Spectask: Specification-First Methodology for AI-Assisted Development and Spawn Extension

Spectask is a set of AI-assisted development methodologies that enforces the 'specification before implementation' principle. It provides structured task workflows, design indexes, and Agent skills via Spawn extensions. Its core goal is to solve the problem that AI programming assistants tend to deviate from requirements when generating code, leading to high rework costs. It ensures that AI completes specification formulation and undergoes human review before coding to reduce directional deviations.

With the popularization of AI programming assistants like Claude Code and Cursor, developers have found that AI generates code quickly but often 'goes off track'—functions do not meet expectations or miss details, and rework costs exceed those of handwritten code. The root cause is that traditional AI interactions lack a structured planning phase: humans design and think before coding, while AI directly jumps into implementation. Spectask was born to solve this problem, establishing a complete AI-assisted development workflow based on the 'specification before implementation' principle.

1. Specifications as a Shared Source of Truth: Before coding, AI needs to write a detailed specification document, including a one-sentence goal, before-and-after comparison description, affected modules, and clarification of ambiguous points, serving as a shared basis for human-AI collaboration;
2. Two-Stage Human Review: After the specification is completed, it undergoes AI self-review (independent sub-Agent reviews architecture impact and correctness) before being submitted for human review. Similarly, after code implementation, it undergoes AI self-review before human review, requiring an explicit approval signal;
3. Explicit Task Decomposition: Complex tasks need to have an execution plan, split into sequential/parallel phases, each step handled by an independent sub-Agent, and one-time full modifications are prohibited.

Spectask is distributed as a Spawn extension and can be integrated into environments supporting the Spawn toolchain:

• Installation: Install the Spawn CLI via the uv toolchain, then add the Spectask extension after initializing the project (commands are in the input);
• Directory Structure: The spec/ directory contains main.md (process rules), design.yaml (architecture index), design/ (architecture documents), tasks/ (task specifications), and seeds/ (initial ideas);
• File Types: Static files (provided by the extension, overwritten on update) and product files (project-specific, retained).

There are 7 steps in total: 1-2. Specification Drafting and Self-Review: AI drafts the specification (including elements like goals, proactively clarifies ambiguous points), and an independent sub-Agent conducts architecture review; 3. Human Specification Review: Submit the specification for human approval; 4-5. Implementation and Code Self-Review: Implement according to the execution plan (steps handled by sub-Agents), and an independent sub-Agent reviews code style and consistency with the specification; 6. Human Code Review: Submit the implementation for final review; 7. Document Update and Archiving: Update architecture documents and archive the task folder.

• Skill System: Predefined skills like spectask-create (draft specification), spectask-execute (implementation and code self-review), etc., encapsulate workflow steps;
• Seeds Concept: The spec/seeds/ directory records informal ideas, which are converted into formal tasks when mature;
• Living Documentation: Architecture documents are included in version control and updated with tasks (e.g., hla.md) to avoid outdated documentation.

• Complex Feature Development: Reduces rework rate; two-stage review ensures alignment of direction;
• Team Collaboration: Unifies process standards; new members quickly understand the project via the spec/ directory;
• Long-Term Maintenance: Living Documentation retains architectural knowledge and forms a traceable evolution history.

Spectask represents the evolution direction of AI-assisted development from 'code generation' to 'process management'. It solves the core challenges of AI programming: ensuring code meets expectations, is maintainable, and coordinates with the architecture. Through methods like specification before implementation and two-stage review, it provides an engineering framework for AI-assisted development, which is worth evaluating by teams exploring AI tools.