How to use the Wiki Enrich agent skill (wiki-enrich)

What the wiki-enrich skill does

Literature-ingest skills in the ARIS toolkit create one page per paper under research-wiki/papers/. Each page has frontmatter, the abstract, and ten fillable sections marked as TODO. Nothing in the toolkit fills those ten sections, so the wiki stays as scaffolds until someone reads every paper.

wiki-enrich reads each scaffolded page and fetches the paper's content through a fallback chain of sources. It then rewrites the empty sections, such as the one-line thesis, method, key results, and limitations, as 1-3 sentence prose summaries.

• Fills: One-line thesis, Problem / Gap, Method, Key Results, Assumptions, Limitations / Failure Modes, Reusable Ingredients, Open Questions, Claims, and Relevance to This Project.
• Never edits: the Connections section (generated from the graph edges file) and the Abstract (original) section (source data from arXiv).

Install the wiki-enrich skill

Download the ZIP for the skill from the directory page, then choose one of the install locations below. A personal install makes the skill available across projects. A project install limits it to one repository.

If you use the Claude apps, you can instead upload the skill under Settings, then Capabilities, then Skills.

• Personal install: unzip the download into ~/.claude/skills/.
• Project install: unzip the download into .claude/skills/ inside your project.
• Claude apps: go to Settings, then Capabilities, then Skills, and upload the skill.

Set up a research-wiki before you run it

The skill only works on an existing research wiki. It hard-fails if research-wiki/ is missing and will not run blind. Create the wiki first with the research-wiki init command.

The skill also needs the research_wiki.py helper script. It looks for this script in the ARIS repository, in a tools/ folder in the current project, or in ~/.codex/skills/research-wiki. It resolves the repository root with git and falls back to the current directory if git is not available. If it cannot find the helper script, it stops with an error.

It also needs network access to the alphaxiv, deepxiv, and arXiv APIs. The skill file does not state any API keys.

• A research-wiki/ directory created by /research-wiki init
• research_wiki.py from the ARIS repository or ~/.codex/skills/research-wiki
• git, used to find the repository root
• Network access to alphaxiv, deepxiv, and arXiv

How to trigger the skill with natural requests

You can call the skill directly with a slash command, or make a natural request. The skill's description lists phrases that should activate it, including 'enrich wiki', 'fill paper TODOs', 'research-wiki 自動填', and 'wiki body 補完'. It also activates after a batch ingest that left papers/ as TODO scaffolds.

By default it only processes papers that still have at least one TODO section, up to 20 papers per run. You can pass a target to work on one paper by slug or on all papers, and you can use flags to choose the source and limit the batch size.

• Plain request: 'enrich wiki' or 'fill paper TODOs' (processes papers with open TODOs)
• One paper: /wiki-enrich vllm
• Rewrite everything: /wiki-enrich all --force
• Choose a source and limit: /wiki-enrich --source alphaxiv --max 5
• Larger batch: /wiki-enrich missing --max 50

How the skill fetches paper content and what it protects

With the default auto source, the skill tries sources in this order and uses the first non-empty result: alphaxiv overview, alphaxiv abstract, deepxiv brief, the arXiv API abstract, and finally the abstract already on the page. You can pin one source with the --source flag.

Without --force, the skill skips sections that already have real content. With --force, it overwrites the fillable sections but still leaves Connections and Abstract (original) untouched. If no 'claim' edges point to a paper, the Claims section is filled with 'No claims tracked yet'. For Relevance to This Project, it reads RESEARCH_BRIEF.md, AGENTS.md or CLAUDE.md, and gap_map.md. If none of these exist, the section stays as TODO and the skill reports it.

• Default source order: alphaxiv overview, alphaxiv abstract, deepxiv brief, arXiv API abstract, page abstract.
• Default batch cap: 20 papers per run.
• Without --force, existing content is kept.

Safety notes: scripts and network access

This skill runs scripts and makes network requests. It fetches content from external paper hosts, so anyone who does not want outbound requests should not use it. It also writes to your wiki files, so keep the project under version control so you can review and revert changes.

The summaries are generated from fetched abstracts and other source text. They can overstate or miss details, so check them before you use them in anything that will be published.

• It runs shell commands and reads and writes files in research-wiki/.
• It sends requests to alphaxiv, deepxiv, and arXiv.
• Review the generated text against the original paper before publishing.