The Agent Skills Directory

Isolated Orchestration: The skill employs a multi-phase orchestration pattern using the delegate tool to manage sub-sessions. Each review lens is executed in an isolated environment with no shared history (context_depth="none"), which is a recommended practice to ensure independent analysis and prevent cross-contamination between different perspectives.
Data Boundary Management: To address considerations related to processing external information (Indirect Prompt Injection), the skill includes a specific phase for creating a "Review Brief." This brief serves as a clean, distilled version of the target under review, separating it from the conversational history before it is passed to the review lenses. This provides a clear boundary for untrusted data ingestion points (like conversation context or external files) and leverages the agent's capability to distill and sanitize content.
Template-Based Generation: The skill functions as a generator, using predefined templates to scaffold new skill files and configurations. This dynamic generation approach is used to create standardized artifacts that include built-in trust guardrails, such as roster manifests and attribution requirements, ensuring that generated councils maintain a consistent security posture.
External Resource References: The skill's templates reference external configuration files and modules. These references point to repositories within the author's official GitHub organization and are used to provide the necessary foundation for the generated skills. These are documented as standard external dependencies for the skill's infrastructure.
Automated Validation: The instructions include steps for running validation recipes to ensure the generated artifacts meet specific quality and structural standards. This includes using the bash tool to invoke local validation utilities, providing an additional layer of verification before a new council is finalized.

councilify