feat(api): pagination + search + PDF/DOCX extract in folder agent tools

app/agents/folder_agent.py

@@ -1,10 +1,15 @@
-"""Scoped file-read tool for the project folder feature."""
+"""Scoped file-read and search tools for the project folder feature."""
 from __future__ import annotations
 
 from langchain_core.tools import tool
 
+from app.core.folder_indexer import _extract_docx_text, _extract_pdf_text
 from app.core.ws_context import execute_on_client
 
+# Cap returned slice size to keep tool output under control.
+_MAX_RETURN_CHARS = 50_000
+_MAX_SEARCH_MATCHES = 20
+
 
 def _is_unsafe_path(rel: str) -> bool:
     if not rel:
@@ -19,19 +24,145 @@ def _is_unsafe_path(rel: str) -> bool:
     return ".." in parts
 
 
+async def _fetch_file(project_id: str, relative_path: str, offset: int, length: int) -> dict:
+    """Return the raw Electron tool_result dict for a file read."""
+    return await execute_on_client(
+        action="read_project_folder_file",
+        data={
+            "projectId": project_id,
+            "relativePath": relative_path,
+            "offset": offset,
+            "length": length,
+        },
+    )
+
+
+def _decode(result: dict) -> tuple[str, str, int]:
+    """Decode a tool_result into (text, kind, total_size). For pdf/docx,
+    extracts text from base64. For images, returns a placeholder string.
+    For text, content is already a sliced utf-8 string.
+    """
+    kind = result.get("kind", "text")
+    content = result.get("content", "") or ""
+    total = int(result.get("totalSize", 0) or 0)
+    if kind == "image":
+        return ("[Image file — cannot be navigated as text. See manifest summary.]", kind, total)
+    if kind == "pdf":
+        return (_extract_pdf_text(content), kind, total)
+    if kind == "docx":
+        return (_extract_docx_text(content), kind, total)
+    return (content, kind, total)
+
+
 @tool
-async def read_project_folder_file(project_id: str, relative_path: str) -> str:
-    """Read full content of a file inside the project's linked folder."""
+async def read_project_folder_file(
+    project_id: str,
+    relative_path: str,
+    offset: int = 0,
+    length: int = _MAX_RETURN_CHARS,
+) -> str:
+    """Read a slice of a file inside the project's linked folder.
+
+    Args:
+        project_id: project ID.
+        relative_path: path relative to the linked folder root.
+        offset: char offset to start reading from (0 = beginning).
+        length: max chars to return. Default 50000. Use smaller values to save tokens.
+
+    Returns text content slice with a header showing position. Header tells you
+    when more content is available; call again with the suggested next offset.
+
+    For PDF / DOCX files the backend extracts text first, then applies offset/length
+    on the extracted text. For images returns a placeholder; navigate with the
+    manifest summary instead.
+    """
     if _is_unsafe_path(relative_path):
         return "Access denied"
-    result = await execute_on_client(
-        action="read_project_folder_file",
-        data={"projectId": project_id, "relativePath": relative_path},
-    )
-    content = result.get("content", "")
-    if not content:
-        return f"File not found: {relative_path}"
-    return content
+
+    result = await _fetch_file(project_id, relative_path, offset, length)
+    text, kind, total_size = _decode(result)
+
+    if not text and kind in ("missing", "error"):
+        return f"File not found or unreadable: {relative_path}"
+
+    if kind in ("pdf", "docx"):
+        # Backend extracted full text — apply offset/length on chars.
+        sliced = text[offset:offset + length]
+        slice_end = min(offset + length, len(text))
+        header = (
+            f"[file={relative_path} kind={kind} offset={offset} end={slice_end} "
+            f"totalChars={len(text)}]"
+        )
+        if slice_end < len(text):
+            header += f"\n[More content available — call again with offset={slice_end}.]"
+        return header + "\n" + sliced
+
+    if kind == "text":
+        slice_end = offset + len(text)
+        header = (
+            f"[file={relative_path} kind=text offset={offset} end={slice_end} "
+            f"totalBytes={total_size}]"
+        )
+        if slice_end < total_size:
+            header += f"\n[More content available — call again with offset={slice_end}.]"
+        return header + "\n" + text
+
+    # image or unknown
+    return text
 
 
-FOLDER_TOOLS = [read_project_folder_file]
+@tool
+async def search_project_folder_file(
+    project_id: str,
+    relative_path: str,
+    query: str,
+    context_lines: int = 3,
+) -> str:
+    """Search a project folder file for a query string (case-insensitive substring).
+
+    Args:
+        project_id: project ID.
+        relative_path: path relative to the linked folder root.
+        query: text to search for.
+        context_lines: number of lines of context around each match (default 3).
+
+    Returns matching line ranges with surrounding context and 1-based line numbers.
+    Capped at 20 matches; if more exist the header shows the total.
+
+    Works on text, code, markdown, PDF (extracted), and DOCX (extracted).
+    Images and binary files are not searchable.
+    """
+    if _is_unsafe_path(relative_path):
+        return "Access denied"
+    if not query:
+        return "Empty query."
+
+    # For text we still need full file; pass length=very large.
+    result = await _fetch_file(project_id, relative_path, offset=0, length=10_000_000)
+    text, kind, _ = _decode(result)
+
+    if not text and kind in ("missing", "error"):
+        return f"File not found or unreadable: {relative_path}"
+    if kind == "image":
+        return "Cannot search inside images."
+
+    lines = text.splitlines()
+    q = query.lower()
+    matches = [i for i, line in enumerate(lines) if q in line.lower()]
+    if not matches:
+        return f"No matches for '{query}' in {relative_path}."
+
+    shown = matches[:_MAX_SEARCH_MATCHES]
+    snippets: list[str] = []
+    for i in shown:
+        start = max(0, i - context_lines)
+        end = min(len(lines), i + context_lines + 1)
+        block = "\n".join(f"{n + 1:5d}: {lines[n]}" for n in range(start, end))
+        snippets.append(block)
+
+    header = f"[file={relative_path} matches={len(matches)} showing={len(shown)} query='{query}']"
+    body = "\n---\n".join(snippets)
+    return header + "\n" + body
+
+
+FOLDER_TOOLS = [read_project_folder_file, search_project_folder_file]

