LLM Wiki is a cross-platform desktop application that turns your documents into an organized, interlinked knowledge base — automatically. Instead of traditional RAG (retrieve-and-answer from scratch every time), the LLM incrementally builds and maintains a persistent wiki from your sources。
TypeScript
import { describe, it, expect } from "vitest"
import { extractJsonObject } from "./sweep-reviews"
describe("extractJsonObject", () => {
describe("bare JSON", () => {
it("extracts a simple object", () => {
expect(extractJsonObject('{"resolved":["a","b"]}')).toBe('{"resolved":["a","b"]}')
})
it("extracts an empty object", () => {
expect(extractJsonObject("{}")).toBe("{}")
})
it("extracts JSON with empty array", () => {
expect(extractJsonObject('{"resolved":[]}')).toBe('{"resolved":[]}')
})
it("preserves whitespace inside the object", () => {
const raw = '{\n "resolved": [\n "id-1"\n ]\n}'
expect(extractJsonObject(raw)).toBe(raw)
})
})
describe("markdown fences", () => {
it("strips ```json ... ``` multi-line fence", () => {
const raw = '```json\n{"resolved":["a"]}\n```'
expect(extractJsonObject(raw)).toBe('{"resolved":["a"]}')
})
it("strips bare ``` ... ``` multi-line fence", () => {
const raw = '```\n{"resolved":["a"]}\n```'
expect(extractJsonObject(raw)).toBe('{"resolved":["a"]}')
})
it("strips single-line ```json {...}``` fence", () => {
const raw = '```json {"resolved":["x"]}```'
expect(extractJsonObject(raw)).toBe('{"resolved":["x"]}')
})
it("handles fences with surrounding whitespace", () => {
const raw = ' \n ```json\n{"resolved":[]}\n``` \n '
expect(extractJsonObject(raw)).toBe('{"resolved":[]}')
})
it("is case-insensitive on the 'json' language tag", () => {
const raw = "```JSON\n{}\n```"
expect(extractJsonObject(raw)).toBe("{}")
})
})
describe("prose-wrapped JSON", () => {
it("finds JSON at the end of prose", () => {
const raw = 'Here is the answer: {"resolved":["a"]}'
expect(extractJsonObject(raw)).toBe('{"resolved":["a"]}')
})
it("returns the FIRST balanced object when prose has other braces before", () => {
// First balanced {...} is the prose one — expected behavior,
// callers then try JSON.parse and fall back on failure.
const raw = 'An example: {maybe like this}. Real answer: {"resolved":["a"]}'
const result = extractJsonObject(raw)
expect(result).toBe("{maybe like this}")
})
it("handles JSON with nested objects", () => {
const raw = '{"outer":{"inner":[1,2,3]}}'
expect(extractJsonObject(raw)).toBe(raw)
})
})
describe("string / escape handling", () => {
it("ignores braces inside string values", () => {
const raw = '{"note":"this { has } braces"}'
expect(extractJsonObject(raw)).toBe(raw)
})
it("handles escaped quotes inside strings", () => {
const raw = '{"q":"she said \\"hi\\""}'
expect(extractJsonObject(raw)).toBe(raw)
})
it("handles escaped backslash", () => {
const raw = '{"path":"C:\\\\foo"}'
expect(extractJsonObject(raw)).toBe(raw)
})
})
describe("malformed input", () => {
it("returns empty string for no JSON at all", () => {
expect(extractJsonObject("no json here")).toBe("")
})
it("returns empty string for empty input", () => {
expect(extractJsonObject("")).toBe("")
})
it("returns empty string for whitespace-only input", () => {
expect(extractJsonObject(" \n \t ")).toBe("")
})
it("returns empty string for unclosed object", () => {
expect(extractJsonObject('{"resolved":')).toBe("")
})
it("returns empty string when only opening brace", () => {
expect(extractJsonObject("{")).toBe("")
})
it("handles a fence with no inner JSON", () => {
expect(extractJsonObject("```json\n```")).toBe("")
})
})
describe("realistic LLM responses", () => {
it("parses the expected fenced output from our prompt", () => {
const raw = '```json\n{"resolved": ["review-1", "review-5"]}\n```'
const extracted = extractJsonObject(raw)
expect(JSON.parse(extracted)).toEqual({ resolved: ["review-1", "review-5"] })
})
it("parses a bare response with no fence", () => {
const raw = '{"resolved": []}'
expect(JSON.parse(extractJsonObject(raw))).toEqual({ resolved: [] })
})
it("survives a chatty preamble", () => {
const raw = 'I analyzed the reviews. Final answer:\n\n{"resolved": ["abc"]}'
expect(JSON.parse(extractJsonObject(raw))).toEqual({ resolved: ["abc"] })
})
})
})