Restore bounded project context overview

tests/core/test_system_prompt.py

@@ -2,6 +2,7 @@
 
 import os
 from pathlib import Path
+import subprocess
 
 import pytest
 
@@ -25,3 +26,77 @@ def test_run_git_survives_non_utf8_output(tmp_path: Path, monkeypatch) -> None:
 
     # The bad bytes are replaced with U+FFFD instead of crashing
     assert "\ufffd" in result.stdout
+
+
+def test_project_context_includes_git_file_overview(
+    tmp_path: Path, monkeypatch
+) -> None:
+    def fake_run_git(
+        self: ProjectContextProvider, args: list[str], timeout: float
+    ) -> subprocess.CompletedProcess[str]:
+        stdout = ""
+        if args == ["ls-files"]:
+            stdout = "src/main.py\nREADME.md\nsrc/main.py\n\n"
+        elif args == ["branch", "--show-current"]:
+            stdout = "main\n"
+        elif args == ["branch", "-r"]:
+            stdout = "origin/main\n"
+        elif args[:2] == ["log", "--oneline"]:
+            stdout = "abc123 init\n"
+        elif args != ["status", "--porcelain"]:
+            raise AssertionError(args)
+        return subprocess.CompletedProcess(args, 0, stdout=stdout)
+
+    monkeypatch.setattr(ProjectContextProvider, "_run_git", fake_run_git)
+
+    provider = ProjectContextProvider(ProjectContextConfig(), root_path=tmp_path)
+
+    context = provider.get_full_context()
+
+    assert "Project file overview (snapshot at conversation start):" in context
+    assert "- README.md" in context
+    assert "- src/main.py" in context
+    assert context.count("- src/main.py") == 1
+
+
+def test_project_context_file_overview_is_bounded(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    def fake_run_git(
+        self: ProjectContextProvider, args: list[str], timeout: float
+    ) -> subprocess.CompletedProcess[str]:
+        stdout = ""
+        if args == ["ls-files"]:
+            stdout = "\n".join(f"file_{index}.py" for index in range(5))
+        return subprocess.CompletedProcess(args, 0, stdout=stdout)
+
+    monkeypatch.setattr(ProjectContextProvider, "_run_git", fake_run_git)
+    monkeypatch.setattr("vibe.core.system_prompt._MAX_FILE_OVERVIEW_ENTRIES", 2)
+    provider = ProjectContextProvider(ProjectContextConfig(), root_path=tmp_path)
+
+    overview = provider.get_file_overview()
+
+    assert "- file_0.py" in overview
+    assert "- file_1.py" in overview
+    assert "- file_2.py" not in overview
+    assert "... 3 more files omitted" in overview
+
+
+def test_project_context_file_overview_falls_back_to_directory_scan(
+    tmp_path: Path, monkeypatch
+) -> None:
+    (tmp_path / "pkg").mkdir()
+    (tmp_path / "pkg" / "module.py").write_text("", encoding="utf-8")
+    (tmp_path / "node_modules").mkdir()
+    (tmp_path / "node_modules" / "ignored.js").write_text("", encoding="utf-8")
+    (tmp_path / ".hidden").write_text("", encoding="utf-8")
+
+    monkeypatch.setattr(ProjectContextProvider, "_list_git_files", lambda self: [])
+
+    provider = ProjectContextProvider(ProjectContextConfig(), root_path=tmp_path)
+
+    overview = provider.get_file_overview()
+
+    assert "- pkg/module.py" in overview
+    assert "ignored.js" not in overview
+    assert ".hidden" not in overview

vibe/core/prompts/project_context.md

@@ -2,3 +2,5 @@ Absolute path: $abs_path
 
 gitStatus: This is the git status at the start of the conversation. Note that this status is a snapshot in time, and will not update during the conversation.
 $git_status
+
+$file_overview

vibe/core/skills/builtins/vibe.py

@@ -147,7 +147,7 @@
 
 # System prompt composition
 include_model_info = true         # Include model name in system prompt
-include_project_context = true    # Include project context (git info, cwd) in system prompt
+include_project_context = true    # Include project context (git info, cwd, file overview) in system prompt
 include_prompt_detail = true      # Include OS info, tool prompts, skills, and agents in system prompt
 
 # Voice features

vibe/core/system_prompt.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from collections.abc import Sequence
 from concurrent.futures import ThreadPoolExecutor
 from datetime import date
 import html
@@ -31,6 +32,17 @@
     from vibe.core.tools.manager import ToolManager
 
 _git_status_cache: dict[Path, str] = {}
+_MAX_FILE_OVERVIEW_ENTRIES = 200
+
+
+def _normalize_overview_entries(entries: Sequence[str]) -> list[str]:
+    clean_entries = []
+    for entry in entries:
+        clean = entry.strip()
+        if not clean:
+            continue
+        clean_entries.append(clean)
+    return sorted(dict.fromkeys(clean_entries))
 
 
 class ProjectContextProvider:
@@ -40,6 +52,63 @@ def __init__(
         self.root_path = Path(root_path).resolve()
         self.config = config
 
+    def get_file_overview(self) -> str:
+        entries = self._list_project_files()
+        if not entries:
+            return "No project file overview available."
+
+        limit = _MAX_FILE_OVERVIEW_ENTRIES
+        visible = entries[:limit]
+        lines = ["Project file overview (snapshot at conversation start):"]
+        lines.extend(f"- {entry}" for entry in visible)
+        if len(entries) > limit:
+            lines.append(f"... {len(entries) - limit} more files omitted")
+        return "\n".join(lines)
+
+    def _list_project_files(self) -> list[str]:
+        if from_git := self._list_git_files():
+            return from_git
+        return self._list_directory_files()
+
+    def _list_git_files(self) -> list[str]:
+        try:
+            result = self._run_git(["ls-files"], min(self.config.timeout_seconds, 10.0))
+        except (subprocess.CalledProcessError, subprocess.TimeoutExpired):
+            return []
+        return _normalize_overview_entries(result.stdout.splitlines())
+
+    def _list_directory_files(self) -> list[str]:
+        max_entries = _MAX_FILE_OVERVIEW_ENTRIES + 1
+        entries: list[str] = []
+        ignored_dirs = {
+            ".git",
+            ".hg",
+            ".svn",
+            ".mypy_cache",
+            ".pytest_cache",
+            ".ruff_cache",
+            ".tox",
+            ".venv",
+            "__pycache__",
+            "node_modules",
+        }
+        try:
+            for current_root, dirs, files in os.walk(self.root_path):
+                dirs[:] = [
+                    d for d in dirs if d not in ignored_dirs and not d.startswith(".")
+                ]
+                dirs.sort()
+                for file_name in sorted(files):
+                    if file_name.startswith("."):
+                        continue
+                    path = Path(current_root, file_name)
+                    entries.append(path.relative_to(self.root_path).as_posix())
+                    if len(entries) >= max_entries:
+                        return entries
+        except OSError:
+            return []
+        return entries
+
     def get_git_status(self) -> str:
         if self.root_path in _git_status_cache:
             return _git_status_cache[self.root_path]
@@ -147,8 +216,12 @@ def get_full_context(self) -> str:
         git_status = self.get_git_status()
 
         template = UtilityPrompt.PROJECT_CONTEXT.read()
+        file_overview = self.get_file_overview()
+
         return Template(template).safe_substitute(
-            abs_path=str(self.root_path), git_status=git_status
+            abs_path=str(self.root_path),
+            git_status=git_status,
+            file_overview=file_overview,
         )