feat(chat): 结构化用户附件并支持图片预览

This commit is contained in:
caoqianming 2026-08-13 10:32:08 +08:00
parent c4418711ab
commit 054aaf8e8c
20 changed files with 354 additions and 21 deletions

View File

@ -4,6 +4,10 @@
> 维护口径:只记**用户可感知**的变化("能干什么了"而不是"改了哪个模块"),纯内部重构 / 修复不记, > 维护口径:只记**用户可感知**的变化("能干什么了"而不是"改了哪个模块"),纯内部重构 / 修复不记,
> 所以不是每个版本号都有条目。条目格式 `## <版本> — <日期>`,新条目加在最上面。 > 所以不是每个版本号都有条目。条目格式 `## <版本> — <日期>`,新条目加在最上面。
> 工程口径的完整记录见 `PROGRESS.md` / git log。 > 工程口径的完整记录见 `PROGRESS.md` / git log。
## 0.65.2 — 2026-08-13
- 粘贴、拖拽或选择附件发送后,对话中的用户消息会保留附件卡片;图片同时显示可点击放大的缩略图,刷新历史对话后也不会消失。
## 0.65.1 — 2026-08-12 ## 0.65.1 — 2026-08-12

View File

@ -175,6 +175,8 @@ Eval 与生产 core 解耦,通过现有 `/v1` API 创建专用任务、监听
**对话产物与生命周期(0025/0028)**:真实文件仍是内容事实源;`artifacts` 表记录已发布产物的稳定身份和生命周期,包含 user-root 相对当前路径、来源 task、复制来源、哈希/大小及 active/deleted、回收路径。新 `messages.artifact_refs` 使用 `{version:2, artifact_id, scope:"working_dir", path:"reports/a.pdf", label?:"最终报告"}``path` 是兼容快照,预览/下载优先按 `artifact_id` 找当前路径因此移动或重命名后历史卡片仍有效。version 1 和 `NULL` 旧消息继续走原 task-scoped 兼容链。普通源码树、中间文件和配套资源不登记agent 仅用 `publish_artifacts` 显式提升少量最终文件。移动保持身份,复制为每个副本创建新身份并记录直接来源;删除将文件移入 `.zcbot_artifact_trash/` 并软删记录,普通文件仍物理删除。 **对话产物与生命周期(0025/0028)**:真实文件仍是内容事实源;`artifacts` 表记录已发布产物的稳定身份和生命周期,包含 user-root 相对当前路径、来源 task、复制来源、哈希/大小及 active/deleted、回收路径。新 `messages.artifact_refs` 使用 `{version:2, artifact_id, scope:"working_dir", path:"reports/a.pdf", label?:"最终报告"}``path` 是兼容快照,预览/下载优先按 `artifact_id` 找当前路径因此移动或重命名后历史卡片仍有效。version 1 和 `NULL` 旧消息继续走原 task-scoped 兼容链。普通源码树、中间文件和配套资源不登记agent 仅用 `publish_artifacts` 显式提升少量最终文件。移动保持身份,复制为每个副本创建新身份并记录直接来源;删除将文件移入 `.zcbot_artifact_trash/` 并软删记录,普通文件仍物理删除。
**用户消息附件(0031)**:`messages.attachment_refs` 与助手产物分开表达输入附件,元素为 `{version:1, scope:"working_dir", path, label, kind, media_type, size_bytes}`;文件仍是事实源,不登记为已发布 artifact也不承诺独立生命周期。`payload.content` 只保存用户自然语言,模型上下文在内存中按附件类型补兼容提示,避免 UI 协议污染正文。`NULL` 表示旧客户端/旧历史,前端继续解析正文标记;`[]` 表示新消息明确无附件。独立 attachment 表暂缓,只有出现跨消息复用、稳定身份、版本快照或附件级审计需求时再抽象 `file_assets + message_attachments`
### 7.2 资源模型(/v1) ### 7.2 资源模型(/v1)
统一 `/v1` 前缀返 JSON;UI 由 platform 实现(§7.9),本地 dev SPA dogfood。要点(细节见 `web/app.py`): 统一 `/v1` 前缀返 JSON;UI 由 platform 实现(§7.9),本地 dev SPA dogfood。要点(细节见 `web/app.py`):
@ -222,8 +224,8 @@ tasks(task_id pk, user_id fk, name NOT NULL, auto_title_pending default false,
context_base_idx, -- 0019 §8.8 软重置窗口起点 context_base_idx, -- 0019 §8.8 软重置窗口起点
deleted_at, -- 0010 软删 deleted_at, -- 0010 软删
created_at, updated_at) created_at, updated_at)
messages(pk, task_id fk, idx, payload jsonb, artifact_refs jsonb null, -- 0025,task-relative UI 元数据 messages(pk, task_id fk, idx, payload jsonb, artifact_refs jsonb null, attachment_refs jsonb null,
tokens_in/out, model_profile, kind, -- kind=push 等 tokens_in/out, model_profile, kind, -- 0025/0031 UI 元数据;kind=push 等
unique(task_id, idx); gin(payload)) unique(task_id, idx); gin(payload))
usage_events(pk, user_id, task_id, message_id, kind, -- chat/image/video/vision/... 自由文本 usage_events(pk, user_id, task_id, message_id, kind, -- chat/image/video/vision/... 自由文本
model_profile, units jsonb, cost numeric, created_at) -- 多态用量,加媒体不动 schema model_profile, units jsonb, cost numeric, created_at) -- 多态用量,加媒体不动 schema

View File

@ -2,7 +2,7 @@
> 配合 `DESIGN.md`。本文件只记 phase 状态、决策偏差、文件量、下一步。每条 1-2 句:做了啥 + 关键判断;细节查 `git log` / `git diff` / `DESIGN §7.9` > 配合 `DESIGN.md`。本文件只记 phase 状态、决策偏差、文件量、下一步。每条 1-2 句:做了啥 + 关键判断;细节查 `git log` / `git diff` / `DESIGN §7.9`
最后更新:2026-08-12(Windows Node 列表与生命周期管理,bump 0.65.1) 最后更新:2026-08-13(用户消息结构化附件与图片预览,bump 0.65.2)
--- ---
@ -18,9 +18,12 @@
