lich-skills tackles one of the biggest pain points in AI coding assistants: they often bulldoze through assumptions without evidence-based verification, leading to unreliable outcomes. This repo implements structured skill modules that embed engineering methodologies like spec-driven development and scientific-method debugging to address this.
What lich-skills offers and how it works
At its core, lich-skills is a Python-based collection of seven domain-specific skills designed for AI coding assistants like Claude Code, Gemini CLI, and OpenAI Codex. Each skill encapsulates a focused capability, ranging from enforcing a rigorous six-phase software development lifecycle (spec-driven-dev) to applying scientific debugging methods (debug-hypothesis) and agentic research aggregation with provenance tracking (wiki-aggregate).
Beyond these core engineering skills, there are utility modules including a web search skill (tavily-search), a text-to-image generation skill using Gemini (nano-banana), a production-grade UI design helper (frontend-design), and a subagent prompt compression skill (subagent-brief).
The repository takes advantage of native plugin or extension systems for the supported AI platforms, with configuration files like .claude-plugin/marketplace.json or gemini-extension.json defining installation and integration. Credentials for external services are managed exclusively through environment variables, never hardcoded, which enhances security and portability.
This modular architecture allows each skill to be independently maintained and updated, and the repo serves as a structured catalog of engineering-driven AI capabilities rather than a monolithic system.
Engineering rigor and thoughtful tradeoffs in skill design
What sets lich-skills apart is its commitment to embedding engineering methodologies directly into AI assistant skills. For example, the spec-driven-dev skill enforces a six-phase software development lifecycle with explicit exit criteria at each phase. This is not just a checklist but a mechanism to prevent premature assumptions and bulldozing through incomplete specs.
Similarly, the debug-hypothesis skill applies a scientific method to debugging, including anti-bulldozer rules that discourage jumping to conclusions without evidence. This is a direct response to the #1 failure mode observed in AI coding agents — making changes without sufficient verification.
The wiki-aggregate skill performs agentic research aggregation with detailed path:line provenance tracking, which is crucial for traceability in complex AI workflows.
These skills represent a tradeoff: by imposing structured workflows and exit criteria, they may slow down some exploratory or freestyle coding sessions but improve the reliability and reproducibility in production or critical environments.
The code quality across the repo is surprisingly clean and modular, making it easier to extend or adapt individual skills. Relying on environment variables for credentials means there is no risk of leaking secrets in code, but it requires users to manage their environment carefully.
Overall, lich-skills prioritizes engineering discipline and transparency over convenience or rapid prototyping.
Quick start with Claude Code
If you’re using Claude Code, installing lich-skills is straightforward via the plugin marketplace inside a running session:
/plugin marketplace add LichAmnesia/lich-skills
/plugin install lich-skills@lich-skills
Once installed, all seven skills become immediately available. You can verify installation with:
/skills
Alternatively, you can clone the repo directly into tool-specific directories if you prefer manual installation.
This quick start leverages Claude Code’s native plugin system, making it easy to integrate without complex setup.
Verdict
lich-skills is well-suited for developers and teams integrating AI coding assistants who want to enforce engineering rigor and reduce error-prone assumptions. Its focus on spec-driven development and scientific debugging offers a practical way to improve AI agent reliability in real-world coding workflows.
That said, it is currently tied to AI assistants like Claude Code and Gemini CLI, limiting its applicability if you use other platforms. The structured workflows might feel restrictive for rapid prototyping or casual use.
For anyone building production-grade AI coding assistants or aiming for reproducible, evidence-driven AI development cycles, lich-skills is worth exploring. Its modular design and clean codebase also make it a solid starting point for extending AI assistant capabilities with engineering best practices in mind.
If you’re curious how AI agents can be made more trustworthy and methodical, lich-skills offers concrete techniques worth understanding and applying.
Related Articles
- how awesome-claude-skills turns claude into a real-world action agent — Awesome Claude Skills is a modular Python framework that empowers Claude to perform real-world actions by integrating wi
- openai/skills: modular agent skills for reusable AI capabilities — The openai/skills repo offers a catalog of modular ‘Agent Skills’ for OpenAI Codex agents, enabling reusable AI function
- OpenAI Codex CLI: local-first AI coding assistant with ChatGPT integration — OpenAI Codex CLI brings AI coding assistance local to your terminal, integrating with ChatGPT plans for powerful hybrid
- Building a production-ready second brain with agentic RAG and LLMOps — Explore an open-source course that teaches building a production-grade AI assistant using advanced retrieval-augmented g
→ GitHub Repo: LichAmnesia/lich-skills ⭐ 213 · Python