diff --git a/DESCRIPTION b/DESCRIPTION
index 956ff3c..e9923fa 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,8 +1,8 @@
 Package: pensar
 Type: Package
 Title: LLM Wiki Engine
-Version: 0.6.3.3
-Date: 2026-06-03
+Version: 0.6.3.4
+Date: 2026-06-06
 Authors@R: c(
     person("Troy", "Hernandez", role = c("aut", "cre"),
         email = "troy@cornball.ai",
diff --git a/NEWS.md b/NEWS.md
index 2932898..aafdc52 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -1,3 +1,13 @@
+# pensar 0.6.3.4 (dev)
+
+## Changes
+
+* Wikilink parsing now ignores Markdown code. Fenced blocks (` ``` ` or
+  `~~~`) and inline spans (`` `...` ``) are stripped before extracting
+  `[[...]]`, so R's `[[ ]]` list indexing in a code sample no longer
+  registers as a broken wikilink. Affects `lint()`, `outlinks()`,
+  `backlinks()`, and `vault_graph()`.
+
 # pensar 0.6.3.3 (dev)
 
 ## Bug fixes
diff --git a/R/parse.R b/R/parse.R
index ea70f0d..6322074 100644
--- a/R/parse.R
+++ b/R/parse.R
@@ -20,13 +20,33 @@ parse_frontmatter <- function(filepath) {
     tryCatch(yaml::yaml.load(yaml_text), error = function(e) list())
 }
 
+#' Blank out code regions so they are not scanned for wikilinks
+#'
+#' Markdown code is literal text, not wiki markup. R's \code{[[ ]]} list
+#' indexing inside a code block would otherwise be misread as a wikilink.
+#' This drops fenced code blocks (\code{```} or \code{~~~}) entirely and
+#' strips inline code spans (\code{`...`}) from the remaining lines.
+#'
+#' An unterminated fence swallows everything to end of file, which is the
+#' safe choice: better to miss a link in malformed markup than to invent one.
+#' @noRd
+strip_code <- function(lines) {
+    fence <- grepl("^\\s*(`{3,}|~{3,})", lines)
+    # cumsum is odd on the opening fence and its contents; the closing
+    # fence lands on an even count, so include it via `| fence`.
+    inside <- (cumsum(fence) %% 2L == 1L) | fence
+    gsub("`+[^`]*`+", " ", lines[!inside])
+}
+
 #' Parse all wikilinks from a markdown file
 #'
+#' Code regions are ignored via \code{strip_code()}.
+#'
 #' @param filepath Path to a markdown file.
 #' @return Character vector of link targets.
 #' @noRd
 parse_wikilinks <- function(filepath) {
-    lines <- readLines(filepath, warn = FALSE)
+    lines <- strip_code(readLines(filepath, warn = FALSE))
     all_links <- regmatches(lines, gregexpr("\\[\\[([^]]+)\\]\\]", lines))
     all_links <- unlist(all_links)
     if (length(all_links) == 0L) {
diff --git a/inst/tinytest/test_parse.R b/inst/tinytest/test_parse.R
index 1ba3253..f32c080 100644
--- a/inst/tinytest/test_parse.R
+++ b/inst/tinytest/test_parse.R
@@ -52,6 +52,26 @@ parsed <- pensar:::parse_wikilink("[[Delta|display text]]")
 expect_equal(parsed$target, "Delta")
 expect_equal(parsed$label, "display text")
 
+# Code regions are not scanned: R's [[ ]] indexing is not a wikilink.
+tmp5 <- tempfile(fileext = ".md")
+writeLines(c(
+  "Real link [[Alpha]] in prose.",
+  "Inline code `merged[[name]] <- project[[name]]` is not a link.",
+  "```r",
+  "x <- registry[[uuid]]",
+  "y <- [[Beta]]",
+  "```",
+  "After the fence [[Gamma]] counts again."
+), tmp5)
+
+wl2 <- pensar:::parse_wikilinks(tmp5)
+expect_true("Alpha" %in% wl2)
+expect_true("Gamma" %in% wl2)
+expect_false("name" %in% wl2)
+expect_false("uuid" %in% wl2)
+expect_false("Beta" %in% wl2)
+unlink(tmp5)
+
 # --- name_from_path ---
 expect_equal(pensar:::name_from_path("/vault/Neural Networks.md"), "Neural Networks")