feat: render screenshots inline in TUI via Image component; add parallelization guidance to SKILL.md

- browser_screenshot now has renderResult that creates an Image component
  for inline display in Kitty/iTerm2/Ghostty/WezTerm terminals
- Falls back to dim text placeholder on terminals without image support
- Updated system prompt table description to mention TUI inline rendering
- Added Parallelization section to SKILL.md teaching the agent to batch
  independent operations (browser_open_urls, browser_http_get, etc.)

3 files changed, 81 insertions, 2 deletions

diff --git a/skills/pi-browser-harness/SKILL.md b/skills/pi-browser-harness/SKILL.md
index 81a233c..60cd069 100644
--- a/skills/pi-browser-harness/SKILL.md
+++ b/skills/pi-browser-harness/SKILL.md
@@ -32,6 +32,60 @@ browser_screenshot() → browser_click(x, y) → browser_screenshot()
 Compositor-level clicks (`Input.dispatchMouseEvent`) work through iframes, shadow DOM,
 and cross-origin content. No selectors needed.
 
+## Parallelization
+
+**Always parallelize when possible.** Never sequence independent operations.
+
+### What can run in parallel
+
+- **Opening result pages:** Use `browser_open_urls` to open multiple URLs in parallel tabs
+  instead of navigating one at a time. After opening, use `browser_list_tabs` to
+  see the new tabs and `browser_switch_tab` to visit each one.
+
+- **API calls:** `browser_http_get` calls are independent of browser state — fire
+  multiple GETs in parallel to fetch APIs, static pages, or search results.
+
+- **JS extraction across tabs:** Once tabs are loaded, extract data from each tab
+  with `browser_execute_js` — these can run in parallel across different tabs.
+
+- **Screenshots across tabs:** After opening multiple pages, capture screenshots
+  of each tab in parallel. Each `browser_screenshot` targets the currently active
+  tab, so switch tabs first, then screenshot.
+
+- **Read + browser tools:** Use `read` and `browser_*` tools in the same turn —
+  reading files or previous results is independent of browser actions.
+
+### What must be sequential
+
+- **Page interactions:** Clicks, typing, scrolling, and form submissions on the
+  same page depend on prior state and must be sequential.
+
+- **Tab switching then acting:** `browser_switch_tab` must complete before acting
+  on the switched-to tab.
+
+### Pattern: Parallel research across tabs
+
+```
+# Step 1: Open all result pages at once
+browser_open_urls(["url1", "url2", "url3"])
+
+# Step 2: Wait for them to load, then extract in parallel
+browser_list_tabs()  # get targetIds
+
+# Step 3: Extract data from each tab (parallel)
+browser_switch_tab(targetId1) → browser_screenshot()   # These are sequential per tab
+browser_switch_tab(targetId2) → browser_execute_js(...) # but you batch switch+act
+```
+
+### Pattern: Parallel API + browser
+
+```
+# Fire API calls while interacting with a page
+browser_http_get("https://api.example.com/data")  # parallel
+browser_click(x, y)                                 # parallel (different target)
+browser_type("search term")                         # parallel (different target)
+```
+
 ## Available Tools
 
 ### Navigation
@@ -49,7 +103,7 @@ and cross-origin content. No selectors needed.
 - `browser_scroll` — scroll the page at coordinates
 
 ### Visual
-- `browser_screenshot` — capture PNG screenshot
+- `browser_screenshot` — capture PNG or JPEG screenshot (rendered inline in the TUI)
 
 ### Data Extraction
 - `browser_execute_js` — execute JavaScript and return result
diff --git a/src/domains/screenshot.ts b/src/domains/screenshot.ts
index 03b49f0..60d4b26 100644
--- a/src/domains/screenshot.ts
+++ b/src/domains/screenshot.ts
@@ -1,5 +1,7 @@
+import { readFileSync } from "node:fs";
 import { writeFile, rename } from "node:fs/promises";
 import { Type } from "typebox";
+import { Image, type ImageTheme, Text } from "@mariozechner/pi-tui";
 import type { BrowserClient } from "../client";
 import { type Result, err, ok } from "../util/result";
 import { defineBrowserTool, type ToolErr, type ToolOk } from "../util/tool";
@@ -82,6 +84,29 @@ export const screenshotTool = defineBrowserTool({
       details: { path: cap.data.path, format: cap.data.format, attached: false },
     });
   },
+
+  renderResult(result, _expanded, theme) {
+    const details = result.details as { path?: string; format?: string } | undefined;
+    const filePath = details?.path;
+    if (!filePath) {
+      return new Text(theme.fg("error", "Screenshot path missing"), 0, 0);
+    }
+    try {
+      const buf = readFileSync(filePath);
+      const b64 = buf.toString("base64");
+      const mimeType = details?.format === "jpeg" ? "image/jpeg" : "image/png";
+      const imageTheme: ImageTheme = {
+        fallbackColor: (str: string) => theme.fg("dim", str),
+      };
+      return new Image(b64, mimeType, imageTheme, {
+        maxWidthCells: 80,
+        maxHeightCells: 24,
+        filename: filePath,
+      });
+    } catch {
+      return new Text(theme.fg("warning", `Screenshot saved: ${filePath}`), 0, 0);
+    }
+  },
 });
 
 export const captureWithCrosshair = async (
diff --git a/src/prompt.ts b/src/prompt.ts
index d4ac8a3..6d2b2ab 100644
--- a/src/prompt.ts
+++ b/src/prompt.ts
@@ -21,7 +21,7 @@ to interact with pages visually, and other tools alongside them.
 
 | Tool | Purpose |
 |------|---------|
-| browser_screenshot | Capture a PNG screenshot of the current page (result details include path and attached: false — the LLM reads the image via the saved file path) |
+| browser_screenshot | Capture a PNG or JPEG screenshot of the current page (rendered inline in the TUI; the LLM reads the image via the saved file path) |
 | browser_click | Click at viewport coordinates |
 | browser_type | Type text into the focused element |
 | browser_press_key | Press a keyboard key (Enter, Tab, Escape, arrows, etc.) |