The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes ontology definitions from the Fabric environment, which are used to ground the agent and construct data queries. This creates a surface where a maliciously crafted ontology could potentially influence the agent's query generation logic.
Ingestion points: The skill retrieves ontology definitions via the Fabric API and decodes them in grounding-extraction.md.
Boundary markers: While the data is structured into a grounding JSON, there are no explicit instructions to ignore embedded commands within the ontology schema itself.
Capability inventory: The skill has the ability to compose and delegate queries to sibling skills like sqldw-consumption-cli and eventhouse-consumption-cli.
Sanitization: The skill uses a propertyBindings[] mapping to translate ontology properties to physical column names, which provides a layer of structural validation.
Automated Update Checks: The skill is designed to perform version checks by executing Git commands (git fetch, git show) to compare local files with a remote repository. This is a common practice for maintaining tool integrity in development environments.
Use of Authenticated CLI Tools: The skill relies on the Azure CLI (az) for interacting with Fabric services. This ensures that all operations are performed within the security context and permissions already established by the user.
Dynamic Utility Execution: A compact Python script is used to perform Base64 decoding of definition parts. This is a standard utility pattern for ensuring consistent data processing across different operating systems.

fabriciq-ontology-consumption-cli