ai_co_discuss 0.1.10 → 0.1.11

package/.agents/skills/discuss-host/references/launch.md

@@ -24,7 +24,7 @@
     "boundaries": "讨论边界",
     "host": "Codex",
     "experts": { "Codex": "cli", "Claude": "cli" },
-    "limits": { "max_rounds": 3, "max_minutes": 60, "execution_minutes": 10, "on_limit": "pause", "summary_minutes": 5 }
+    "limits": { "max_rounds": 3, "max_minutes": 60, "execution_minutes": 0, "on_limit": "pause", "summary_minutes": 5 }
   },
   "participants": {
     "Codex": { "adapter": "codex" }
@@ -32,7 +32,7 @@
 }
 ```
 
-此示例的主持在 Codex App；若当前是独立 Codex CLI，应使用 `codex_cli`。可执行文件、App 来源聊天或 Sidecar 等缺项仍须来自实际本机配置。主持也须出现在 `experts` 中并设为 `cli`。示例中的上限不能覆盖用户已经指定的参数；总时长为 0 时，另用顶层 `minutes` 指定本次有限运行时长。
+此示例的主持在 Codex App；若当前是独立 Codex CLI，应使用 `codex_cli`。可执行文件、App 来源聊天或 Sidecar 等缺项仍须来自实际本机配置。主持也须出现在 `experts` 中并设为 `cli`。示例中的上限不能覆盖用户已经指定的参数；总时长为 0 时可不限总时长运行，按轮次上限或手动停止结束。单次执行默认 0（不限），主动设置的单次执行超时仍有效。只有用户另设本次时长时才提供顶层 `minutes`，不能用它延长房间既定截止。
 
 已有房间改为指定顶层 `room_id`，省略 `room`；受保护房间另提供原私有配置的 `access_file` 路径，不读取或打印令牌。首次启动命令不替代已暂停房间的继续操作，已有运行按当前房间状态接续。
 

package/.agents/skills/discuss-participant/SKILL.md

@@ -42,6 +42,8 @@ description: "AI CoDiscuss 的通用参与技能，所有参与者（包括主
 
 增量读取用实际已读编号 `message list ROOM --after ID`，按 `next_after` / `has_more` 续读；`latest_id` 不等于已读。筛选后的游标只代表该查询范围，不能用来声称已读全场。阶段和安排以最新服务端状态为准。
 
+消息的 `attachments` 提供文件名、类型、大小与编号；需要核对内容时用 `attachment view ROOM ID`、`attachment download ROOM ID --dest 明确的新文件路径`，再用当前软件的实际工具读取。独立阶段仍遵循上述材料边界；下载或看见文件名不等于已读，无法解析时如实说明。
+
 ## 发言权与提交边界
 
 专家只回答属于自己的当前未完成安排（pending/notified），使用真实 `--turn`；没有安排就等待，不自行接管主持。提交前核对阶段、安排和上限。暂停、收束、`limits.reached` 或 `limits.closing` 时不新增专家观点，保留草稿；主持人按主持技能处理到限或总结。
@@ -56,6 +58,8 @@ description: "AI CoDiscuss 的通用参与技能，所有参与者（包括主
 
 观点用 `message post`，摘要用 `--summary` / `--summary-file`，完整正文优先 UTF-8 `--body-file`；自己的观点署名自己。主持消息用 `moderation`。代发保留真实作者、完整原文及引用关系，摘要归实际撰写者，不把自己的改写冒充对方回答。
 
