diff --git a/.github/actions/spell-check/expect.txt b/.github/actions/spell-check/expect.txt
index ef086ec29e..b3680616b9 100644
--- a/.github/actions/spell-check/expect.txt
+++ b/.github/actions/spell-check/expect.txt
@@ -94,7 +94,7 @@ ASSOCSTR
 ASYNCWINDOWPLACEMENT
 ASYNCWINDOWPOS
 atl
-atx
+ATX
 ATRIOX
 aumid
 Authenticode
diff --git a/.github/prompts/fix-spelling.prompt.md b/.github/prompts/fix-spelling.prompt.md
new file mode 100644
index 0000000000..9364474b87
--- /dev/null
+++ b/.github/prompts/fix-spelling.prompt.md
@@ -0,0 +1,18 @@
+---
+mode: 'agent'
+model: GPT-5-Codex (Preview)
+description: 'Resolve Code scanning / check-spelling comments on the active PR.'
+---
+
+**Goal:** Clear every outstanding GitHub pull request comment created by the `Code scanning / check-spelling` workflow by explicitly allowing intentional terms.
+
+**Guardrails:**
+- Update only discussion threads authored by `github-actions` or `github-actions[bot]` that mention `Code scanning results / check-spelling`.
+- Resolve findings solely by editing `.github/actions/spell-check/expect.txt`; reuse existing entries.
+- Leave all other files and topics untouched.
+
+**Workflow:**
+1. Determine the active pull request with a single `gh pr view --json number` call (default to the current branch).
+2. Fetch all PR discussion data once via `gh pr view --json comments,reviews` and filter to check-spelling comments authored by `github-actions` or `github-actions[bot]` that are not minimized; when several remain, process only the most recent comment body.
+3. For each flagged token, review `.github/actions/spell-check/expect.txt` for an equivalent term (for example an existing lowercase variant); when found, reuse that normalized term rather than adding a new entry, even if the flagged token differs only by casing. Only add a new entry after confirming no equivalent already exists.
+4. Add any remaining missing token to `.github/actions/spell-check/expect.txt`, keeping surrounding formatting intact.
\ No newline at end of file