The Agent Skills Directory

Command Execution: The skill utilizes the Bash tool to run local configuration checks, invoke the Power Platform CLI (pac env who), and execute specific metadata discovery scripts (query-table-lookups.js, query-table-relationships.js, and validate-permissions-schema.js). These operations are restricted to querying the environment's own API endpoint and local workspace settings, which is standard for an analysis and auditing tool.
Indirect Prompt Injection Consideration: The skill processes local site source code and Dataverse metadata to generate its report, creating a potential surface for encountering untrusted data.
Ingestion points: Source code files and table configuration YAML files are read dynamically from the workspace via the Glob, Grep, and Read tools in Step 2.
Boundary markers: Specific delimiters instructing the model to ignore natural language instructions embedded within the target files are absent during the analysis phase.
Capability inventory: The skill executes local helper scripts via the Bash tool, writes files to the workspace or temp directory, and can optionally spawn a dedicated agent using the Agent tool if authorized by the user.
Sanitization: The generated HTML report template (assets/audit-report.html) contains explicit client-side escaping routines (esc and escAttr) to ensure that data populated from the findings is safely rendered in the web browser.

audit-permissions