Zissa Wiki workflow
A markdown research wiki template: the agent ingests sources, updates pages and checks every quote word for word against the file.
- 62 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The template works with local markdown files and the chosen model reads the sources.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/zissa-wikiDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The template works with local markdown files and the chosen model reads the sources.
This is third-party code. Review the repository files before installing.
What it does
Zissa Wiki implements Andrej Karpathy's LLM Wiki pattern: instead of searching raw documents again each time, the model builds and maintains an interlinked wiki that grows with every source. The user drops an article into `raw/` and tells the agent to ingest. The agent reads the source, writes a note and updates the affected concept, author and debate pages. Every direct quote is then checked word for word against the source file by `scripts/wiki.py`, which never calls a model: it checks links, quotes and structure. Claude Code has `/ingest`, `/query` and `/lint` commands, the schema sits in `AGENTS.md`, so Codex, Gemini CLI, Cursor and OpenCode fit too. Chat apps get the `wiki.py prompt` and `apply` commands.
Who it is for. Researchers and students who keep a knowledge base from papers and books and want the agent to maintain it without inventing quotes.
Good fit when
- A new paper should update a dozen linked wiki pages.
- Before publication you must confirm quotes match the original.
- An answer to a research question is sought across accumulated pages.
- The wiki is kept in ChatGPT or another chat with no file access.
Not a fit when
- A ready wiki with content is wanted: the repository gives an empty skeleton.
- Sources must not be shown to a model.
- Vector database search is required: the template deliberately avoids it.
Example request
Ingest raw/articles/smith-2020.pdf following the wiki rules, update the linked pages and check that all quotes match the file.Limitations
The repository gives a skeleton and the user fills the wiki, and the author advises supervising the first five to ten ingests by hand. `wiki.py` catches broken links and inexact quotes but cannot find semantic errors in the model's retelling. The author tested end to end with Claude Code, Codex CLI and, through chat, a Mistral model. Windows and other models need manual checks. Sources in `raw/` reach the chosen model in full.
How to disable. Delete the wiki folder along with `.claude/skills/ingest`, `query` and `lint`. Turn off the commit hook with `git config --unset core.hooksPath`.
FAQ
How do I start?
Clone the repository, open the folder in the agent, fill Domain Context in `AGENTS.md` and add a first note in `raw/notes/`.
Which commands exist in Claude Code?
`/ingest`, `/query` and `/lint`. In other agents say the same in words, for example ingest raw/….
How are quotes checked?
The `wiki.py` script compares every direct quote word for word with the source file and reports mismatches.
Can the wiki be kept in a chat?
Yes. `wiki.py prompt` bundles a message for the chat and `wiki.py apply` writes the model reply back into the wiki.
Is a database needed?
No, the wiki is plain markdown files and needs no embeddings or plugins.
Discussion
No questions or reviews yet.
Related
WorkflowsA Claude Code template that builds a knowledge base from articles and PDFs: five newsroom roles and contradiction checks.
LLM Wiki Manager keeps a personal Markdown wiki through an agent: it ingests sources, links pages and answers with citations
An Obsidian vault template where an agent follows a CLAUDE.md schema to compile sources into a linked wiki and extend it.
CLI for a personal Markdown knowledge base: the agent writes, searches and lints pages with wiki commands; the tool calls no model
A workflow for Claude Code, Codex and Gemini CLI where the agent files source documents into a persistent, linked wiki
Turns session history of Claude Code, Codex, Cursor and Gemini CLI into a searchable wiki on Karpathy's LLM Wiki pattern, locally
README in short
The README covers quick start, an agent table, three workflows (ingest, query, lint), quote checks, wiki graph viewing, folder structure, page types and scaling advice.