| §7 SaaS | DESIGN §7 路线 | 🟡 | A 事件流化 ✅;B 完工 ✅;D `/v1` JSON API ✅;D' 过渡 auth + dev SPA ✅;单活 run 锁 + cancel ✅;0004 schema 瘦身 ✅;入口归位 ✅;真 OIDC 待;**C Step 1-3 + 3d ✅(Executor + Docker 池 + DockerExecutor + fs 工具进容器)+ Step 5 部署前置对账 ✅ + 容器资源 yaml + 应用层磁盘配额 ✅ + dogfood 网络放开 + 容器内 pip/npm 源持久化 ✅**;**Step 4 完整 egress proxy + Step 3b PGID kill 协议延后到外部用户开放前**(还需 egress proxy + xfs project quota OS 层硬化,§7.5 落地清单 #2 #4)。 | | §7 SaaS | DESIGN §7 路线 | 🟡 | A 事件流化 ✅;B 完工 ✅;D `/v1` JSON API ✅;D' 过渡 auth + dev SPA ✅;单活 run 锁 + cancel ✅;0004 schema 瘦身 ✅;入口归位 ✅;真 OIDC 待;**C Step 1-3 + 3d ✅(Executor + Docker 池 + DockerExecutor + fs 工具进容器)+ Step 5 部署前置对账 ✅ + 容器资源 yaml + 应用层磁盘配额 ✅ + dogfood 网络放开 + 容器内 pip/npm 源持久化 ✅**;**Step 4 完整 egress proxy + Step 3b PGID kill 协议延后到外部用户开放前**(还需 egress proxy + xfs project quota OS 层硬化,§7.5 落地清单 #2 #4)。 |
--- ---
## 已完成关键能力 ## 已完成关键能力
### 2026-08-13
- **08-13 / 0.65.2 / 用户消息结构化附件 + 对话内图片预览**:新增 0031 `messages.attachment_refs`,新客户端将附件作为结构化字段发送,数据库正文仅保留用户自然语言;后端按 task working_dir 校验路径并在内存模型上下文中补附件提示,旧客户端与历史正文标记继续兼容。用户消息即时态和历史态统一展示附件 chip图片额外显示可点击缩略图相关 Python 34 项、全部前端 Node 26 项、Python/JavaScript 语法、Alembic 单 head 与 diff 检查通过,未连接或迁移生产 DB。
### 2026-08-12 ### 2026-08-12
- **08-12 / 0.65.1 / Windows Node 列表与生命周期管理**:管理后台按 10 秒轮询展示节点名称/ID、在线状态、系统与桌面会话、版本、能力和最近心跳并提供可恢复的禁用/启用与永久删除;禁用或删除在线节点先主动关闭 WebSocket删除后的本机身份必须清除并重新注册。生产实测补齐 nginx 节点 WebSocket Upgrade 专用 location 与故障说明,避免注册成功后长连接被普通 HTTP location 当作 GET。相关后端、鉴权路由和管理前端 45 项 unittest、Python 编译、JavaScript 语法与 diff 检查通过;无 schema、migration 或依赖变化,未连接或写入生产 DB。 - **08-12 / 0.65.1 / Windows Node 列表与生命周期管理**:管理后台按 10 秒轮询展示节点名称/ID、在线状态、系统与桌面会话、版本、能力和最近心跳并提供可恢复的禁用/启用与永久删除;禁用或删除在线节点先主动关闭 WebSocket删除后的本机身份必须清除并重新注册。生产实测补齐 nginx 节点 WebSocket Upgrade 专用 location 与故障说明,避免注册成功后长连接被普通 HTTP location 当作 GET。相关后端、鉴权路由和管理前端 45 项 unittest、Python 编译、JavaScript 语法与 diff 检查通过;无 schema、migration 或依赖变化,未连接或写入生产 DB。

5
RUN.md
View File

@ -152,6 +152,7 @@
- **PG**:`ZCBOT_DB_URL` 必填。本地 docker compose / 远端 dev / 生产任选;未设置时启动清晰报错,不引导 docker(§7.4)。 - **PG**:`ZCBOT_DB_URL` 必填。本地 docker compose / 远端 dev / 生产任选;未设置时启动清晰报错,不引导 docker(§7.4)。
- **OpenAPI / MCP 外部系统**:① `.env` 配置独立的 `ZCBOT_CREDENTIAL_MASTER_KEY`,可选 `ZCBOT_CREDENTIAL_KEY_ID` 标识当前密钥;轮换时把旧 key 以 JSON 对象放入 `ZCBOT_CREDENTIAL_PREVIOUS_KEYS`,待用户凭据完成重写后再移除。② 执行 `main.py db upgrade head`。③ admin 进入管理后台「外部系统」,选择通用 OpenAPI 或通用 MCP具体 MES/ERP/LIMS 都作为数据库 definition 配置,不新增专用 provider。MCP 填写与登录 Base URL 同源的 Streamable HTTP URL可选填写期望 Server 名称;连接后以 `tools/list` 为事实源。④ 普通用户点击左栏 **「外部」**,页面按 definition 动态显示用户名密码、API Key 或 Bearer Token目标、Server 身份、登录、认证绑定或 TLS 变化后保留密文并暂停调用重新测试成功后恢复。OpenAPI spec 和 MCP tool catalog 只在进程内按连接身份有界缓存,登录与业务响应均限长,普通用户和模型不能传任意 URL。 - **OpenAPI / MCP 外部系统**:① `.env` 配置独立的 `ZCBOT_CREDENTIAL_MASTER_KEY`,可选 `ZCBOT_CREDENTIAL_KEY_ID` 标识当前密钥;轮换时把旧 key 以 JSON 对象放入 `ZCBOT_CREDENTIAL_PREVIOUS_KEYS`,待用户凭据完成重写后再移除。② 执行 `main.py db upgrade head`。③ admin 进入管理后台「外部系统」,选择通用 OpenAPI 或通用 MCP具体 MES/ERP/LIMS 都作为数据库 definition 配置,不新增专用 provider。MCP 填写与登录 Base URL 同源的 Streamable HTTP URL可选填写期望 Server 名称;连接后以 `tools/list` 为事实源。④ 普通用户点击左栏 **「外部」**,页面按 definition 动态显示用户名密码、API Key 或 Bearer Token目标、Server 身份、登录、认证绑定或 TLS 变化后保留密文并暂停调用重新测试成功后恢复。OpenAPI spec 和 MCP tool catalog 只在进程内按连接身份有界缓存,登录与业务响应均限长,普通用户和模型不能传任意 URL。
- **Artifact 生命周期(0028)**:部署本版本必须先执行 `.venv/Scripts/python.exe main.py db upgrade head`。migration 会从存量 `messages.artifact_refs` 回填 active artifact 身份;删除后的已发布产物保存在用户根目录隐藏区 `.zcbot_artifact_trash/`,默认不自动清理且继续计入磁盘配额。普通文件删除语义不变。 - **Artifact 生命周期(0028)**:部署本版本必须先执行 `.venv/Scripts/python.exe main.py db upgrade head`。migration 会从存量 `messages.artifact_refs` 回填 active artifact 身份;删除后的已发布产物保存在用户根目录隐藏区 `.zcbot_artifact_trash/`,默认不自动清理且继续计入磁盘配额。普通文件删除语义不变。
