From 88b22b0abc2e5106d43fccec607e2ab1396ae0a8 Mon Sep 17 00:00:00 2001
From: DreamLinx <dreamlinx@gmail.com>
Date: Tue, 18 Aug 2026 16:08:24 +0800
Subject: [PATCH] fix(writer): preserve complete sections when truncating

---
 CHANGELOG.md                       |  9 ++++++++
 src/codeindex/writers/README_AI.md |  2 +-
 src/codeindex/writers/utils.py     | 35 ++++++++++++++++++++----------
 tests/test_smart_writer.py         |  2 +-
 tests/writers/test_utils.py        | 21 ++++++++++++++++++
 5 files changed, 56 insertions(+), 13 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9a81c77..e325013 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Fixed
+
+- **Detailed `README_AI.md` truncation preserves complete file sections** (GH
+  #178). The 10KB navigation budget now reserves its notice first, then cuts
+  at the last complete Markdown heading (`##` through `######`) instead of
+  raw bytes inside a `### file` symbol list. The fallback notice now directs
+  agents to the relevant source files rather than assuming per-module README
+  files exist. Tiny configured budgets remain bounded too.
+
 ## [0.37.0] - 2026-08-15
 
 ### Removed (BREAKING)
diff --git a/src/codeindex/writers/README_AI.md b/src/codeindex/writers/README_AI.md
index 96e08b2..1363d84 100644
--- a/src/codeindex/writers/README_AI.md
+++ b/src/codeindex/writers/README_AI.md
@@ -1,5 +1,5 @@
 <!-- codeindex navigation index — agent: drill into source via Read/Grep for precise mechanism; do not treat this as final word. -->
-<!-- Generated by codeindex (detailed) at 2026-08-15T23:23:40.946606 -->
+<!-- Generated by codeindex (detailed) at 2026-08-18T16:06:36.207750 -->
 
 # writers
 
diff --git a/src/codeindex/writers/utils.py b/src/codeindex/writers/utils.py
index 8d4047d..282bd36 100644
--- a/src/codeindex/writers/utils.py
+++ b/src/codeindex/writers/utils.py
@@ -299,19 +299,32 @@ def truncate_content(content: str, max_size: int) -> tuple[str, bool]:
     if len(content_bytes) <= max_size:
         return content, False
 
-    # Find a good truncation point
-    truncated = content_bytes[:max_size - 200].decode('utf-8', errors='ignore')
-
-    # Try to truncate at a section boundary
-    last_section = truncated.rfind("\n## ")
-    if last_section > len(truncated) // 2:
-        truncated = truncated[:last_section]
-
-    # Add truncation notice
-    truncated += (
+    notice = (
         "\n\n---\n"
         "_Content truncated due to size limit. "
-        "See individual module README files for details._\n"
+        "Read the relevant source files for details._\n"
     )
+    if len(notice.encode('utf-8')) > max_size:
+        notice = "_Truncated._\n"
+        if len(notice.encode('utf-8')) > max_size:
+            return notice[:max_size], True
+    budget = max(0, max_size - len(notice.encode('utf-8')))
+    truncated = content_bytes[:budget].decode('utf-8', errors='ignore')
+
+    # README generators use ## for document sections and ### for files. Prefer
+    # the last complete heading boundary so detailed output never ends midway
+    # through a file's symbol list.
+    last_section = max(
+        truncated.rfind(f"\n{'#' * level} ")
+        for level in range(2, 7)
+    )
+    if last_section >= 0:
+        truncated = truncated[:last_section]
+    else:
+        last_line = truncated.rfind("\n")
+        if last_line > 0:
+            truncated = truncated[:last_line]
+
+    truncated = truncated.rstrip() + notice
 
     return truncated, True
diff --git a/tests/test_smart_writer.py b/tests/test_smart_writer.py
index 0921d4c..a52ff83 100644
--- a/tests/test_smart_writer.py
+++ b/tests/test_smart_writer.py
@@ -267,7 +267,7 @@ def test_smart_writer_size_limit():
 
         assert result.success
         assert result.truncated
-        assert result.size_bytes <= 1024 + 200  # Allow some margin for truncation notice
+        assert result.size_bytes <= 1024
 
 
 def test_determine_level():
diff --git a/tests/writers/test_utils.py b/tests/writers/test_utils.py
index 419f0b2..2356508 100644
--- a/tests/writers/test_utils.py
+++ b/tests/writers/test_utils.py
@@ -325,9 +325,30 @@ def test_truncates_at_section_boundary(self):
         result, truncated = truncate_content(sections, 60)
         assert truncated
         assert "truncated" in result.lower()
+        assert len(result.encode("utf-8")) <= 60
 
     def test_truncation_adds_notice(self):
         content = "x" * 2000
         result, truncated = truncate_content(content, 500)
         assert truncated
         assert "Content truncated" in result
+
+    def test_detailed_file_section_is_never_cut_mid_list_item(self):
+        """Detailed output uses ``### file`` headings, not only ``##``."""
+        content = (
+            "# cli\n\n"
+            "## Files\n\n"
+            "### alpha.py\n"
+            "- `def alpha() -> None`\n\n"
+            "### beta.py\n"
+            "- `def beta_with_a_very_long_signature(" + "arg, " * 80 + ") -> None`\n"
+        )
+
+        result, truncated = truncate_content(content, 300)
+
+        assert truncated
+        assert "### alpha.py" in result
+        assert "### beta.py" not in result
+        assert result.count("`") % 2 == 0
+        assert "Read the relevant source files" in result
+        assert len(result.encode("utf-8")) <= 300