+需要提交原文件时，先 `attachment upload ROOM --as 本人 --file FILE --submission-id 上传编号`，再把回执中的附件 ID 通过 `message post --attachment ID` 关联，可重复指定（最多 10 个，单个 20 MiB）。受保护上传带本次 `--access-file`，须在回答完成、执行资格结束前上传。上传和消息分别保存提交编号；结果不明用 `attachment list ROOM --submission-id 上传编号` 核对，重试沿用原文件与原编号。
+
 创建房间、安排、消息前保存 `--submission-id` 与原内容；同次重试沿用，新内容换新编号。成功须核对服务端返回的持久记录、作者和关联；返回不含所需证据时按编号补读，避免重复获取同一份已确认内容。答案落库后只读核验即可结束，不补启停回执或使用旧资格继续写入。
 
 结果不明先按原 `submission_id` 精确查询，找到就沿用；明确可重试时仅用原编号原内容重试一次，再不明则保留草稿并报告。不换身份、职责或编号绕过拒绝。首次使用某类写入、代发、超长正文或遇到失败时，读取 [写入与恢复](references/writing.md) 的相关部分；不把全部故障流程作为每次发言的检查表。

package/README.md

@@ -63,7 +63,7 @@ codiscuss setup
 
 1. 安装完成后进入自动打开的网站；以后运行 `codiscuss local open` 打开或复用本机网站，默认地址为 `http://127.0.0.1:8765`。
 2. 点击“新建讨论”，填写**话题名称**和**用户原文**。把问题、背景与顾虑写在原文中，讨论目标、背景和边界由 AI 主持人据此补全。
-3. 选择“全自动讨论”，勾选本场专家并从中指定 AI 主持人；分别设置轮次和时长上限，点击数值展开滚轮或直接输入。新建默认 **3 轮 / 不限时长**；网页启动全自动讨论前需设置有限时长。
+3. 选择“全自动讨论”，点击“邀请 AI”选择本场专家，并从中指定 AI 主持人；参与区只显示已加入的 AI，点击“移出”调整名单。已配置的 CLI 专家点击卡片正面，在背面配置模型与思考强度，留空使用 CLI 默认值。分别设置轮次和时长上限，点击数值展开滚轮或直接输入。新建默认 **3 轮 / 不限时长**，可直接启动全自动讨论，按轮次上限或手动停止结束；单次 AI 执行也默认不限时，可按需设置 1–60 分钟上限。
 4. 勾选“创建后立即启动 AI”，复核软件配置。需要修改时展开“调整启动配置”；未配置的软件需选择 CLI / App 类型并补齐路径。项目目录、读写权限和已有会话按本场需要填写。
 5. 点击“创建并启动讨论”。程序完成建房、登记与首次任务交付；在进度看板核对 AI 是否实际进入，并查看保存的发言。
 
@@ -105,7 +105,7 @@ codiscuss local connect https://你的站点
 
 下次打开会保留原讨论，已暂停的房间需在进度看板明确重启。网页已连接或创建成功不代表 AI 已回答，应核对实际执行进展与讨论记录。独立服务管理、停止归属及故障排查见 [接入说明](docs/runner.md#常驻讨论服务)；私有配置需保留在原位置。
 
-软件失败时先查看暂停原因，登录和权限问题在对应软件中处理。本机网站的已登记房间可在“接入 → 修正本场软件连接”修改路径，待服务确认采用后重启；线上连接暂不提供已登记房间的路径修正。保存本场路径不会改动后续讨论的默认配置，详见 [配置修正](docs/runner.md#已启动后的配置修正)。
+软件失败时先查看暂停原因，登录和权限问题在对应软件中处理。本机网站的已登记房间，可从进度看板 Codex 旁的“…”菜单直接选择“切换 App / CLI”，在弹窗内暂停并保存；软件路径在“接入 → 修正本场软件连接”修改。原会话与讨论进度保留，待服务确认采用后重启；线上连接暂不提供此修正入口。切到 App 后模型和文件权限按 App 设置，切回 CLI 前须让 App 释放原聊天。本场修改不影响后续讨论的默认配置，详见 [配置修正](docs/runner.md#已启动后的配置修正)。
 
 **手动接入：** 选择“手动复制冷启动”，将“接入”页的参与指令交给对应 AI 会话；接入后可由软件原生定时或人工触发，无需每轮复制本地 AI 的回答。在线专家由主持人操作浏览器或由用户转交，首次确定分工后沿用。详见 [讨论工作流](docs/workflow.md)。
 
@@ -135,7 +135,9 @@ codiscuss local connect https://你的站点
 
