juxt/allium
5 skills6.7k installs1.9k starsGitHub
Install
npx skills add https://github.com/juxt/alliumSkills in this repo
1Elicitelicit is an agent skill from the Allium ecosystem that helps solo builders and small teams validate process and rule specs before implementation drift becomes expensive. It packages assumption-checking rituals: reflecting what you heard in plain language, questioning whether steps must be sequential, walking scenario traces when rules exist, and verifying actors across entity lifecycles. It is most often shelved under Validate scope work because that is where wrong ordering or missing transitions hurt timelines, but the same techniques apply while refining Idea research outputs or Build PM artifacts. The skill expects you to already have a coarse or complete Allium-oriented spec—not to replace discovery—and pairs well with iterative correction loops when stakeholders say “you’re missing X” or “Y happens before Z.” Use it whenever mental models and formal graphs might diverge.1.3kinstalls2Distilldistill is a JUXT Allium agent skill that teaches solo builders how to pull a formal specification out of real code. Instead of guessing requirements from scattered handlers and models, you follow worked examples—starting with a Python/Flask password reset implementation covering users, hashed passwords, lockout, and reset tokens—and learn what to capture in Allium. That makes the skill valuable during validate when you are proving what the product actually does, and during build docs when you need maintainable spec artifacts for agents or teammates. It is methodology-heavy documentation skill content, not a deploy or test runner. Use it when you inherit a feature branch and need a spec before rewriting auth, billing, or similar flows.1.3kinstalls3Propagatepropagate is an agent skill for solo builders who treat specifications as the source of truth and want tests that actually cover every obligation. It reads Allium `.allium` specs and uses `allium plan` and `allium model` (or the reference taxonomy) to enumerate required tests, then writes them in your repo's frameworks and style— including surface boundary tests and deeper property or state-machine cases. Use after you have a spec and implementation and need confidence that nothing in the spec is untested, not when you are still brainstorming features without written behavior. The skill emphasizes deterministic completeness from the parent spec while the agent handles naming, file layout, and correlating entities to code. Ideal for API-first SaaS and contract-heavy services validated before release and maintained in Ship.1.3kinstalls4TendTend is a senior, opinionated agent skill for maintaining Juxt Allium gardens: the `.allium` files that describe how a system should behave. Solo builders and small teams invoke it when they need to add entities, wire triggers and surfaces, fix validation errors, rename domains safely, or migrate specs to a newer language version. The workflow starts from the Allium language reference, discovers relevant spec files, optionally validates syntax with `allium check`, and only then proposes precise edits. Unlike freeform coding, tend optimizes for specification integrity—translating requirements into well-formed Allium rather than improvising implementation details. It is built for agentic workflows where Claude Code, Cursor, or Codex act as a spec editor with guardrails. Use it whenever behavior is meant to live in Allium, not scattered across comments and one-off scripts.1.3kinstalls5Weedweed is an agent skill for Allium projects that treats specification files and code as two gardens that must stay in sync. On startup it loads the Allium language reference, finds `.allium` specs, optionally validates them with the `allium` CLI, and reads the matching implementation. In check mode it reports divergences only; in update spec it rewrites `.allium` to describe current behavior; in update code it changes implementation to honor the spec. Solo builders maintaining spec-driven backends or libraries use it when specs have rotted after fast iteration, before a release audit, or when onboarding needs a trustworthy contract. It pairs naturally with build-time spec authoring and ship-time quality gates. Not a test runner or generic linter—it is Allium-specific alignment work. Intermediate complexity assumes you already commit `.allium` beside code.1.3kinstalls