- **用户消息结构化附件(0031)**:部署新版前执行 `.venv/Scripts/python.exe main.py db upgrade head`,为 `messages` 增加 nullable `attachment_refs`。migration 不回写或改动存量正文;旧客户端仍可只传 `content`,新版会发送结构化 `attachments`
- **旧 `factory_mes` definition 一次性转换**:新版代码不再识别 `factory_mes`;部署时保持旧服务进程运行,先从新代码目录执行数据脚本,转换成功后再重启到新版。脚本不加载 `.env`、不读取 `ZCBOT_DB_URL`,只认显式的 `ZCBOT_MIGRATION_DB_URL`;默认 dry-run检查同名冲突与配置合法性。确认输出后加 `--apply`,脚本把 definition 转为 `generic_openapi + query`、物化 JWT/提示/只读 POST 配置并同步 active connection revision不解密或改写用户凭据。 - **旧 `factory_mes` definition 一次性转换**:新版代码不再识别 `factory_mes`;部署时保持旧服务进程运行,先从新代码目录执行数据脚本,转换成功后再重启到新版。脚本不加载 `.env`、不读取 `ZCBOT_DB_URL`,只认显式的 `ZCBOT_MIGRATION_DB_URL`;默认 dry-run检查同名冲突与配置合法性。确认输出后加 `--apply`,脚本把 definition 转为 `generic_openapi + query`、物化 JWT/提示/只读 POST 配置并同步 active connection revision不解密或改写用户凭据。
```powershell ```powershell
$env:ZCBOT_MIGRATION_DB_URL="postgresql+psycopg://user:pass@host:5432/zcbot" $env:ZCBOT_MIGRATION_DB_URL="postgresql+psycopg://user:pass@host:5432/zcbot"
@ -361,8 +362,8 @@ $env:ZCBOT_EVAL_TOKEN = "<dedicated-eval-user-jwt>"
| `DELETE /v1/external-systems/{id}` | 清除当前用户连接及密文凭据;指定用户模式保留管理员授予的可见权 | 必填 | | `DELETE /v1/external-systems/{id}` | 清除当前用户连接及密文凭据;指定用户模式保留管理员授予的可见权 | 必填 |
| `GET/POST /v1/admin/external-system-definitions` | 管理员列出或新增可信外部系统目录 | admin | | `GET/POST /v1/admin/external-system-definitions` | 管理员列出或新增可信外部系统目录 | admin |
| `PUT/DELETE /v1/admin/external-system-definitions/{id}` | 管理员编辑、停用或删除目录项;已有用户连接时拒绝删除 | admin | | `PUT/DELETE /v1/admin/external-system-definitions/{id}` | 管理员编辑、停用或删除目录项;已有用户连接时拒绝删除 | admin |
| `GET /v1/tasks/{id}/messages` | LiteLLM payload 透传;0025 起每条另带 `artifact_refs``null`=旧消息、`[]`=新消息无产物、非空数组=相对该 task 当前 working_dir 的结构化产物引用 | 必填 | | `GET /v1/tasks/{id}/messages` | LiteLLM payload 透传;另带 `artifact_refs`(助手产物)与 `attachment_refs`(用户附件)。两者均以 `null` 表示旧消息、`[]` 表示新消息明确为空、非空数组表示 task-relative 结构化引用 | 必填 |
| `POST /v1/tasks/{id}/messages` | `{content, image_model?=""}` 发消息;返 `{events_url}`;**`run_status` 是 running/cancelling → 409**(单活 run;error 起新 run 时清);`image_model` 是 `config/media/doubao.yaml` image 段的 variant key(空 → 沿用 yaml 第一个),仅本 run 装配 SeedreamTool 时使用,不入 DB;UI 应 disable send 直到 SSE `done` | 必填 | | `POST /v1/tasks/{id}/messages` | `{content, attachments?:[{path,kind,label?}], image_model?=""}` 发消息;`attachments` 路径由后端按当前 task working_dir 校验并规范化,允许纯附件消息;旧客户端省略该字段继续兼容。返 `{events_url}`**`run_status` 是 running/cancelling → 409**UI 应 disable send 直到 SSE `done` | 必填 |
| `GET /v1/tasks/{id}/events` | SSE 流(`event: <type>` + `data: <json>`);订阅 task 当前活动 | 必填 | | `GET /v1/tasks/{id}/events` | SSE 流(`event: <type>` + `data: <json>`);订阅 task 当前活动 | 必填 |
| `POST /v1/tasks/{id}/cancel` | 协作式 cancel;`run_status != running` → 409;LLM 走 streaming,chunk 间 poll cancel — 延迟 100ms 级,基本秒退 | 必填 | | `POST /v1/tasks/{id}/cancel` | 协作式 cancel;`run_status != running` → 409;LLM 走 streaming,chunk 间 poll cancel — 延迟 100ms 级,基本秒退 | 必填 |
| `GET /v1/procs` | 当前用户全部后台进程(bg proc,§8.12;shell/run_python `background=true` 启动);纯文件系统读取,前端运行条 5s 轮询用 | 必填 | | `GET /v1/procs` | 当前用户全部后台进程(bg proc,§8.12;shell/run_python `background=true` 启动);纯文件系统读取,前端运行条 5s 轮询用 | 必填 |

View File

@ -1,3 +1,3 @@
# zcbot 版本号单一事实源:web/app.py 的 FastAPI version、/healthz 返回、前端展示都引这里。 # zcbot 版本号单一事实源:web/app.py 的 FastAPI version、/healthz 返回、前端展示都引这里。
# 改版本只动这一行。 # 改版本只动这一行。
__version__ = "0.65.1" __version__ = "0.65.2"

76
core/attachments.py Normal file
View File

@ -0,0 +1,76 @@
"""Structured user-message attachments and model-context compatibility rendering."""
