diff --git a/.changeset/tangy-waves-shout.md b/.changeset/tangy-waves-shout.md
new file mode 100644
index 000000000..dd4d51683
--- /dev/null
+++ b/.changeset/tangy-waves-shout.md
@@ -0,0 +1,13 @@
+---
+"eve": patch
+---
+
+fix(slack): route HITL input-request prompts through GFM → mrkdwn conversion (#1293)
+
+Model-authored prompts in Slack HITL input-request widgets were being placed into `{ type: "mrkdwn" }` blocks verbatim — `**bold**`, `__bold__`, `~~strike~~`, and `[label](url)` rendered as literal punctuation instead of formatting. The asymmetry with `thread.post(string)` (which converts via eve's `gfmToSlackMrkdwn`) made every eve+Slack app rediscover the same workaround.
+
+`renderInputRequestBlocks` and `buildFreeformModalView` now both pass the prompt through `gfmToSlackMrkdwn` before truncating, matching the behavior of the rest of the Slack surface. Code fences and inline code are preserved untouched by the fence-aware splitter already used elsewhere.
+
+Includes regression coverage for both the block-renderer path and the modal path.
+
+Closes #1293
diff --git a/packages/eve/src/public/channels/slack/hitl.test.ts b/packages/eve/src/public/channels/slack/hitl.test.ts
index 6ff2db34a..9237e56a9 100644
--- a/packages/eve/src/public/channels/slack/hitl.test.ts
+++ b/packages/eve/src/public/channels/slack/hitl.test.ts
@@ -360,6 +360,25 @@ describe("renderInputRequestBlocks", () => {
       optionId: "weekly_report",
     });
   });
+
+  it("converts model-authored GFM in the prompt block to Slack mrkdwn (#1293)", () => {
+    const request = makeRequest({
+      prompt:
+        "Got it — a couple things to confirm:\n\n1. **Which environment** should I deploy to?\n2. Should the deploy *[skip the build cache](https://example.com)*?\n\n```ts\n// ** keep literal in code **\nconst x = 1;\n```",
+    });
+
+    const blocks = renderInputRequestBlocks(request);
+    const section = blocks[0] as { text: { type: string; text: string }; type: string };
+
+    // `**bold**` and `*italic*` collapse to Slack's single-asterisk bold.
+    expect(section.text.text).toContain("*Which environment*");
+    // Markdown links become Slack's `<url|label>` form.
+    expect(section.text.text).toContain("<https://example.com|skip the build cache>");
+    // Code fence content is passed through untouched — asterisks stay literal.
+    expect(section.text.text).toContain("** keep literal in code **");
+    // The prompt surface must advertise Slack's mrkdwn dialect.
+    expect(section.text.type).toBe("mrkdwn");
+  });
 });
 
 describe("formatInputRequestFallbackText", () => {
@@ -454,6 +473,28 @@ describe("buildFreeformModalView", () => {
     const blocks = view.blocks as Array<Record<string, unknown>>;
     expect(blocks.find((b) => b.type === "section")).toBeUndefined();
   });
+
+  it("converts model-authored GFM in the modal prompt to Slack mrkdwn (#1293)", () => {
+    const view = buildFreeformModalView({
+      metadata: {
+        continuationToken: "slack:C01:1.0",
+        channelId: "C01",
+        threadTs: "1.0",
+        messageTs: "1.1",
+        requestId: "call_abc",
+      },
+      prompt:
+        "One more before I proceed — **which environment** should I target? See [the runbook](https://example.com/runbook).",
+    });
+
+    const blocks = view.blocks as Array<{ type: string; text?: { type: string; text: string } }>;
+    const section = blocks.find((b) => b.type === "section");
+
+    expect(section?.text?.text).toContain("*which environment*");
+    expect(section?.text?.text).toContain("<https://example.com/runbook|the runbook>");
+    // The prompt surface must advertise Slack's mrkdwn dialect.
+    expect(section?.text?.type).toBe("mrkdwn");
+  });
 });
 
 describe("buildAnsweredBlocks", () => {
diff --git a/packages/eve/src/public/channels/slack/hitl.ts b/packages/eve/src/public/channels/slack/hitl.ts
index 9f5d0f30b..5205c6110 100644
--- a/packages/eve/src/public/channels/slack/hitl.ts
+++ b/packages/eve/src/public/channels/slack/hitl.ts
@@ -19,6 +19,7 @@ import {
   truncatePlainText,
   truncateSectionText,
 } from "#public/channels/slack/limits.js";
+import { gfmToSlackMrkdwn } from "#public/channels/slack/mrkdwn.js";
 import type { InputRequest } from "#runtime/input/types.js";
 
 /**
@@ -144,8 +145,17 @@ export function isHitlAction(actionId: string): boolean {
  * Always emits at least the prompt section.
  */
 export function renderInputRequestBlocks(request: InputRequest): unknown[] {
+  // Model-authored prompts are GitHub-flavored markdown by default; convert
+  // to Slack's mrkdwn dialect before embedding so `**bold**`, `__bold__`,
+  // `~~strike~~`, and `[label](url)` render natively instead of showing up
+  // as literal punctuation. The conversion is fence-aware — inline code and
+  // code blocks are passed through untouched. Truncate after conversion so
+  // the limit applies to the rendered text, not the source markdown. (#1293)
   const prompt = {
-    text: { text: truncateSectionText(request.prompt), type: "mrkdwn" },
+    text: {
+      text: truncateSectionText(gfmToSlackMrkdwn(request.prompt)),
+      type: "mrkdwn",
+    },
     type: "section",
   };
   const details = renderInputRequestDetailBlocks(request);
@@ -231,8 +241,16 @@ export function buildFreeformModalView(input: {
   readonly prompt?: string;
 }): Record<string, unknown> {
   const title = input.prompt ? truncateModalTitle(input.prompt) : "Your answer";
+  // See the matching conversion in `renderInputRequestBlocks` — the modal
+  // prompt is the same model-authored markdown surface and needs the same
+  // gfm → mrkdwn pass before reaching Slack. (#1293)
   const promptBlocks = input.prompt
-    ? [{ type: "section", text: { type: "mrkdwn", text: truncateSectionText(input.prompt) } }]
+    ? [
+        {
+          type: "section",
+          text: { type: "mrkdwn", text: truncateSectionText(gfmToSlackMrkdwn(input.prompt)) },
+        },
+      ]
     : [];
   return {
     type: "modal",