The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes dynamic context injection in SKILL.md to retrieve project metadata upon loading. It executes commands such as 'cat .infrahub.yml' to access configuration and 'find' to index existing Python checks and GraphQL queries. These operations are scoped to the local directory and are used exclusively to provide relevant context for the AI's assistance in development tasks.
[INDIRECT_PROMPT_INJECTION]: The skill defines a framework for processing external data from GraphQL APIs, which presents a surface for indirect prompt injection. Ingestion points: Untrusted data enters the agent context via GraphQL query results delivered to the 'validate(data)' method of check classes, such as those shown in 'checks/rack_unit_collision.py'. Boundary markers: No explicit delimiters or instructions to ignore embedded commands are mandated for the data being processed. Capability inventory: User-defined checks have access to the Infrahub SDK client ('self.client') and standard Python libraries, allowing for secondary API interactions or system operations that could be exploited by maliciously crafted data. Sanitization: Although the skill focuses on data validation, it does not specify methods for sanitizing the ingested data to prevent it from influencing the agent's instruction flow.
[COMMAND_EXECUTION]: The skill provides documentation for various CLI and build operations. This includes usage of 'infrahubctl' for local check testing and a Dockerfile template using 'uv sync' for managing dependencies within worker environments.
[EXTERNAL_DOWNLOADS]: The skill references external assets required for its documented workflows. A Dockerfile example specifies pulling a base image from the vendor's registry at 'registry.opsmill.io/opsmill/infrahub' and references documentation at 'docs.infrahub.app'. These resources are vendor-owned and considered safe for the intended use case.

infrahub-managing-checks