from __future__ import annotations
import mimetypes
from collections.abc import Iterable
from pathlib import Path
from typing import Any
from .artifacts import ArtifactPathError, resolve_artifact_path
ATTACHMENT_REF_VERSION = 1
MAX_ATTACHMENTS_PER_MESSAGE = 10
_IMAGE_EXTENSIONS = {
".avif", ".bmp", ".gif", ".heic", ".heif", ".jpeg", ".jpg",
".png", ".svg", ".tif", ".tiff", ".webp",
}
def _attachment_kind(requested: str, path: Path, media_type: str) -> str:
detected_image = media_type.startswith("image/") or path.suffix.lower() in _IMAGE_EXTENSIONS
if requested == "image" and detected_image:
return "image"
return "file"
def normalize_attachment_refs(
refs: Iterable[Any], *, working_dir: Path, user_root: Path,
) -> list[dict]:
"""Validate client refs and return task-relative, display-ready metadata."""
output: list[dict] = []
seen: set[str] = set()
for raw in refs:
if len(output) >= MAX_ATTACHMENTS_PER_MESSAGE:
raise ArtifactPathError(
f"at most {MAX_ATTACHMENTS_PER_MESSAGE} attachments are allowed"
)
data = raw.model_dump() if hasattr(raw, "model_dump") else dict(raw or {})
absolute, rel = resolve_artifact_path(
str(data.get("path") or ""),
working_dir=working_dir,
user_root=user_root,
require_file=True,
)
if rel in seen:
continue
seen.add(rel)
media_type = mimetypes.guess_type(absolute.name)[0] or "application/octet-stream"
label = str(data.get("label") or absolute.name).strip() or absolute.name
output.append({
"version": ATTACHMENT_REF_VERSION,
"scope": "working_dir",
"path": rel,
"label": label,
"kind": _attachment_kind(str(data.get("kind") or "file"), absolute, media_type),
"media_type": media_type,
"size_bytes": absolute.stat().st_size,
})
return output
def content_for_model(content: str, refs: Iterable[dict], working_dir_name: str) -> str:
"""Append the legacy-readable file hints only to the provider-bound message."""
lines: list[str] = []
wd = str(working_dir_name or "").strip().strip("/\\")
for ref in refs or ():
path = str((ref or {}).get("path") or "").strip().replace("\\", "/")
if not path:
continue
full_path = f"{wd}/{path}" if wd else path
marker = "[用户上传的参考图]" if (ref or {}).get("kind") == "image" else "[用户上传的文件]"
lines.append(f"{marker} {full_path}")
text = str(content or "").strip()
if not lines:
return text
suffix = "\n".join(lines)
return f"{text}\n\n{suffix}" if text else suffix

View File

@ -226,6 +226,16 @@ class Session:
).scalars().all() ).scalars().all()
for row in rows: for row in rows:
payload = dict(row.payload) payload = dict(row.payload)
attachment_refs = getattr(row, "attachment_refs", None)
if payload.get("role") == "user" and attachment_refs:
from .attachments import content_for_model
working_dir = str((sess.meta or {}).get("working_dir") or "")
wd_name = working_dir.replace("\\", "/").rstrip("/").rsplit("/", 1)[-1]
payload["content"] = content_for_model(
str(payload.get("content") or ""),
attachment_refs,
wd_name,
)
if payload.get("role") == "assistant" and isinstance(payload.get("content"), str): if payload.get("role") == "assistant" and isinstance(payload.get("content"), str):
# 历史行不回写生产库;只在重建 LLM 上下文时应用同一窄修复, # 历史行不回写生产库;只在重建 LLM 上下文时应用同一窄修复,
# 与 Web 展示层保持一致,避免旧坏围栏继续污染后续轮次。 # 与 Web 展示层保持一致,避免旧坏围栏继续污染后续轮次。

View File

@ -192,6 +192,10 @@ class Message(Base):
# fall back to path extraction); [] means a new message explicitly published no artifacts. # fall back to path extraction); [] means a new message explicitly published no artifacts.
# Kept outside payload so provider-bound conversation messages remain protocol-clean. # Kept outside payload so provider-bound conversation messages remain protocol-clean.
artifact_refs: Mapped[Optional[list[dict[str, Any]]]] = mapped_column(JSONB, nullable=True) artifact_refs: Mapped[Optional[list[dict[str, Any]]]] = mapped_column(JSONB, nullable=True)
# Structured user input attachments. NULL means legacy; [] means explicitly none.
attachment_refs: Mapped[Optional[list[dict[str, Any]]]] = mapped_column(
JSONB, nullable=True
)
tokens_in: Mapped[Optional[int]] = mapped_column(Integer, nullable=True) tokens_in: Mapped[Optional[int]] = mapped_column(Integer, nullable=True)
tokens_out: Mapped[Optional[int]] = mapped_column(Integer, nullable=True) tokens_out: Mapped[Optional[int]] = mapped_column(Integer, nullable=True)
# 0006:产生该 message 的模型(只在 assistant 行有值;user/tool/system 为 NULL)。 # 0006:产生该 message 的模型(只在 assistant 行有值;user/tool/system 为 NULL)。

View File

@ -0,0 +1,27 @@
"""Add structured user attachment references to messages.