 默认上限为 **3 轮、不限时长、到限暂停**，可改为请主持人总结，并设置单次执行上限与收束时间。计数和停止规则见 [讨论上限](docs/workflow.md#讨论上限)。
 
-讨论结束后，可在左上方频道条目的“更多操作 → 导出对话”下载原始数据和全文，窄屏从频道导航进入；经典主题使用“备份当前讨论”按钮。也可用 `discuss-report` 整理带来源的综合报告。备份目前不支持导入恢复，也不下载外部附件。
+消息支持选择文件、拖拽或粘贴截图，每条最多 10 个附件、单个最多 20 MiB。PNG、JPEG、GIF、WebP 可放大查看，PDF 可在站内翻页、缩放，其他文件提供下载（不在线解析 Office 文件）。本地 AI 可通过 CLI 按需下载原文件；实际能否理解图片或文档取决于参与软件，上传不代表 AI 已读取。
+
+讨论结束后，可在左上方频道条目的“更多操作 → 导出对话”下载原始数据、全文及已发送附件原文件，窄屏从频道导航进入。也可用 `discuss-report` 整理带来源的综合报告。备份目前不支持导入恢复，也不下载外链内容。
 
 网站目前**没有用户登录和读取权限隔离**；公开部署前请按 [安全说明](.github/SECURITY.md) 配置访问边界。同一模型名称只能配置一个参与者，安排串行进行；在线读取、原生定时和浏览器操作能力取决于参与软件。
 

package/backend/runtime.py

@@ -6,9 +6,8 @@ import time
 import uuid
 
 LEASE_SECONDS = 120
-WORK_SECONDS = 1200
 MAX_RUN_SECONDS = 86400
-LIMIT_DEFAULTS = {"max_rounds": 0, "max_minutes": 0, "execution_minutes": 20,
+LIMIT_DEFAULTS = {"max_rounds": 0, "max_minutes": 0, "execution_minutes": 0,
                   "on_limit": "pause", "summary_minutes": 5}
 PUBLIC_EVENT_FIELDS = ("id", "room_id", "run_id", "generation", "cause", "participant",
                        "role", "turn_id", "status", "session_id", "lease_until", "result", "created_at")
@@ -45,7 +44,7 @@ def limit_values(payload, current=None):
     result = {key: (current or {}).get(key, value) for key, value in LIMIT_DEFAULTS.items()}
     result.update(payload)
     for key, minimum, maximum in (("max_rounds", 0, 100), ("max_minutes", 0, 1440),
-                                  ("execution_minutes", 1, 60), ("summary_minutes", 1, 60)):
+                                  ("execution_minutes", 0, 60), ("summary_minutes", 1, 60)):
         if type(result[key]) is not int or not minimum <= result[key] <= maximum:
             raise ValueError(f"{key} 须为 {minimum}–{maximum} 的整数")
     if result["on_limit"] not in ("pause", "summarize"):
@@ -93,7 +92,7 @@ def require_within_limits(room):
 def execution_seconds(connection, room_id):
     """Read the per-execution budget shared by entry, renewal and writes."""