tests/test_folder_agent_tool.py

@@ -4,7 +4,10 @@ from unittest.mock import AsyncMock, patch
 
 import pytest
 
-from app.agents.folder_agent import read_project_folder_file
+from app.agents.folder_agent import (
+    read_project_folder_file,
+    search_project_folder_file,
+)
 
 pytestmark = pytest.mark.asyncio
 
@@ -12,10 +15,11 @@ pytestmark = pytest.mark.asyncio
 async def test_happy_path():
     with patch(
         "app.agents.folder_agent.execute_on_client",
-        new=AsyncMock(return_value={"content": "file body"}),
+        new=AsyncMock(return_value={"content": "file body", "kind": "text", "totalSize": 9}),
     ):
         out = await read_project_folder_file.ainvoke({"project_id": "p1", "relative_path": "docs/x.md"})
-    assert out == "file body"
+    assert "file body" in out
+    assert "kind=text" in out
 
 
 async def test_traversal_rejected():
@@ -31,7 +35,105 @@ async def test_absolute_path_rejected():
 async def test_missing_file():
     with patch(
         "app.agents.folder_agent.execute_on_client",
-        new=AsyncMock(return_value={"content": ""}),
+        new=AsyncMock(return_value={"content": "", "kind": "missing", "totalSize": 0}),
     ):
         out = await read_project_folder_file.ainvoke({"project_id": "p1", "relative_path": "ghost.md"})
     assert "not found" in out.lower()
+
+
+async def test_pagination_signals_more_available():
+    # Electron returned the first slice, totalSize larger than slice length.
+    with patch(
+        "app.agents.folder_agent.execute_on_client",
+        new=AsyncMock(return_value={"content": "first chunk", "kind": "text", "totalSize": 1000}),
+    ):
+        out = await read_project_folder_file.ainvoke({
+            "project_id": "p1",
+            "relative_path": "big.txt",
+            "offset": 0,
+            "length": 11,
+        })
+    assert "first chunk" in out
+    assert "More content available" in out
+    assert "offset=11" in out
+
+
+async def test_pdf_extracted_then_sliced(monkeypatch):
+    from app.agents import folder_agent
+    monkeypatch.setattr(folder_agent, "_extract_pdf_text", lambda b: "ABC " * 100)
+    with patch(
+        "app.agents.folder_agent.execute_on_client",
+        new=AsyncMock(return_value={"content": "JVBERi0xLg==", "kind": "pdf", "totalSize": 12}),
+    ):
+        out = await read_project_folder_file.ainvoke({
+            "project_id": "p1",
+            "relative_path": "doc.pdf",
+            "offset": 0,
+            "length": 8,
+        })
+    assert "kind=pdf" in out
+    assert "ABC ABC " in out
+    assert "More content available" in out
+
+
+async def test_image_returns_placeholder():
+    with patch(
+        "app.agents.folder_agent.execute_on_client",
+        new=AsyncMock(return_value={"content": "iVBORw0K", "kind": "image", "totalSize": 1024}),
+    ):
+        out = await read_project_folder_file.ainvoke({"project_id": "p1", "relative_path": "logo.png"})
+    assert "image" in out.lower()
+
+
+async def test_search_finds_match_with_context():
+    body = "alpha\nbeta\nthe needle is here\ngamma\ndelta"
+    with patch(
+        "app.agents.folder_agent.execute_on_client",
+        new=AsyncMock(return_value={"content": body, "kind": "text", "totalSize": len(body)}),
+    ):
+        out = await search_project_folder_file.ainvoke({
+            "project_id": "p1",
+            "relative_path": "log.txt",
+            "query": "needle",
+            "context_lines": 1,
+        })
+    assert "needle" in out
+    assert "matches=1" in out
+    # Context lines included
+    assert "beta" in out
+    assert "gamma" in out
+
+
+async def test_search_no_match():
+    with patch(
+        "app.agents.folder_agent.execute_on_client",
+        new=AsyncMock(return_value={"content": "nothing here", "kind": "text", "totalSize": 12}),
+    ):
+        out = await search_project_folder_file.ainvoke({
+            "project_id": "p1",
+            "relative_path": "x.txt",
+            "query": "zzz",
+        })
+    assert "No matches" in out
+
+
+async def test_search_rejects_traversal():
+    out = await search_project_folder_file.ainvoke({
+        "project_id": "p1",
+        "relative_path": "../etc/passwd",
+        "query": "root",
+    })
+    assert out == "Access denied"
+
+
+async def test_search_image_rejected():
+    with patch(
+        "app.agents.folder_agent.execute_on_client",
+        new=AsyncMock(return_value={"content": "b64data", "kind": "image", "totalSize": 100}),
+    ):
+        out = await search_project_folder_file.ainvoke({
+            "project_id": "p1",
+            "relative_path": "logo.png",
+            "query": "anything",
+        })
+    assert "Cannot search" in out