deployments (pac pages upload-code-site).
Sanitization: Technical references (references/framework-equivalents.md) provide framework-specific guidance to avoid unsafe HTML injection by using safe rendering components instead of raw HTML interpolation.
Dynamic Code Generation and Execution: The skill's primary purpose is to dynamically generate service code, framework wrappers (React, Vue, Angular, Astro), and UI components. It also executes local build scripts (npm run build) to verify that the generated code compiles correctly within the user's project environment.
System Command Execution: The workflow involves executing local scripts (scripts/validate-ai-webapi.js) and system tools such as the Power Platform CLI (pac), git, and node to manage site settings, perform deployments, and track project status. These operations are conducted within the scope of the skill's intended development and integration tasks.

add-ai-webapi