     row = connection.execute("SELECT execution_minutes FROM discussion_limits WHERE room_id=?", (room_id,)).fetchone()
-    return row[0] * 60 if row else WORK_SECONDS
+    return (row[0] if row else LIMIT_DEFAULTS["execution_minutes"]) * 60
 
 
 def runtime_view(connection, room_id):
@@ -103,7 +102,7 @@ def runtime_view(connection, room_id):
         return {"mode": "manual", "status": "ready", "revision": 0, "protected": False}
     result = {key: row[key] for key in row.keys() if key != "control_hash"}
     result["protected"] = True
-    result["expired"] = row["status"] == "running" and row["deadline"] <= int(time.time())
+    result["expired"] = row["status"] == "running" and 0 < row["deadline"] <= int(time.time())
     event = connection.execute("SELECT id,participant,role,status,result FROM discussion_events WHERE room_id=? ORDER BY id DESC LIMIT 1", (room_id,)).fetchone()
     result["latest_execution"] = dict(event) if event else None
     result["participants"] = participant_executions(connection, room_id)
@@ -157,11 +156,12 @@ def authorize_write(connection, room_id, token, operation, payload):
     actor = payload.get("actor", payload.get("submitted_by", payload.get("author", "用户")))
     if actor == "用户":
         return
+    work_seconds = execution_seconds(connection, room_id)
     event = connection.execute("""SELECT * FROM discussion_events WHERE room_id=? AND token_hash=?
         AND run_id=? AND generation=? AND status='started' AND lease_until>?
-        AND started_at+?>?""", (room_id, hashed, run["run_id"], run["generation"],
-                                int(time.time()), execution_seconds(connection, room_id), int(time.time()))).fetchone()
-    if not hashed or not event or run["status"] != "running" or run["deadline"] <= int(time.time()):
+        AND (?=0 OR started_at+?>?)""", (room_id, hashed, run["run_id"], run["generation"],
+                                       int(time.time()), work_seconds, work_seconds, int(time.time()))).fetchone()
+    if not hashed or not event or run["status"] != "running" or 0 < run["deadline"] <= int(time.time()):
         raise PermissionError("执行资格无效、已结束或已到期；请重新读取运行状态")
     if actor != event["participant"]:
         raise PermissionError("执行凭据与实际参与身份不符")
@@ -173,6 +173,8 @@ def authorize_write(connection, room_id, token, operation, payload):
     elif operation == "floor":
         if payload.get("speaker", actor) != actor:
             raise PermissionError("自动执行只能操作本人发言权")
+    elif operation == "attachments":
+        pass
     elif event["role"] != "host" or operation not in ("topic", "state", "turns", "limits"):
         raise PermissionError("本次执行不能进行此操作")
 
@@ -212,7 +214,7 @@ def configure_runtime(connection, room_id, token, payload):
 
 
 def validate_start(connection, room_id, payload):
-    """Require an unarchived AI-hosted room and a finite user-specified deadline."""
+    """Validate an AI-hosted run; zero deadline leaves only the configured limits."""
     row = connection.execute("SELECT * FROM topic_settings JOIN room_directory USING(room_id) WHERE room_id=?",
                              (room_id,)).fetchone()
     limits = connection.execute(LIMITS_SQL, (room_id,)).fetchone()
@@ -224,9 +226,9 @@ def validate_start(connection, room_id, payload):
         configured = min(value for value in (configured, closing_deadline, previous_deadline) if value)
     deadline = payload.get("deadline", configured)
     if type(deadline) is int and configured:
-        deadline = min(deadline, configured)
-    if type(deadline) is not int or not int(time.time()) < deadline <= int(time.time()) + MAX_RUN_SECONDS:
