---
id: "822cd59e-7511-4ab0-b366-8b83ae6dae52"
name: "extract_text()"
micro_summary: "Read me if you need to get text from a page. Antifragile extraction: handles still-loading pages, React SPAs, and empty bodies. The go-to helper for \"let me read the page.\""
status: "resolved"
see_also: []
scope: "Callable helper: extract_text(selector=None) — antifragile text extraction from any page. Handles React/Next.js hydration, pages still loading, SPAs with deferred rendering. Falls back through multiple extraction strategies. Never returns empty string."
warnings: "Cannot reach into iframes — extract_text() runs in the parent page context and silently skips iframe content. If a page has iframes (check with `js('document.querySelectorAll(\"iframe\").length')`), use parent-page JS to reach in: `contentWindow.document.body.innerHTML`."
---
## `extract_text(selector=None)`

Call this when you need to read text from any page. Handles loading state, React hydration, and page shocks automatically.

### Why This Exists

`document.body.innerHTML` on React/Next.js sites includes component definitions (`_jsx` calls), compiled MDX, and scripts — often 5-10x larger than readable text. Example: the glossary page had 976KB of innerHTML but only 17KB of actual text. `extract_text()` bypasses all of that.

### Signature

```python
text = extract_text()           # Full page
text = extract_text(".prose")   # Specific element
```

### Fallback Chain

| Step | Strategy | Why |
|------|----------|-----|
| 0 | `wait_for_load()` + `wait(1)` | React SPA hydration |
| 1 | `innerText` on body/selector | Filters scripts, styles, hidden |
| 2 | `cloneNode()` → remove scripts → `textContent` | Handles pages where innerText is empty |
| 3 | `innerText` on `documentElement` | Pages with no `<body>` |
| 4 | `innerHTML` on target (last resort) | Keeps structure if all else fails |
| Fail | `"[extract_text: no visible text found]"` | Never returns empty |

### Usage Patterns

**Read the page inline (no screenshot needed):**
```python
text = extract_text()
print(text[:500])   # preview top without context bloat
```

**Search (Python's rtk like token-efficient grep):**
```python
import re
text = extract_text()
pattern = re.compile(r"subagent", re.IGNORECASE)
matches = [(i, line[:120]) for i, line in enumerate(text.split("\n")) if pattern.search(line)]
print(f"{len(matches)} lines:" if matches else "no matches")
for ln, line in matches[:20]:
    print(f"  {ln}: {line}")
if len(matches) > 20:
    print(f"  [+{len(matches) - 20} more]")
```

**Search with surrounding context (like rg -C):**
```python
import re
text = extract_text()
lines = text.split("\n")
for m in re.finditer(r"(?i)subagent", text):
    line_no = text[:m.start()].count("\n")
    ctx_start = max(0, line_no - 1)
    ctx_end = min(len(lines), line_no + 2)
    for i in range(ctx_start, ctx_end):
        prefix = "→ " if i == line_no else "  "
        print(f"{prefix}{i}: {lines[i][:120]}")
    print("---")
```

**Preview top and bottom to understand page shape:**
```python
text = extract_text()
print(text[:400])      # what's at the top?
print("...")
print(text[-400:])     # what's at the bottom?
```

**Save to file for later:**
```python
from pathlib import Path
Path("/tmp/page.md").write_text(text)
```

**Target a specific section:**
```python
text = extract_text("#subagent")  # just the subagent definition
print(text[:200])
```