## Summary epic r04 - begin refactor ## Change Type - [x] Cowork feature - [ ] Bug fix - [ ] Core AI contribution - [ ] Test / hardening - [ ] Performance - [ ] Documentation ## Related Work Cowork Task: Core Repo: http://34.143.229.138/gitea-admin/fsg-ai-core-assets Core AI Issue: Core Task: Related PR: ## Scope What is intentionally included? What is intentionally NOT included? ## Validation - [ ] Unit tests - [ ] Integration tests - [ ] Manual verification - [ ] Regression check Commands / evidence: ## Security Impact Permission / credential / network / customer data impact: ## Compatibility - [ ] No breaking change - [ ] Breaking change documented ## Reviewer Notes Anything Cowork reviewers should pay attention to. --------- Co-authored-by: Anh Tran Nguyen Minh <anhtnm1@fpt.com> Co-authored-by: Huong Le Thi Thien <huongltt35@fpt.com> Co-authored-by: Nam Pham Dinh Thanh <nampdt@fpt.com> Co-authored-by: Vu Dam Tuan <vudt15@fpt.com> Co-authored-by: Hiep Ha Van <hiephv3@fpt.com> Co-authored-by: Lam Hoang Van <lamhv7@fpt.com> Reviewed-on: #7 Co-authored-by: Duy Le Huu <duylh19@fpt.com>
This commit was merged in pull request #7.
This commit is contained in:
Nam Pham Dinh Thanh
@@ -0,0 +1,220 @@
|
||||
"""Tệp đính kèm của một lượt chat — R08-T03.
|
||||
|
||||
Đọc nội dung tệp người dùng kèm vào rồi ghép vào câu hỏi. Ba thứ đáng
|
||||
chú ý:
|
||||
|
||||
* ``_enforce_attachment_security`` chạy TRƯỚC khi nội dung vào ngữ cảnh
|
||||
model — đây là một trong ba tầng kiểm của R09.
|
||||
* ``_attach_char_limit`` cắt bớt tệp quá dài; không cắt thì một tệp log
|
||||
vài chục MB đủ làm hỏng cả lượt.
|
||||
* Kèm cả thư mục thì chỉ lấy DANH SÁCH tệp, không đọc nội dung từng cái.
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from pathlib import Path
|
||||
from typing import List
|
||||
from PySide6.QtCore import Qt
|
||||
from ...core.worker import AgentWorker
|
||||
from ...i18n import tr
|
||||
from ...ui.osutil import is_image
|
||||
|
||||
|
||||
class AttachmentMixin:
|
||||
"""Trộn vào ChatPanel."""
|
||||
|
||||
def _on_attachments_added(self, paths: List[str]) -> None:
|
||||
# Push attachments into the Input box as soon as they're attached.
|
||||
"""Tệp vừa được đính kèm: đưa luôn vào mục "Tệp đầu vào" để người dùng thấy ngay."""
|
||||
for p in paths:
|
||||
self.input_section.add(p)
|
||||
|
||||
def _on_attachment_removed(self, path: str) -> None:
|
||||
# A file added by mistake was removed in the composer — drop it from the
|
||||
# Input panel too (only matters before the message is sent).
|
||||
"""Gỡ tệp đính kèm nhầm khỏi cả mục "Tệp đầu vào" (chỉ có ý nghĩa trước khi gửi)."""
|
||||
self.input_section.remove(path)
|
||||
|
||||
def _attach_char_limit(self) -> int:
|
||||
"""Per-file content cap (characters) from the Settings token limit
|
||||
(~4 chars/token)."""
|
||||
try:
|
||||
tokens = int(self.ctx.config.data.get("attachments", {}).get("max_tokens", 500000))
|
||||
except (TypeError, ValueError):
|
||||
tokens = 500000
|
||||
return max(1000, tokens) * 4
|
||||
|
||||
def _augment(self, text: str, attachments: List[str], notify=None) -> str:
|
||||
"""Embed attachment paths AND their extracted contents into the prompt so
|
||||
the agent actually reads and analyses each attached file.
|
||||
|
||||
Additionally, scans the workspace/output folder for existing files and
|
||||
loads them as input data so the agent can read/process them automatically.
|
||||
|
||||
``notify``, if given, is called with UI-visible events (a live "reading
|
||||
page X/Y" progress notice, and a warning when a file's content could not
|
||||
be read) instead of failures being silently handed to the model as an
|
||||
opaque inline note."""
