agently-playbook

The agently-playbook serves as the foundational skill for developers and architects building model-powered applications with the Agently framework. It is specifically designed to handle broad, under-specified product requests where the user aims to build assistants, internal tools, automated workflows, or evaluators without knowing the exact implementation path. Instead of jumping straight into code, this playbook guides the agent to perform requirement reduction, structure a repository skeleton, and decide on appropriate owner layers and async boundaries before selecting specific framework capabilities. It acts as the intelligent bridge between natural language business goals and technical execution, ensuring the resulting project adheres to Agently's native design philosophy rather than adopting generic, framework-agnostic habits.

• Project scaffolding and architectural design for Agently-native applications.

• Automated mapping of business scenarios to core capabilities like TriggerFlow, model-setup, prompt-management, and output-control.

• Native-first guidance prioritizing async-first execution, streaming support, and clear separation between settings, prompts, services, and domain contracts.

• Strategic decision-making for complex integrations, such as deciding between local models (Ollama) versus cloud providers, or choosing when to wrap a tool as an MCP or a custom service.

• Refactoring existing, cluttered implementations into a modular, testable structure suitable for production-ready model applications.

• Always start with this skill for unresolved, high-level product intent or broad refactor requests.

• Follow the five-step workflow: reduce scenario, choose architecture/skeleton, map to native capabilities, define concrete operations, and establish validation rules.

• Default to async-first patterns for service code, streaming, and orchestration unless specific requirements strictly demand synchronous execution.

• Encourage the use of YAML-backed configurations for prompts and model settings to ensure project deployability and cleaner code boundaries.

• Treat this as the primary routing layer; once the architecture is defined, hand off to specialized leaf skills for specific implementations (e.g., knowledge-base, output control, or session memory).