The Agent Skills Directory

Indirect Prompt Injection: The skill analyzes authored source code (e.g., JavaScript, HTML, Liquid) to identify Web API usage patterns. Processing untrusted source files introduces a potential surface for indirect prompt injection, where malicious instructions could be embedded in the code to influence the agent. The skill mitigates this through several layers of defense.
Ingestion points: Source files and configuration artifacts are searched in Phase 2.2.
Boundary markers: The prompt does not define specific boundary markers for the ingested code during LLM reasoning.
Capability inventory: The agent has capabilities to edit source files (Phase 5) and upload configuration to Dataverse (Phase 7).
Sanitization: All identified table and column names are validated against an authoritative schema snapshot retrieved directly from the Dataverse environment (Phase 3).
Command Execution: The skill utilizes shell commands to execute local Node.js scripts and the Power Platform CLI (pac). These commands are used for legitimate purposes such as schema retrieval, report generation, and site deployment.
Local scripts: Scripts such as query-table-schema.js and render-migration-report.js are executed using node.
CLI usage: Official pac pages commands are used for environment interaction.
User Consent and Verification: The skill enforces multiple human-in-the-loop checkpoints before any state-changing operations are performed, including site identity verification, plan review, and deployment approval.

migrate-webapi-selectall