|
||||
has_attachments = bool(attachments)
|
||||
limit = self._attach_char_limit()
|
||||
lines = [text] if text else []
|
||||
|
||||
# --- User-attached files ---
|
||||
if has_attachments:
|
||||
lines.append("\n[Attachments] — read and use these files to answer the request:")
|
||||
for p in attachments:
|
||||
lines.extend(self._read_one_attachment(p, limit, notify))
|
||||
|
||||
# --- Auto-load existing workspace/output folder files as input data ---
|
||||
# This is what makes "📁 Chọn thư mục khác" useful as an INPUT folder
|
||||
# too: every file already in the chosen folder is read and embedded so
|
||||
# the agent can act on their contents without manual attaching.
|
||||
workspace = self.workspace_dir()
|
||||
max_files = int(self.ctx.config.data.get("attachments", {})
|
||||
.get("max_files", 10) or 0)
|
||||
if workspace is not None:
|
||||
lines.extend(self._folder_input_lines(
|
||||
workspace,
|
||||
"[Workspace files] — existing files in output folder, "
|
||||
"read and use as input data. The user expects you to "
|
||||
"process these files automatically:",
|
||||
limit, max_files, notify))
|
||||
|
||||
# --- Project knowledge (Claude-Projects style) ---
|
||||
# Only scanned separately when it's a DIFFERENT folder from the
|
||||
# session's own workspace — for Cowork the two are now the same
|
||||
# folder (a project has one shared workspace, no per-thread
|
||||
# sub-folder), so this never double-scans the same directory.
|
||||
knowledge = self.project_knowledge_dir()
|
||||
if knowledge is not None and knowledge != workspace:
|
||||
lines.extend(self._folder_input_lines(
|
||||
knowledge,
|
||||
"[Project files] — shared knowledge files of this project, "
|
||||
"available to every conversation in it. Read and use them "
|
||||
"as context for the request:",
|
||||
limit, max_files, notify))
|
||||
|
||||
return "\n".join(lines)
|
||||
|
||||
def _folder_input_lines(self, folder: Path, header: str, limit: int,
|
||||
max_files: int, notify=None) -> list:
|
||||
"""Embed a folder's readable files into the prompt — recursing into
|
||||
every sub-folder, any depth, not just the top level, so files placed
|
||||
in nested folders are read and processed too (same per-message file
|
||||
cap as manual attachments — Settings → Attachments → max files;
|
||||
0 = unlimited — so a folder with dozens of files can't blow the
|
||||
context window)."""
|
||||
from ...core.doc_extract import find_input_files
|
||||
|
||||
out: list = []
|
||||
shown, total = find_input_files(folder, self._INPUT_EXTS, max_files)
|
||||
if shown:
|
||||
out.append("\n" + header)
|
||||
for f in shown:
|
||||
out.extend(self._read_one_attachment(str(f), limit, notify))
|
||||
if total > len(shown):
|
||||
skipped = total - len(shown)
|
||||
out.append(f"…({skipped} more files in the folder were not "
|
||||
"loaded — per-message attachment limit; mention a "
|
||||
"file by name if the user asks about it)")
|
||||
if notify is not None:
|
||||
notify({"type": "notice", "level": "warning",
|
||||
"text": tr("chat.workspace_files_capped",
|
||||
shown=len(shown), total=total)})
|
||||
return out
|
||||
|
||||
def _read_one_attachment(self, path: str, limit: int, notify=None) -> list:
|
||||
"""Read and format one attachment/workspace file. Returns list of lines.
|
||||
|
||||
Handles every file type: images (noted with path), MS Office / PDF /
|
||||
OpenDocument / text (extracted), and ZIP archives — which are auto-
|
||||
extracted into the workspace and their contents read + processed."""
|
||||
name = Path(path).name
|
||||
result = []
|
||||
if is_image(path):
|
||||
result.append(f"- {name} (image at {path})")
|
||||
return result
|
||||
from ...core.doc_extract import is_zip
|
||||
if is_zip(path):
|
||||
result.extend(self._read_zip_attachment(path, name, limit, notify))
|
||||
return result
|
||||
|
||||
def progress(page: int, total: int, _name=name) -> None:
|
||||
"""Báo tiến độ trích nội dung, chỉ với tệp nhiều trang — tệp một trang thì dòng
|
||||
tiến độ chỉ làm nhiễu.