Revision ID: 0031
Revises: 0030
Create Date: 2026-08-13
"""
from collections.abc import Sequence
import sqlalchemy as sa
from alembic import op
from sqlalchemy.dialects import postgresql
revision: str = "0031"
down_revision: str | None = "0030"
branch_labels: str | Sequence[str] | None = None
depends_on: str | Sequence[str] | None = None
def upgrade() -> None:
op.add_column(
"messages",
sa.Column("attachment_refs", postgresql.JSONB(astext_type=sa.Text()), nullable=True),
)
def downgrade() -> None:
op.drop_column("messages", "attachment_refs")

View File

@ -0,0 +1,36 @@
import assert from "node:assert/strict";
import test from "node:test";
import { readFileSync } from "node:fs";
import { parseUserAttachments } from "../web/static/js/attachments.js";
const chatJs = readFileSync("web/static/js/chat.js", "utf8");
test("user attachment protocol lines are separated from visible message text", () => {
const parsed = parseUserAttachments(
"请比较这张图\n\n[用户上传的参考图] 项目/显微照片.png\n[用户上传的文件] 项目/数据表.xlsx",
);
assert.equal(parsed.content, "请比较这张图");
assert.deepEqual(parsed.attachments, [
{ path: "项目/显微照片.png", kind: "image" },
{ path: "项目/数据表.xlsx", kind: "file" },
]);
});
test("attachment-only messages have an empty visible body", () => {
const parsed = parseUserAttachments("[用户上传的参考图] 项目/照片 01.png");
assert.equal(parsed.content, "");
assert.deepEqual(parsed.attachments, [{ path: "项目/照片 01.png", kind: "image" }]);
});
test("ordinary bracketed text remains visible", () => {
const content = "[备注] 这不是附件\n继续处理";
assert.deepEqual(parseUserAttachments(content), { content, attachments: [] });
});
test("new sends use structured attachments and history keeps a legacy fallback", () => {
assert.match(chatJs, /attachments:\s*pendingAttachments\.map/);
assert.match(chatJs, /Array\.isArray\(m\.attachment_refs\)/);
assert.match(chatJs, /structuredAttachments === null/);
assert.doesNotMatch(chatJs, /content = content \? `\$\{content\}\\n\\n\$\{lines\}`/);
});

49
tests/test_attachments.py Normal file
View File

@ -0,0 +1,49 @@
import tempfile
import unittest
from pathlib import Path
from core.artifacts import ArtifactPathError
from core.attachments import content_for_model, normalize_attachment_refs
class AttachmentRefTests(unittest.TestCase):
def test_normalizes_user_relative_path_to_task_relative_metadata(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp)
wd = root / "实验分析"
wd.mkdir()
image = wd / "显微照片.png"
image.write_bytes(b"png")
refs = normalize_attachment_refs(
[{"path": "实验分析/显微照片.png", "kind": "image"}],
working_dir=wd,
user_root=root,
)
self.assertEqual(refs[0]["path"], "显微照片.png")
self.assertEqual(refs[0]["kind"], "image")
self.assertEqual(refs[0]["size_bytes"], 3)
self.assertEqual(refs[0]["scope"], "working_dir")
def test_rejects_attachment_outside_working_dir(self):
with tempfile.TemporaryDirectory() as tmp:
root = Path(tmp)
wd = root / "任务"
wd.mkdir()
(root / "outside.txt").write_text("x", encoding="utf-8")
with self.assertRaises(ArtifactPathError):
normalize_attachment_refs(
[{"path": "../outside.txt", "kind": "file"}],
working_dir=wd,
user_root=root,
)
def test_builds_provider_content_without_changing_stored_text(self):
refs = [{"path": "figures/a.png", "kind": "image"}]
self.assertEqual(
content_for_model("请分析", refs, "实验分析"),
"请分析\n\n[用户上传的参考图] 实验分析/figures/a.png",
)
self.assertEqual(
content_for_model("", refs, "实验分析"),
"[用户上传的参考图] 实验分析/figures/a.png",
)

View File

@ -0,0 +1,27 @@
import unittest
from pydantic import ValidationError
from web.schemas import MessageRequest
class MessageAttachmentSchemaTests(unittest.TestCase):
def test_omitted_attachments_remains_distinguishable_for_legacy_clients(self):
body = MessageRequest(content="旧客户端正文")
self.assertEqual(body.attachments, [])
self.assertNotIn("attachments", body.model_fields_set)
def test_explicit_structured_attachments_are_accepted(self):
body = MessageRequest(
content="分析图片",
attachments=[{"path": "任务/a.png", "kind": "image"}],
)
self.assertIn("attachments", body.model_fields_set)
self.assertEqual(body.attachments[0].path, "任务/a.png")
def test_attachment_count_is_limited(self):
with self.assertRaises(ValidationError):
MessageRequest(
content="太多",
attachments=[{"path": f"任务/{i}.txt"} for i in range(11)],
)

View File

@ -38,6 +38,7 @@ class ClaimRunTests(unittest.TestCase):
with patch.object(run_lifecycle, "session_scope", fake_scope): with patch.object(run_lifecycle, "session_scope", fake_scope):
claim = run_lifecycle.claim_run_with_message( claim = run_lifecycle.claim_run_with_message(
tid, uid, "持久化消息", prepare=prepare, tid, uid, "持久化消息", prepare=prepare,
attachment_refs=[{"path": "a.png", "kind": "image"}],
) )
self.assertEqual(claim.metadata, {"marker": "ok"}) self.assertEqual(claim.metadata, {"marker": "ok"})
@ -48,6 +49,7 @@ class ClaimRunTests(unittest.TestCase):
added.payload, added.payload,
{"role": "user", "content": "持久化消息"}, {"role": "user", "content": "持久化消息"},
) )
self.assertEqual(added.attachment_refs, [{"path": "a.png", "kind": "image"}])