-        raise ValueError("deadline 须为未来 24 小时内的 Unix 秒数")
+        deadline = min(deadline or configured, configured)
+    if type(deadline) is not int or (deadline != 0 and not int(time.time()) < deadline <= int(time.time()) + MAX_RUN_SECONDS):
+        raise ValueError("deadline 须为 0（不限时长）或未来 24 小时内的 Unix 秒数")
     experts = json.loads(row["experts"])
     if row["archived"] or row["phase"] == "concluded" or row["host"] == "用户":
         raise ValueError("自动运行需要未归档、未收束的话题及 AI 主持人")
@@ -262,12 +264,14 @@ def event_action(connection, room_id, event_id, token, payload):
     now = int(time.time())
     work_seconds = execution_seconds(connection, room_id)
     action = payload.get("action")
+    work_deadline = (now if action == "enter" else event["started_at"]) + work_seconds if work_seconds else 0
+    lease_until = min(now + LEASE_SECONDS, work_deadline or now + LEASE_SECONDS, run["deadline"] or now + LEASE_SECONDS)
     controller = token_hash(token) == run["control_hash"]
     if not controller and (not token_hash(token) or token_hash(token) != event["token_hash"]):
         raise PermissionError("执行凭据无效")
     if action == "view":
         return {key: event[key] for key in PUBLIC_EVENT_FIELDS}
-    if run["status"] != "running" or run["deadline"] <= now or event["generation"] != run["generation"]:
+    if run["status"] != "running" or 0 < run["deadline"] <= now or event["generation"] != run["generation"]:
         raise FileExistsError("本次运行已停止或执行已过期")
     if action == "reserve" and controller:
         reserve_event(connection, event, payload)
@@ -275,12 +279,12 @@ def event_action(connection, room_id, event_id, token, payload):
         if event["status"] != "dispatched" or not payload.get("session_id"):
             raise FileExistsError("本次任务已进入或不可进入；不要重复执行")
         connection.execute("UPDATE discussion_events SET status='started',session_id=?,started_at=?,lease_until=? WHERE id=?",
-                           (str(payload["session_id"])[:200], now, min(now + LEASE_SECONDS, now + work_seconds, run["deadline"]), event_id))
+                           (str(payload["session_id"])[:200], now, lease_until, event_id))
     elif action == "renew" and controller and event["status"] == "started":
-        if event["lease_until"] <= now or event["started_at"] + work_seconds <= now:
+        if event["lease_until"] <= now or 0 < work_deadline <= now:
             raise FileExistsError("执行租约已到期，不能自动重新激活")
         connection.execute("UPDATE discussion_events SET lease_until=? WHERE id=?",
-                           (min(now + LEASE_SECONDS, event["started_at"] + work_seconds, run["deadline"]), event_id))
+                           (lease_until, event_id))
     elif action == "fail" and event["status"] in ("dispatched", "started"):
         connection.execute("UPDATE discussion_events SET status='failed',result=? WHERE id=?",
                            (str(payload.get("reason", "需要用户处理"))[:500], event_id))

package/backend/schema.sql

@@ -52,6 +52,27 @@ CREATE TABLE IF NOT EXISTS messages (
     UNIQUE(room_id, submission_id)
 );
 CREATE INDEX IF NOT EXISTS messages_room ON messages(room_id, id);
+CREATE TABLE IF NOT EXISTS attachments (
+    id TEXT PRIMARY KEY, room_id TEXT NOT NULL REFERENCES rooms(id) ON DELETE CASCADE,
+    message_id INTEGER REFERENCES messages(id) ON DELETE CASCADE,
+    position INTEGER NOT NULL DEFAULT 0,
+    file_name TEXT NOT NULL, content_type TEXT NOT NULL,
+    size INTEGER NOT NULL, sha256 TEXT NOT NULL,
+    actor TEXT NOT NULL, submission_id TEXT NOT NULL, created_at TEXT NOT NULL,
+    UNIQUE(room_id, submission_id)
+);