|
||||
"""
|
||||
if notify is not None and total > 1:
|
||||
notify({"type": "notice", "level": "progress",
|
||||
"text": tr("chat.reading_progress", name=_name, page=page, total=total)})
|
||||
|
||||
content, note = self._read_attachment_text(path, progress=progress)
|
||||
if content is None:
|
||||
result.append(f"- {name} ({note}; located at {path})")
|
||||
if notify is not None:
|
||||
notify({"type": "notice", "level": "warning",
|
||||
"text": tr("chat.attachment_failed", name=name, note=note)})
|
||||
return result
|
||||
self._enforce_attachment_security(name, content) # raises SecurityBlocked on a violation
|
||||
extra = ""
|
||||
if len(content) > limit:
|
||||
content = content[:limit]
|
||||
extra = f"\n…(truncated to ~{limit // 4} tokens)…"
|
||||
result.append(f"- {name} ({path})")
|
||||
result.append(f"\n--- Content of {name} ---\n{content}{extra}\n--- end of {name} ---")
|
||||
return result
|
||||
|
||||
def _read_zip_attachment(self, path: str, name: str, limit: int, notify=None) -> list:
|
||||
"""Auto-extract a .zip into the workspace and read+process its files, so
|
||||
an attached archive is unpacked and its contents used automatically."""
|
||||
from ...core.doc_extract import extract_archive
|
||||
ws = self.workspace_dir()
|
||||
dest = (Path(ws) if ws is not None else Path(path).parent) / Path(name).stem
|
||||
files = extract_archive(path, dest)
|
||||
result = [f"- {name} (archive) — extracted {len(files)} file(s) into the workspace at "
|
||||
f"{dest}. Read/edit them there as needed."]
|
||||
if self.workspace_dir() is not None:
|
||||
self.output_changed.emit(str(self.workspace_dir())) # let the graph/folder refresh
|
||||
max_files = int(self.ctx.config.data.get("attachments", {}).get("max_files", 10) or 0)
|
||||
shown = files[:max_files] if max_files else files
|
||||
for f in shown:
|
||||
result.extend(self._read_one_attachment(str(f), limit, notify))
|
||||
if max_files and len(files) > max_files:
|
||||
result.append(f"- …and {len(files) - max_files} more file(s) in {dest} "
|
||||
"(not inlined; open/read them from the workspace as needed).")
|
||||
return result
|
||||
|
||||
def _enforce_attachment_security(self, filename: str, content: str) -> None:
|
||||
"""Agent Security's attachment layer (Settings → 🛡 Agent Security) —
|
||||
scans extracted file content for malicious payloads BEFORE it enters
|
||||
the model's context. No-op when disabled. Raises SecurityBlocked
|
||||
(propagates out of _augment → the worker job → AgentWorker.failed,
|
||||
which the panel shows as a chat error) on a violation."""
|
||||
sec = self.ctx.config.data.get("agent_security", {})
|
||||
if not sec.get("enabled") or not sec.get("validate_attachments", True):
|
||||
return
|
||||
from ...core.agent_security import SecurityBlocked, combined_rules_text, validate_attachment
|
||||
from ...core.agent_security_alert import notify_admin
|
||||
|
||||
rules_text = combined_rules_text(self.ctx.config)
|
||||
verdict = validate_attachment(self.build_provider(), filename, content, rules_text)
|
||||
if verdict.allowed:
|
||||
return
|
||||
notify_admin(self.ctx.config, verdict, detail=f"file: {filename}")
|
||||
raise SecurityBlocked(verdict)
|
||||
|
||||
@staticmethod
|
||||
def _read_attachment_text(path: str, progress=None):
|
||||
"""Best-effort text extraction so the agent can read the attachment.
|
||||
Returns (text, note); text is None when nothing readable was found.
|
||||
|
||||
Delegates to core.doc_extract, which parses docx/xlsx/pptx/odf directly
|
||||
(stdlib, no extra packages), uses pypdf for PDFs (reporting per-page
|
||||
``progress`` for multi-page files), and falls back to a headless
|
||||
LibreOffice conversion for anything else."""
|
||||
from ...core.doc_extract import extract_text
|
||||
|
||||
return extract_text(path, progress=progress)
|
||||
|
||||
def project_knowledge_dir(self):
|
||||
"""Folder of project-level shared knowledge files (None = no project
|
||||
knowledge). Overridden by the Cowork tab for non-default projects."""
|
||||
return None
|
||||
Reference in New Issue
Block a user