update_stmt = session.execute.call_args_list[-1].args[0] update_stmt = session.execute.call_args_list[-1].args[0]
params = update_stmt.compile().params params = update_stmt.compile().params
self.assertEqual(params["run_status"], "running") self.assertEqual(params["run_status"], "running")

View File

@ -123,7 +123,8 @@ class StaticVendorTests(unittest.TestCase):
self.assertIn(".composer-tool .composer-label", css) self.assertIn(".composer-tool .composer-label", css)
self.assertIn('aria-label="添加附件"', html) self.assertIn('aria-label="添加附件"', html)
self.assertIn("await ensureTaskFromDraft();", frontend) self.assertIn("await ensureTaskFromDraft();", frontend)
self.assertIn('"[用户上传的文件]"', frontend) self.assertIn("attachments: pendingAttachments.map", frontend)
self.assertIn("parseUserAttachments", frontend)
self.assertNotIn("发送第一条消息创建对话后即可添加附件", frontend) self.assertNotIn("发送第一条消息创建对话后即可添加附件", frontend)
self.assertIn('aria-label="语音输入"', html) self.assertIn('aria-label="语音输入"', html)
self.assertIn('aria-label="润色提示词"', html) self.assertIn('aria-label="润色提示词"', html)

View File

@ -100,6 +100,7 @@ def register_message_routes(app, *, require_user) -> None:
Message.idx, Message.payload, Message.tokens_in, Message.idx, Message.payload, Message.tokens_in,
Message.tokens_out, Message.model_profile, Message.created_at, Message.tokens_out, Message.model_profile, Message.created_at,
Message.artifact_refs, Message.artifact_refs,
Message.attachment_refs,
) )
if limit is None: if limit is None:
# 旧行为:升序全量 # 旧行为:升序全量
@ -148,6 +149,7 @@ def register_message_routes(app, *, require_user) -> None:
"model_profile": r.model_profile, # 0006:assistant 行非空,标产生该 msg 的模型 "model_profile": r.model_profile, # 0006:assistant 行非空,标产生该 msg 的模型
"created_at": iso(r.created_at), "created_at": iso(r.created_at),
"artifact_refs": r.artifact_refs, "artifact_refs": r.artifact_refs,
"attachment_refs": r.attachment_refs,
} }
for r in rows for r in rows
] ]
@ -205,7 +207,7 @@ def register_message_routes(app, *, require_user) -> None:
headers={"Retry-After": "3"}, headers={"Retry-After": "3"},
) )
content = (body.content or "").strip() content = (body.content or "").strip()
if not content: if not content and not body.attachments:
raise HTTPException(400, "empty content") raise HTTPException(400, "empty content")
# 快捷指令展开(与渠道入口共用 core/shortcuts.py):整条精确命中触发词 → 换成完整 # 快捷指令展开(与渠道入口共用 core/shortcuts.py):整条精确命中触发词 → 换成完整
# 指令。在起 run 之前、落库之前展开,模型看到的就是完整指令(不进上下文、不问模型)。 # 指令。在起 run 之前、落库之前展开,模型看到的就是完整指令(不进上下文、不问模型)。
@ -219,6 +221,31 @@ def register_message_routes(app, *, require_user) -> None:
# running 后才发现参数非法,会留下一个实际上没有 worker 的假活跃任务。 # running 后才发现参数非法,会留下一个实际上没有 worker 的假活跃任务。
image_variant = resolve_image_model(body.image_model, user_id=user_id) image_variant = resolve_image_model(body.image_model, user_id=user_id)
video_variant = resolve_video_model(body.video_model, user_id=user_id) video_variant = resolve_video_model(body.video_model, user_id=user_id)
# Omitted means a legacy client; explicit [] means a new message has no attachments.
attachments_supplied = "attachments" in body.model_fields_set
attachment_refs: Optional[list[dict]] = [] if attachments_supplied else None
if body.attachments:
from core.artifacts import ArtifactPathError
from core.attachments import normalize_attachment_refs
from core.paths import from_db_path
root = load_user_root(user_id)
with session_scope() as s:
task_working_dir = s.execute(
select(Task.working_dir).where(
Task.task_id == tid, Task.user_id == user_id,
)
).scalar_one_or_none()
if not task_working_dir:
raise HTTPException(404, f"task not found: {tid}")
working_dir = from_db_path(task_working_dir).resolve()
try:
attachment_refs = normalize_attachment_refs(
body.attachments,
working_dir=working_dir,
user_root=root,
)
except ArtifactPathError as e:
raise HTTPException(400, str(e))
def _prepare_claim(s, task): def _prepare_claim(s, task):
values: dict = {} values: dict = {}
# 档位门控:存量 task 的模型已不在用户档位内(如管理员下调了档位)→ 本次起 # 档位门控:存量 task 的模型已不在用户档位内(如管理员下调了档位)→ 本次起
@ -248,6 +275,7 @@ def register_message_routes(app, *, require_user) -> None:
try: try:
claim = claim_run_with_message( claim = claim_run_with_message(
tid, user_id, content, prepare=_prepare_claim, tid, user_id, content, prepare=_prepare_claim,
attachment_refs=attachment_refs,
) )
except RunTaskNotFound: except RunTaskNotFound:
raise HTTPException(404, f"task not found: {tid}") raise HTTPException(404, f"task not found: {tid}")
@ -276,7 +304,7 @@ def register_message_routes(app, *, require_user) -> None:
should_auto_title = claim.metadata["should_auto_title"] should_auto_title = claim.metadata["should_auto_title"]
if should_auto_title: if should_auto_title:
from core.task_title import is_attachment_only_message from core.task_title import is_attachment_only_message
should_auto_title = not is_attachment_only_message(content) should_auto_title = bool(content) and not is_attachment_only_message(content)
