Skip to main content
Lab Grimoire
TW EN
☕ Coffee
Spellbook / lich-skills ↗ Visit project
ai-tools

lich-skills

A 6-skill engineering toolkit that runs natively across Claude Code, Gemini CLI, and OpenAI Codex. Design principle: telegraph style, opinionated, zero filler. Includes debug-hypothesis (scientific debugging), spec-driven-dev (6-gate development), wiki-aggregate (multi-source research with citation tracking), and nano-banana (Gemini image generation). MIT licensed.

Best For

Developers who work across multiple AI coding agents (Claude Code + Gemini CLI, or Claude Code + Codex). Specifically:

  • Multi-agent users who need shared workflow skills across different agent platforms
  • Engineers who find AI debugging too intuition-dependent and want a more systematic methodology
  • Researchers who need to aggregate information across multiple sources with per-claim citation tracking
  • Anyone who wants Gemini image generation directly integrated into a Claude Code workflow

If you use only one agent platform, the cross-platform compatibility is not relevant to you. But individual skills in the package — particularly debug-hypothesis and wiki-aggregate — have standalone value even in single-platform environments.

How I Actually Use It

lich-skills has a different design philosophy from most skill packages: it forces you to do the right things. Every skill includes anti-rationalization tables — an explicit list of "the excuse you might use to skip this step, and why that excuse is wrong."

This design targets AI's number-one failure mode: stacking actions on top of an unverified assumption rather than verifying the assumption first.

The six skills each have a clear purpose:

  • debug-hypothesis: Scientific method debugging (observe, hypothesize, experiment, conclude). Enforces maximum 5-line changes per experiment and prohibits fix code before hypothesis verification.
  • spec-driven-dev: Spec, Plan, Build, Test, Review, Ship — six gates, each with an exit criteria checklist.
  • wiki-aggregate: Research aggregation from N sources with per-claim path:line citation tracking and cross-source contradiction detection.
  • tavily-search handles Tavily API web search integration.
  • nano-banana wraps Google Gemini Flash Image for 512–4K text-to-image, usable directly within Claude Code.
  • The same skill package ships with .claude-plugin/, gemini-extension.json, and Codex adapters simultaneously.

The cross-platform architecture is a serious technical commitment. Almost no other skill packages maintain synchronized versions across all three platforms.

image

Where It Is Strong

  • Cross-platform native compatibility: Claude Code, Gemini CLI, OpenAI Codex from one skill library, no separate versions needed
  • Every skill includes anti-rationalization tables that explicitly address common step-skipping excuses
  • The debug-hypothesis 5-line experiment limit forces atomic experiments, cutting down on "I changed a lot and don't know what fixed it"
  • wiki-aggregate citation tracking: Each claim traced to path:line, making research auditable instead of a black-box conclusion
  • nano-banana puts Gemini text-to-image directly in Claude Code without tool switching

Where It Fails

  • Tavily search requires a Tavily API key; nano-banana requires a Gemini API key. Each skill has different prerequisites.
  • If the GSD skill series (gsd-plan-phase, gsd-execute-phase) is already installed, spec-driven-dev is probably redundant.
  • Small community: Personal skill library with limited stars; troubleshooting means reading source code.
  • Maintenance rhythm depends on the individual author. Cross-skill update synchronization needs watching over time.

Pricing, Difficulty, and Risk

Free and MIT licensed. Intermediate difficulty — individual skill installation is low-barrier, but fully leveraging cross-platform functionality (Gemini CLI + Codex sync) requires multi-platform agent environment setup. Tavily/Gemini API keys are optional and do not affect core debug/spec/wiki functionality.

Verdict

The three core skills (debug-hypothesis, wiki-aggregate, nano-banana) justify installing the entire package. For multi-agent workflows, the cross-platform compatibility is real differentiated value.

If you do not use Gemini CLI or Codex, spec-driven-dev may overlap with existing GSD tools — but the other three do not. Recommend installing wiki-aggregate and debug-hypothesis first; the impact is immediate.

Source