AutoSci Profile

AutoSci is a built-in profile template for research projects. It is useful both as a ready-to-use research workspace and as a concrete example of how profiles compose entities, relations, lifecycles, artifacts, workflows, connector bindings, and trust rules without profile-specific code. AutoSci is the largest built-in example of Configurable Lifecycle Profiles: the domain behavior comes from profile data, while the core engine stays generic.

Install AutoSci

llmwiki template inspect autosci
llmwiki template init autosci
llmwiki profile validate

template init autosci writes .llmwiki/profile.json after confirming the typed corpus is empty. The installed profile has profileId: autosci.

What AutoSci declares

Capability	AutoSci
Entities	12
Relations	12
Workflows	5
Workflow actions	5
Artifact types	7
Connector bindings	`crossref`
Content tiers	yes
Relation preconditions	yes
Artifact preconditions	yes

Entity vocabulary

AutoSci’s entities model a research workflow:

Entity type	Directory	Role
`papers`	`wiki/papers`	Imported and distilled literature
`sources`	`wiki/sources`	Repositories, videos, web pages, and paper sources
`ideas`	`wiki/ideas`	Proposed, explored, validated, or rejected hypotheses
`experiments`	`wiki/experiments`	Designs, running experiments, and completed results
`manuscripts`	`wiki/manuscripts`	Drafting, citation-checking, and submission
`topics`	`wiki/topics`	Research areas and gaps
`research-concepts`	`wiki/research-concepts`	Domain concepts extracted from literature
`methods`	`wiki/methods`	Proposed, validated, or deprecated methods
`foundations`	`wiki/foundations`	Datasets, tools, theories, and benchmarks
`people`	`wiki/people`	Researchers and collaborators
`reviews`	`wiki/reviews`	Peer review and internal review records
`research-outputs`	`wiki/research-outputs`	Released models, datasets, reports, and code

Relations

AutoSci relations are typed and endpoint-checked:

{
  "tests": {
    "from": ["experiments"],
    "to": ["ideas"],
    "direction": "directed"
  },
  "cites": {
    "from": ["papers", "manuscripts"],
    "to": ["sources", "papers"],
    "direction": "directed"
  },
  "builds-on": {
    "from": ["ideas"],
    "to": ["papers", "ideas"],
    "direction": "directed"
  }
}

The full profile also includes challenges, introduces-concept, uses-concept, proposes-method, extends-method, supports, contradicts, derived-from, and addresses-gap.

Lifecycle gates

AutoSci uses lifecycle fields to keep state transitions explicit. For example, an experiment starts as designed, moves to running, and can enter complete only when it has a resultSummary, a live tests relation to an idea, and a healthy experiment-result artifact ref.

---
title: Sparse attention ablation
hypothesis: Sparse attention improves latency.
stage: complete
resultSummary: Sparse attention improved p95 latency.
result: experiment-result/sparse-attention@sha256:4b7...
---

That frontmatter is accepted only if the referenced artifact exists and matches the pinned hash, and the relation graph contains the required tests edge.

Artifacts

AutoSci declares seven artifact types:

Artifact type	File name	Content kind
`experiment-result`	`result.json`	JSON
`paper-source-metadata`	`source-metadata.json`	JSON
`experiment-plan`	`plan.json`	JSON
`run-log`	`run.txt`	text
`manuscript-draft`	`draft.md`	text
`review-packet`	`packet.json`	JSON
`rebuttal-response`	`rebuttal.md`	text

Artifacts are written with llmwiki artifact write or as workflow artifact outputs. Pages reference them with artifactRef fields.

Workflows

AutoSci ships five workflows:

Workflow	Purpose
`literature-review`	Gather papers, extract concepts, and synthesize topics
`research`	Move from paper import through idea and experiment completion
`manuscript-writing`	Draft, cite, check, and submit a manuscript
`experiment-design`	Propose methods and design experiments
`review-response`	Intake reviews and draft responses

Each workflow is declared in the profile. The core workflow engine has no AutoSci-specific branches.

Crossref binding

AutoSci binds the first-party crossref connector to papers:

{
  "crossref": {
    "entityType": "papers",
    "fields": {
      "title": "title",
      "doi": "doi",
      "year": "year",
      "authors": "authors",
      "stage": "stage"
    },
    "contentField": "abstract"
  }
}

Running Crossref stages a typed papers review candidate. Approval requires the operator-supplied draft content hash printed by review show.

Why AutoSci is a configuration example

AutoSci proves the profile model by expressing a rich research workflow as data:

entity directories and field contracts;
typed relations;
finite-state lifecycles;
relation and artifact preconditions;
declared workflow stages and actions;
artifact contracts;
connector bindings;
content tiers.

The default profile, AutoSci, and Newsroom all use the same core engine. Profile behavior comes from .llmwiki/profile.json, not from domain-specific branches.

Get Started

Core Concepts

CLI Reference

Configuration

Guides

Troubleshooting

AutoSci Profile

Install AutoSci

What AutoSci declares

Entity vocabulary

Relations

Lifecycle gates

Artifacts

Workflows

Crossref binding

Why AutoSci is a configuration example

​Install AutoSci

​What AutoSci declares

​Entity vocabulary

​Relations

​Lifecycle gates

​Artifacts

​Workflows

​Crossref binding

​Why AutoSci is a configuration example

Install AutoSci

What AutoSci declares

Entity vocabulary

Relations

Lifecycle gates

Artifacts

Workflows

Crossref binding

Why AutoSci is a configuration example