# 快速入口只在首条消息时 pending=true。辅助标题与主 run 并行,不阻塞 # 快速入口只在首条消息时 pending=true。辅助标题与主 run 并行,不阻塞
# TTFT/SSE纯附件消息保留 pending等下一条自然语言再命名避免标题暴露 # TTFT/SSE纯附件消息保留 pending等下一条自然语言再命名避免标题暴露
# `[用户上传的文件]` 等内部标记。generate_task_title 内再次查闸并用条件 UPDATE # `[用户上传的文件]` 等内部标记。generate_task_title 内再次查闸并用条件 UPDATE

View File

@ -53,6 +53,7 @@ def claim_run_with_message(
user_message: str, user_message: str,
*, *,
prepare: Optional[PrepareClaim] = None, prepare: Optional[PrepareClaim] = None,
attachment_refs: Optional[list[dict[str, Any]]] = None,
) -> RunClaim: ) -> RunClaim:
"""在 task 行锁下原子提交 user 消息和 `run_status=running`。 """在 task 行锁下原子提交 user 消息和 `run_status=running`。
@ -82,6 +83,7 @@ def claim_run_with_message(
task_id=task_id, task_id=task_id,
idx=int(next_idx), idx=int(next_idx),
payload={"role": "user", "content": user_message}, payload={"role": "user", "content": user_message},
attachment_refs=attachment_refs,
)) ))
values = { values = {
"run_status": "running", "run_status": "running",

View File

@ -33,8 +33,15 @@ class SchedulePatchRequest(BaseModel):
enabled: Optional[bool] = None enabled: Optional[bool] = None
class MessageAttachmentRequest(BaseModel):
path: str
kind: str = "file"
label: str = ""
class MessageRequest(BaseModel): class MessageRequest(BaseModel):
content: str content: str
attachments: list[MessageAttachmentRequest] = Field(default_factory=list, max_length=10)
# 该条消息触发的生图 / 生视频模型 variant key(config/media/doubao.yaml image/video 段)。 # 该条消息触发的生图 / 生视频模型 variant key(config/media/doubao.yaml image/video 段)。
# 空 → 对应 tool 走 yaml 第一个 variant;非空 → 本次 run 装配指定 variant。 # 空 → 对应 tool 走 yaml 第一个 variant;非空 → 本次 run 装配指定 variant。
# 仅作用于本 run,不入 DB,UI 下拉的选择跟在消息 POST body 上。 # 仅作用于本 run,不入 DB,UI 下拉的选择跟在消息 POST body 上。

View File

@ -1132,6 +1132,14 @@
width: auto; height: auto; cursor: zoom-in; width: auto; height: auto; cursor: zoom-in;
} }
.art-media-video { flex: 0 1 720px; width: min(100%, 720px); } .art-media-video { flex: 0 1 720px; width: min(100%, 720px); }
/* 用户发出的图片先以紧凑缩略图确认内容,下面仍保留统一附件 chip。 */
.user-attachments { margin-top: 6px; }
.user-attachments .artifact-bar { margin-top: 4px; }
.user-attachments .art-media-image { border-radius: 8px; }
.user-attachments .art-media-image img {
width: auto; height: auto; max-width: min(240px, 100%); max-height: 180px;
object-fit: contain;
}
.art-media video { .art-media video {
display: block; width: 100%; height: auto; max-height: 405px; background: #000; display: block; width: 100%; height: auto; max-height: 405px; background: #000;
} }

View File

@ -0,0 +1,19 @@
// 新消息使用结构化 attachment_refs这里只解析旧客户端与存量消息的正文标记。
const USER_ATTACHMENT_RE = /^\[(用户上传的参考图|用户上传的文件)\]\s+(.+?)\s*$/;
export function parseUserAttachments(content) {
const attachments = [];
const bodyLines = [];
for (const line of String(content || "").split("\n")) {
const match = line.match(USER_ATTACHMENT_RE);
if (!match) {
bodyLines.push(line);
continue;
}
attachments.push({
path: match[2],
kind: match[1] === "用户上传的参考图" ? "image" : "file",
});
}
return { content: bodyLines.join("\n").trim(), attachments };
}

View File

@ -33,6 +33,7 @@ import {
} from "./preview.js"; } from "./preview.js";
import { loadFiles, scheduleFilesRefresh, uploadFiles, formatUploadProgress } from "./files.js"; import { loadFiles, scheduleFilesRefresh, uploadFiles, formatUploadProgress } from "./files.js";
import { toolActivityLabel, _workingDirName, extractMediaBanner, extractArtifactRels, renderArtifactBarHtml, renderArtifactChipContent, upgradeMediaArtifacts, ARTIFACT_PRODUCING_TOOLS, _flushMediaArtifactCache } from "./media.js"; import { toolActivityLabel, _workingDirName, extractMediaBanner, extractArtifactRels, renderArtifactBarHtml, renderArtifactChipContent, upgradeMediaArtifacts, ARTIFACT_PRODUCING_TOOLS, _flushMediaArtifactCache } from "./media.js";
import { parseUserAttachments } from "./attachments.js";
import { applyProgressAction, cloneProgressSteps, progressActionsFromToolCalls } from "./progress.js"; import { applyProgressAction, cloneProgressSteps, progressActionsFromToolCalls } from "./progress.js";
import { refreshProcs, decorateBgprocCard, hasRunningProc, killTaskProcs } from "./procs.js"; import { refreshProcs, decorateBgprocCard, hasRunningProc, killTaskProcs } from "./procs.js";
@ -1708,8 +1709,13 @@ function renderMessages(msgs, { stickBottom = true } = {}) {
hasVisibleAssistantContent = true; hasVisibleAssistantContent = true;
} }
} }
if (typeof p.content === "string" && p.content) { const structuredAttachments = role === "user" && Array.isArray(m.attachment_refs)
html += `<div class="body">${renderMd(p.content)}</div>`; ? m.attachment_refs : null;
const userParts = role === "user" && structuredAttachments === null
? parseUserAttachments(p.content) : null;
const visibleContent = userParts ? userParts.content : p.content;
if (typeof visibleContent === "string" && visibleContent) {
html += `<div class="body">${renderMd(visibleContent)}</div>`;
if (role === "assistant") hasVisibleAssistantContent = true; if (role === "assistant") hasVisibleAssistantContent = true;
