feat(tools): name the dead end — past-EOF and empty-file notes in read_file

A read past EOF returned content '900|' (a phantom line-number prefix
that looks like a real line) and an empty file returned '1|' — both
ambiguous silence: indistinguishable, from inside the model, from a
broken tool, so it re-reads and widens windows. Name the dead end and
its recovery instead: 'offset 900 is beyond the end of the file (412
lines total). Retry with offset <= 412.' / 'File is empty (0 bytes).'
Notes, not errors — a fact about the file is not a failure.

Boundary pinned by test: offset == total_lines still reads (an
off-by-one in a resume hint is a silently corrupted read).

Measured (file-only arm, 3 reps, control vs feature): qwen3.8-max
-18% tokens, -26% tool calls, -17% turns across the two affected
tasks; opus-4.8 flat (within rep noise); accuracy held 1.00.

tests/tools/test_read_past_eof_note.py

@@ -0,0 +1,44 @@
+"""Tests for the past-EOF and empty-file notes in read_file.
+
+An empty content string is ambiguous from inside the model (empty file?
+bad offset? broken tool?) — the tool names the dead end and the recovery.
+"""
+
+import json
+
+from tools.file_tools import read_file_tool
+
+
+class TestPastEofNote:
+    def test_offset_beyond_eof_names_recovery(self, tmp_path, monkeypatch):
+        monkeypatch.setenv("TERMINAL_CWD", str(tmp_path))
+        p = tmp_path / "r.txt"
+        p.write_text("\n".join(f"l{i}" for i in range(1, 51)) + "\n")
+        result = json.loads(read_file_tool(str(p), offset=900, limit=50))
+        hint = result.get("hint") or ""
+        assert "beyond the end" in hint
+        assert "50" in hint  # states actual line count
+        assert not result.get("error"), "a fact about the file is not an error"
+
+    def test_offset_at_last_line_still_reads(self, tmp_path, monkeypatch):
+        monkeypatch.setenv("TERMINAL_CWD", str(tmp_path))
+        p = tmp_path / "r.txt"
+        p.write_text("a\nb\nc\n")
+        result = json.loads(read_file_tool(str(p), offset=3, limit=10))
+        assert "c" in result.get("content", "")
+        assert "beyond" not in (result.get("hint") or "")
+
+    def test_empty_file_says_so(self, tmp_path, monkeypatch):
+        monkeypatch.setenv("TERMINAL_CWD", str(tmp_path))
+        p = tmp_path / "e.txt"
+        p.write_text("")
+        result = json.loads(read_file_tool(str(p)))
+        assert "empty" in (result.get("hint") or "").lower()
+        assert not result.get("error")
+
+    def test_normal_pagination_hint_unchanged(self, tmp_path, monkeypatch):
+        monkeypatch.setenv("TERMINAL_CWD", str(tmp_path))
+        p = tmp_path / "r.txt"
+        p.write_text("\n".join(f"l{i}" for i in range(1, 300)) + "\n")
+        result = json.loads(read_file_tool(str(p), offset=1, limit=100))
+        assert "offset=101" in (result.get("hint") or "")

tools/file_operations.py

@@ -1331,7 +1331,30 @@ class ShellFileOperations(FileOperations):
         hint = None
         if truncated:
             hint = f"Use offset={end_line + 1} to continue reading (showing {offset}-{end_line} of {total_lines} lines)"
-        
+
+        # Ambiguous-silence guards: an empty content string is
+        # indistinguishable, from inside the model, from a broken tool —
+        # it re-reads, widens the window, tries another path. Name the
+        # dead end and its recovery instead.
+        if file_size == 0:
+            return ReadResult(
+                content="",
+                total_lines=0,
+                file_size=0,
+                hint="File is empty (0 bytes).",
+            )
+        if offset > total_lines > 0:
+            return ReadResult(
+                content="",
+                total_lines=total_lines,
+                file_size=file_size,
+                hint=(
+                    f"Note: offset {offset} is beyond the end of the file "
+                    f"({total_lines} lines total). Retry with offset <= "
+                    f"{total_lines}."
+                ),
+            )
+
         return ReadResult(
             content=self._add_line_numbers(read_output, offset),
             total_lines=total_lines,