+CREATE INDEX IF NOT EXISTS attachments_message ON attachments(message_id, position);
+CREATE TRIGGER IF NOT EXISTS validate_attachment_link BEFORE UPDATE OF message_id ON attachments
+WHEN NEW.message_id IS NOT NULL AND (
+    (OLD.message_id IS NOT NULL AND OLD.message_id != NEW.message_id)
+    OR NOT EXISTS (
+        SELECT 1 FROM messages m JOIN message_details d ON d.message_id=m.id
+        WHERE m.id=NEW.message_id AND m.room_id=NEW.room_id AND d.submitted_by=NEW.actor
+    )
+)
+BEGIN
+    SELECT RAISE(ABORT, 'attachment_unavailable');
+END;
 CREATE TABLE IF NOT EXISTS message_summaries (
     message_id INTEGER PRIMARY KEY REFERENCES messages(id),
     text TEXT NOT NULL DEFAULT '', author TEXT NOT NULL DEFAULT ''
@@ -191,7 +212,7 @@ CREATE TABLE IF NOT EXISTS discussion_limits (
     room_id TEXT PRIMARY KEY REFERENCES rooms(id) ON DELETE CASCADE,
     max_rounds INTEGER NOT NULL DEFAULT 0,
     max_minutes INTEGER NOT NULL DEFAULT 0,
-    execution_minutes INTEGER NOT NULL DEFAULT 20,
+    execution_minutes INTEGER NOT NULL DEFAULT 0,
     on_limit TEXT NOT NULL DEFAULT 'pause',
     summary_minutes INTEGER NOT NULL DEFAULT 5,
     started_at INTEGER NOT NULL DEFAULT 0,
@@ -199,10 +220,11 @@ CREATE TABLE IF NOT EXISTS discussion_limits (
     stop_reason TEXT NOT NULL DEFAULT '',
     revision INTEGER NOT NULL DEFAULT 0
 );
-INSERT OR IGNORE INTO discussion_limits(room_id) SELECT id FROM rooms;
-CREATE TRIGGER IF NOT EXISTS limits_room_created AFTER INSERT ON rooms
+INSERT OR IGNORE INTO discussion_limits(room_id,execution_minutes) SELECT id,0 FROM rooms;
+DROP TRIGGER IF EXISTS limits_room_created;
+CREATE TRIGGER limits_room_created AFTER INSERT ON rooms
 BEGIN
-    INSERT INTO discussion_limits(room_id) VALUES (NEW.id);
+    INSERT INTO discussion_limits(room_id,execution_minutes) VALUES (NEW.id,0);
 END;
 CREATE TRIGGER IF NOT EXISTS limits_topic_started AFTER INSERT ON topic_settings
 WHEN NEW.phase IN ('initial','discussing','reviewing')
@@ -225,7 +247,7 @@ CREATE TABLE IF NOT EXISTS discussion_runtime (
     generation INTEGER NOT NULL DEFAULT 1,
     status TEXT NOT NULL CHECK(status IN ('ready','running','paused','stopped','concluded')),
     control_hash TEXT NOT NULL,
-    deadline INTEGER NOT NULL DEFAULT 0,
+    deadline INTEGER NOT NULL DEFAULT 0, -- 0 means no run duration limit; execution leases remain bounded.
     revision INTEGER NOT NULL DEFAULT 1,
     resume_phase TEXT NOT NULL DEFAULT 'initial',
     updated_at TEXT NOT NULL
@@ -258,6 +280,20 @@ CREATE TABLE IF NOT EXISTS event_timings (
     dispatched_at INTEGER NOT NULL DEFAULT 0,
     finished_at INTEGER NOT NULL DEFAULT 0
 );
+-- Freeze dispatch-to-publication time independently of the recent execution list.
+CREATE TABLE IF NOT EXISTS message_timings (
+    message_id INTEGER PRIMARY KEY REFERENCES messages(id) ON DELETE CASCADE,
+    elapsed_seconds INTEGER NOT NULL CHECK(elapsed_seconds>=0)
+);
+-- Only completed expert events identify a historical message unambiguously.
+INSERT INTO message_timings(message_id,elapsed_seconds)
+    SELECT m.id,MAX(0,CAST(strftime('%s',m.created_at) AS INTEGER)-t.dispatched_at)
+    FROM messages AS m JOIN discussion_events AS e
+        ON e.room_id=m.room_id AND e.result='message:'||m.id AND e.participant=m.author
+    JOIN event_timings AS t ON t.event_id=e.id
+    WHERE m.source='cli' AND e.role='expert' AND e.status='completed'
+        AND e.started_at>0 AND t.dispatched_at>0 AND t.finished_at>0
+    ON CONFLICT(message_id) DO NOTHING;
 CREATE TRIGGER IF NOT EXISTS record_event_timing AFTER UPDATE OF status ON discussion_events
 WHEN OLD.status<>NEW.status
 BEGIN