// assistant 正文里 echo 的 <wd>/... 路径**永远**展示(绕开 seenRels)。图片/视频 // assistant 正文里 echo 的 <wd>/... 路径**永远**展示(绕开 seenRels)。图片/视频
// 已可能在产物工具结果中内联,仍用 chip 防重复HTML 通常由 write/shell 产出, // 已可能在产物工具结果中内联,仍用 chip 防重复HTML 通常由 write/shell 产出,
@ -1719,6 +1725,12 @@ function renderMessages(msgs, { stickBottom = true } = {}) {
html += renderArtifactBarHtml(extractArtifactRels(p.content, wd), "html", state.taskId || "", true); html += renderArtifactBarHtml(extractArtifactRels(p.content, wd), "html", state.taskId || "", true);
} }
} }
const visibleAttachments = structuredAttachments || (userParts && userParts.attachments) || [];
if (visibleAttachments.length) {
html += renderUserAttachmentsHtml(
visibleAttachments, state.taskId || "", structuredAttachments === null,
);
}
if (role === "assistant" && Array.isArray(m.artifact_refs)) { if (role === "assistant" && Array.isArray(m.artifact_refs)) {
html += renderArtifactBarHtml(m.artifact_refs, true, state.taskId || ""); html += renderArtifactBarHtml(m.artifact_refs, true, state.taskId || "");
if (m.artifact_refs.length) hasVisibleAssistantContent = true; if (m.artifact_refs.length) hasVisibleAssistantContent = true;
@ -1784,6 +1796,7 @@ function renderMessages(msgs, { stickBottom = true } = {}) {
wrap.appendChild(sb); wrap.appendChild(sb);
_msgScrollObserver.observe(sb); _msgScrollObserver.observe(sb);
} }
if (stickBottom) wrap.scrollTop = wrap.scrollHeight; if (stickBottom) wrap.scrollTop = wrap.scrollHeight;
setTaskProgress(state.taskId, currentProgressSteps); setTaskProgress(state.taskId, currentProgressSteps);
upgradeMediaArtifacts(wrap); upgradeMediaArtifacts(wrap);
@ -1791,6 +1804,18 @@ function renderMessages(msgs, { stickBottom = true } = {}) {
renderLiveRunIfVisible(); renderLiveRunIfVisible();
} }
// 用户附件与助手产物语义不同:每个附件都保留可辨认、可点击的 chip图片另外在
// chip 上方给缩略图,兼顾快速确认内容和统一的预览/下载入口。
function renderUserAttachmentsHtml(attachments, taskId = "", legacy = false) {
if (!attachments || !attachments.length) return "";
const images = attachments.filter((item) => item.kind === "image");
const refs = attachments.map((item) => ({ path: item.path, label: item.label || "" }));
return `<div class="user-attachments">`
+ (images.length ? renderArtifactBarHtml(images, true, taskId, legacy) : "")
+ renderArtifactBarHtml(refs, false, taskId, legacy)
+ `</div>`;
}
// ───── send + SSE ───── // ───── send + SSE ─────
// 发送 / 停止 单按钮:idle → 发送(primary 红实心);streaming → 停止(danger 红边); // 发送 / 停止 单按钮:idle → 发送(primary 红实心);streaming → 停止(danger 红边);
// cancelling 是过渡态 — 用户点过停止后到 SSE 收到 cancelled/done 之间。 // cancelling 是过渡态 — 用户点过停止后到 SSE 收到 cancelled/done 之间。
@ -2661,16 +2686,9 @@ async function sendMessage(overrideText) {
$("chat-hint").textContent = `${_chCfg.label}对话请在${_chCfg.label}里进行 — web 端为只读镜像`; $("chat-hint").textContent = `${_chCfg.label}对话请在${_chCfg.label}里进行 — web 端为只读镜像`;
return; return;
} }
// 附件路径按类型注入正文:图片走视觉/改图工具,其他文件走文档读取、音频转写等工具。
const pendingAttachments = fromInput ? takePendingAttachments() : []; const pendingAttachments = fromInput ? takePendingAttachments() : [];
const attachmentOnly = !content && pendingAttachments.length > 0; const attachmentOnly = !content && pendingAttachments.length > 0;
if (pendingAttachments.length) { if (!content && !pendingAttachments.length) return;
const lines = pendingAttachments.map(({ rel, kind }) =>
`${kind === "image" ? "[用户上传的参考图]" : "[用户上传的文件]"} ${rel}`
).join("\n");
content = content ? `${content}\n\n${lines}` : lines;
}
if (!content) return;
setActionMode("cancelling"); // 临时锁住,等 events_url 拿到再切 streaming setActionMode("cancelling"); // 临时锁住,等 events_url 拿到再切 streaming
$("chat-hint").textContent = "发送中…"; $("chat-hint").textContent = "发送中…";
const taskId = state.taskId; const taskId = state.taskId;
@ -2680,8 +2698,13 @@ async function sendMessage(overrideText) {
const wrap = $("chat-stream"); const wrap = $("chat-stream");
const userCard = document.createElement("div"); const userCard = document.createElement("div");
userCard.className = "msg user"; userCard.className = "msg user";
userCard.innerHTML = `<div class="role">我</div><div class="body">${escapeHtml(content)}</div>`; userCard.innerHTML = `<div class="role">我</div>`
+ (content ? `<div class="body">${escapeHtml(content)}</div>` : "")
+ renderUserAttachmentsHtml(
pendingAttachments.map(({ rel, kind }) => ({ path: rel, kind })), taskId, true,
);
wrap.appendChild(userCard); wrap.appendChild(userCard);
upgradeMediaArtifacts(userCard);
// assistant 流式占位卡 // assistant 流式占位卡
const asstCard = document.createElement("div"); const asstCard = document.createElement("div");
@ -2692,6 +2715,10 @@ async function sendMessage(overrideText) {
const r = await postMessageWithRetry(taskId, { const r = await postMessageWithRetry(taskId, {
content, content,
attachments: pendingAttachments.map(({ rel, kind }) => ({
path: rel,
kind,
})),
image_model: state.imageModel || "", image_model: state.imageModel || "",
video_model: state.videoModel || "", video_model: state.videoModel || "",
}); });