Skip to content

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

fix(lint): drop missing-page findings for pages that already exist

Andrew Chen840f42af72e4
diff --git a/src/lib/lint.test.ts b/src/lib/lint.test.ts
index 8c6b03ff..f7512703 100644
--- a/src/lib/lint.test.ts
+++ b/src/lib/lint.test.ts
@@ -118,6 +118,48 @@ describe("runSemanticLint — language directive", () => {
})
})
+describe("runSemanticLint — missing-page false positives (#537)", () => {
+ it("drops missing-page findings when the entity page already exists", async () => {
+ const pages = [
+ makeFileNode("概念页.md", "介绍 [[维特根斯坦]]、[[阿德勒]] 和 [[尼采]] 的概念页"),
+ makeFileNode("维特根斯坦.md", "维特根斯坦是一位哲学家。"), // entity page EXISTS
+ makeFileNode("阿德勒.md", "阿德勒是一位心理学家。"), // entity page EXISTS
+ ]
+ mockListDirectory.mockResolvedValue(pages.map((p) => p.node))
+ mockReadFile.mockImplementation(async (path) => {
+ const match = pages.find((p) => p.node.path === path)
+ return match?.content ?? ""
+ })
+ // The LLM wrongly reports missing-page for entities that already have pages,
+ // using free-form Chinese titles that don't match a known "缺失页面:" prefix.
+ const llm = [
+ "---LINT: missing-page | warning | 维特根斯坦---", // bare name, page exists
+ "维特根斯坦在概念页中被引用但未建立独立实体页。",
+ "PAGES: 概念页.md",
+ "---END LINT---",
+ "---LINT: missing-page | warning | 阿德勒实体页缺失---", // name + suffix, page exists
+ "阿德勒被引用但缺少页面。",
+ "PAGES: 概念页.md",
+ "---END LINT---",
+ "---LINT: missing-page | warning | 尼采---", // genuinely missing (no page)
+ "尼采被引用但没有页面。",
+ "PAGES: 概念页.md",
+ "---END LINT---",
+ ].join("\n")
+ mockStreamChat.mockImplementation(async (_c, _m, cb) => {
+ cb.onToken(llm)
+ cb.onDone()
+ })
+
+ const results = await runSemanticLint("/project", fakeLlmConfig())
+
+ const missingPages = results.filter((r) => r.detail.includes("[missing-page]"))
+ // The two existing-entity false positives are filtered; only 尼采 remains.
+ expect(missingPages).toHaveLength(1)
+ expect(missingPages[0].page).toBe("尼采")
+ })
+})
+
describe("runSemanticLint — activity & early returns", () => {
it("logs a running activity item and marks done", async () => {
mockListDirectory.mockResolvedValue([makeFileNode("a.md", "content").node])
diff --git a/src/lib/lint.ts b/src/lib/lint.ts
index 12b69402..ee772a67 100644
--- a/src/lib/lint.ts
+++ b/src/lib/lint.ts
@@ -62,6 +62,35 @@ function normalizeLinkTarget(target: string): string {
.toLowerCase()
}
+/**
+ * Normalize a name for missing-page existence comparison. NFKC folds full-width
+ * and compatibility forms so CJK / full-width variants compare equal.
+ */
+function normalizeForExistence(s: string): string {
+ return s.normalize("NFKC").trim().toLowerCase()
+}
+
+/**
+ * Decide whether an LLM `missing-page` finding actually refers to a page that
+ * already exists. The LLM does not reliably cross-reference the file list, so it
+ * flags entities whose page is already present — especially in non-English wikis
+ * where its free-form titles ("维特根斯坦", "阿德勒实体页缺失", "缺失页面:X") don't
+ * match a fixed prefix. Drop the finding when a known page name matches exactly
+ * or appears within the finding title (guarded by length to avoid over-matching).
+ */
+function missingPageAlreadyExists(
+ llmTitle: string,
+ existingPageNames: Set<string>,
+): boolean {
+ const norm = normalizeForExistence(llmTitle)
+ if (!norm) return false
+ if (existingPageNames.has(norm)) return true
+ for (const name of existingPageNames) {
+ if (name.length >= 2 && norm.includes(name)) return true
+ }
+ return false
+}
+
function extractTitle(content: string, fallbackPath: string): string {
const frontmatter = content.match(/^---\s*\n([\s\S]*?)\n---/)
if (frontmatter) {
@@ -329,13 +358,20 @@ export async function runSemanticLint(
(f) => f.name !== "log.md"
)
- // Build a compact summary of each page (frontmatter + first 500 chars)
+ // Build a compact summary of each page (frontmatter + first 500 chars), and
+ // collect the set of existing page names (basename + frontmatter title) used
+ // to filter out `missing-page` findings for pages that already exist (#537).
const summaries: string[] = []
+ const existingPageNames = new Set<string>()
for (const f of wikiFiles) {
+ const basename = f.name.replace(/\.md$/i, "")
+ if (basename) existingPageNames.add(normalizeForExistence(basename))
try {
const content = await readFile(f.path)
const preview = content.slice(0, 500) + (content.length > 500 ? "..." : "")
const shortPath = getRelativePath(f.path, wikiRoot)
+ const title = extractTitle(content, shortPath)
+ if (title) existingPageNames.add(normalizeForExistence(title))
summaries.push(`### ${shortPath}\n${preview}`)
} catch {
// skip
@@ -412,8 +448,12 @@ export async function runSemanticLint(
const title = match[3].trim()
const body = match[4].trim()
- // semantic results always use type "semantic"
- void rawType
+ // Drop `missing-page` findings whose page already exists — the LLM often
+ // flags entities that already have a page, especially in non-English wikis
+ // where its free-form titles don't match a fixed prefix (#537).
+ if (rawType === "missing-page" && missingPageAlreadyExists(title, existingPageNames)) {
+ continue
+ }
const pagesMatch = body.match(/^PAGES:\s*(.+)$/m)
const affectedPages = pagesMatch