# Allium

> Allium behavioral specs integrated with /core:agent-loop. Use when attaching a formal spec to an epic, propagating tests from a spec before TDD, or weed-checking spec/code divergence after CI passes.

- **Type:** Skill
- **Install:** `agentstack add skill-vinnie357-claude-skills-allium`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [vinnie357](https://agentstack.voostack.com/s/vinnie357)
- **Installs:** 0
- **Category:** [Developer Tools](https://agentstack.voostack.com/c/developer-tools)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [vinnie357](https://github.com/vinnie357)
- **Source:** https://github.com/vinnie357/claude-skills/tree/main/plugins/tools/allium/skills/allium

## Install

```sh
agentstack add skill-vinnie357-claude-skills-allium
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Allium

Opinionated integration between [juxt/allium](https://github.com/juxt/allium) and the `/core:agent-loop` workflow. Allium captures observable behavior in `.allium` files using `entity`, `rule`, and `config` blocks — implementation-agnostic, co-located with code.

## When to Use

- Epic author is attaching a formal behavioral spec to a new epic
- Worker needs to seed failing TDD tests from a spec before implementing
- Validator has passed CI and needs to confirm the code matches the spec semantically
- Refactor epic requires a behavioral baseline before decomposition

## Prerequisites

Upstream Allium Claude plugin must be installed separately:

```
/plugin install allium@juxt
```

The CLI (`allium-tools` binary) installs via mise — see `references/installation.md`.

## Spec Location Convention

Recommended convention for projects adopting this skill: specs live at `docs/specs/.allium` with shared specs in `docs/specs/shared/`. A project-root `allium.config.json` declares `specPaths: ["docs/specs"]` so the CLI and LSP resolve them. Copy `templates/allium.config.json` into the project root to bootstrap this layout.

## Integration Points by Tier

See `references/agent-loop-integration.md` for the complete 6-tier breakdown. Summary:

| Tier | Allium action |
|---|---|
| Epic Author | Run `/allium:elicit` or copy a template from `templates/`; set `spec:` field in epic markdown |
| Team Leader | If refactor label and no spec: run `/allium:distill` for baseline |
| Worker | If spec attached: run `/allium:propagate` to seed failing tests BEFORE implementing |
| Validator | After CI passes: run `/allium:weed`; divergences route to fix-agent same as CI failures |

## Spec Syntax (v3)

```allium
entity Order {
    status: pending | confirmed | shipped | delivered | cancelled
    tracking_number: String when status = shipped | delivered

    transitions status {
        pending -> confirmed
        confirmed -> shipped
        terminal: delivered, cancelled
    }

    invariant NonNegativeTotal { this.total >= 0 }
}

rule ShipOrder {
    when: ShipOrder(order, tracking)
    requires: order.status = confirmed
    ensures:
        order.status = shipped
        order.tracking_number = tracking
}
```

Key constructs:
- `entity` — domain object with typed fields, conditional attributes (`when`), state transitions, and invariants
- `rule` — behavioral rule with `when` (trigger), `requires` (preconditions), `ensures` (postconditions)
- `config` — project-level settings (thresholds, retry counts, timeout values)
- `use` — import another spec: `use "./shared/oauth.allium" as oauth`

## Templates

Copy-pasteable specs in `templates/`:

- `oauth-pkce-flow.allium` — OAuth2 PKCE flow matching the agent-loop epic-authoring example
- `circuit-breaker.allium` — infrastructure contract with `entity` + `config` + `rule`
- `runex-workflow.allium` — spec for a Runex workflow step
- `epic-with-spec.md` — VantageEx epic showing the `spec:` field in use

See `references/epic-spec-template.md` for a guided walkthrough.

## Anti-Fabrication

All claims about spec content must derive from reading actual `.allium` files with Read tool. Never assert that a spec covers a behavior without verifying the `ensures` or `invariant` blocks. See `/core:anti-fabrication`.

## References

- `references/installation.md` — mise backend details and upstream plugin install
- `references/agent-loop-integration.md` — 6-tier integration procedure
- `references/epic-spec-template.md` — template walkthrough
- `references/distilling-legacy-code.md` — `/allium:distill` procedure for existing modules

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [vinnie357](https://github.com/vinnie357)
- **Source:** [vinnie357/claude-skills](https://github.com/vinnie357/claude-skills)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-vinnie357-claude-skills-allium
- Seller: https://agentstack.voostack.com/s/vinnie357
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.