Skillscript: A Declarative Language for Agent Workflows

Skillscript is a declarative domain-specific language designed for writing agent workflows, addressing pain points of using general-purpose languages (like Python) for this task. Key features include dependency DAGs, typed operations, connector mechanisms, and support for both interpretive execution and compilation into prompt products. Currently in v1 development stage, its public API, syntax, and connector contracts may change before v1.0.0 (no backward compatibility guaranteed).

With LLM capabilities advancing, agent-based automated workflows are increasingly common. However, using general languages like Python for orchestration brings challenges: mix of code and configuration, hard-to-visualize dependencies, cross-platform portability issues, and difficulty compiling workflows into model-consumable prompts. Skillscript was created to solve these problems by providing a declarative, domain-specific way to describe agent workflows, focusing on "what to do" rather than "how to do it."

Skillscript's core design includes:

1. Declarative Recipe: Treats workflows as recipes (declarative programs describing dependencies and data flow), supporting parallel execution and cache optimization.
2. Once Write, Multiple Executions: Can be interpreted (via interpreter or scheduled tasks) or compiled into prompts for agents, adapting to different environments.
3. Typed Operations & Connectors: Each operation has clear type signatures (input/output checked), and connectors enable interaction with external systems (filesystem, SQLite, Ollama local models, etc.)

Skillscript's technical stack includes:

• CLI Toolchain: skillfile CLI supports init (new project), run (execute scripts), compile (to prompts), lint (static check), list (available skills/ops).
• Default Connectors: Built-in connectors like SkillStore (filesystem-based), MemoryStore (SQLite memory storage), LocalModel (Ollama integration), MCP scaffold.
• Containerization: Multi-architecture images with docker-compose.yml preconfigured for Ollama/SQLite integration.
• Library Export: Can be embedded into apps via imports (e.g., import { compile, execute } from "skillscript-runtime")

Skillscript offers:

1. Lower Dev Threshold: Declarative syntax and type safety reduce cognitive load for building complex agent workflows, ideal for rapid prototyping.
2. Better Portability: Declarative definitions enable seamless migration across local dev, CI/CD, and production environments.
3. New Human-AI Collaboration: Compilation to prompts allows structured workflow definitions by devs, consumed by agents—leveraging strengths of both.

Limitations: Unstable API/syntax (pre-v1.0), immature ecosystem (limited connectors), small community. Future: As agent workflow demand grows, declarative languages like Skillscript will gain value. Its design (declarative, type-safe, compilable) aligns with the field's direction.

Skillscript is an experimental project redefining agent workflow development with declarative language. Though in early stages, its core idea (workflows as DAG-based typed programs) addresses key pain points in current agent orchestration. It's a project worth tracking for developers focused on AI workflow infrastructure.