fix: salvage truncated worker envelopes and log worker lifecycle

Models occasionally drop the closing tag of the result envelope while
the JSON itself is complete; the strict parser rejected these and one
repair attempt could not always recover, cascading into worker_error
and dropping valid attachments. The parser now falls back to
brace-balanced salvage when the closing tag is missing, and worker
lifecycle events (start, resume, invalid envelope, repair, settle,
worker_error) are logged.
This commit is contained in:
zenord
2026-08-19 15:44:53 +08:00
parent 6b506b8c55
commit 9c3139380d
5 changed files with 183 additions and 11 deletions
+4 -3
View File
@@ -87,8 +87,9 @@ gori-agent list
- 每个 ACP worker 使用独立进程组和随机 `GORI_AGENT_WORKER_TOKEN`;cancel、timeout、crash 或 assistant 隔离违约必须清理同进程组工具后代;bootstrap 禁止 `setsid`、`nohup`、detached/daemon/background 遗留进程,主动脱组属于无 cgroup/Bubblewrap 方案的边界。
- assistant 只接受 Kimi Code ACP,并依赖项目级 `tools: []`、`subagents: []` profile;permission deny 只是附加层。
- `src/acp/assistant-manager.ts`
- 每 conversation(chat + user)一个无工具 Assistant 会话,同群不同用户互相隔离(`GORI_ASSISTANT_ACTION_V2` envelope:create_proposal/confirm/adjust_proposal/follow_up/finish/start_next/cancel/stop,格式只修复一次);runtime 执行 action 后在 reply 末尾追加人话纠正(如 start_next 被阻塞、无 owner 匹配),Assistant 不得自行宣称 action 已生效。
- 唯一活跃 Worker 执行已确认 Proposal(`GORI_WORKER_RESULT_V2`:仅 `PENDING`,summary 必填,可带 question/workspaceDirty);pending 不区分 success/fail,只有 finish 落定为 finished(done),cancel 落定为 finished(cancelled),不自动开始下一个。
- 每 conversation(chat + user)一个无工具 Assistant 会话,同群不同用户互相隔离(`GORI_ASSISTANT_ACTION_V2` envelope:create_proposal/confirm/adjust_proposal/follow_up/finish/send_image/start_next/cancel/stop,格式只修复一次);runtime 执行 action 后在 reply 末尾追加人话纠正(如 start_next 被阻塞、无 owner 匹配),Assistant 不得自行宣称 action 已生效。
- 唯一活跃 Worker 执行已确认 Proposal(`GORI_WORKER_RESULT_V2`:仅 `PENDING`,summary 必填,可带 question/workspaceDirty/attachments);pending 不区分 success/fail,只有 finish 落定为 finished(done),cancel 落定为 finished(cancelled),不自动开始下一个。
- envelope 解析先严格匹配(闭合标签 + 文本末尾 anchor);缺闭合标签时从起始标签后做花括号配平(字符串/转义感知)salvage,配平点必须在文本末尾才接受,否则仍判无效走修复。worker 启动(new/resume session)、envelope 无效、repair 成败、settle(attachments/dropped 数)、worker_error、follow_up resume/fresh 兜底均有单行安全日志(proposal id 前 8 位,不含正文)。
- capacity(maxAssistantSessions/maxProcesses)、idle sweep、cancel/confirm/finish/stop、worker_lost 恢复、owner 事件通知。
- `src/core/durable-session-store.ts`
- state v3 只保存 bot/platform identity 和 assistant binding(conversation key,即 chat + user、agent ID、native session ID、assistant workspace、fingerprint、时间戳);不保存消息正文。
@@ -209,7 +210,7 @@ Kimi Code agent `args` 必须严格为 `["acp"]`,不能添加可能绕过 assi
Proposal 状态流:`proposed → queued → working → pending → finished`。`proposed` 必须用户确认才进入 `queued`;`pending` 就是「等用户决定」,不区分 success/failure;只有 `finish` 把 pending 落定为 `finished(done)`,`cancel`(proposed/queued/pending)落定为 `finished(cancelled)`;没有 working、owner 自己没有 pending、全局没有 workspaceDirty pending 时,最早确认的 queued Proposal 才可通过 confirm 或 start_next 开始。
Assistant 每条回复以隐藏 `GORI_ASSISTANT_ACTION_V2` envelope 结尾(`reply` + `actions`),action 仅 `create_proposal`、`confirm`、`adjust_proposal`(仅 proposed/queued)、`follow_up`(pending → working,优先 resume 原 worker native session,失败则带 Proposal 上下文新起 session,用户图片附件随 prompt 给 Worker)、`finish`、`send_image`(把 Worker 报告过的 workspace 内图片发给用户)、`start_next`、`cancel`、`stop`,格式错误只修复一次。Worker 每轮以隐藏 `GORI_WORKER_RESULT_V2` envelope 收尾:仅 `PENDING`(`summary` 必填,可选 `question`、`workspaceDirty`、`attachments`),同样只修复一次。
Assistant 每条回复以隐藏 `GORI_ASSISTANT_ACTION_V2` envelope 结尾(`reply` + `actions`),action 仅 `create_proposal`、`confirm`、`adjust_proposal`(仅 proposed/queued)、`follow_up`(pending → working,优先 resume 原 worker native session,失败则带 Proposal 上下文新起 session,用户图片附件随 prompt 给 Worker)、`finish`、`send_image`(把 Worker 报告过的 workspace 内图片发给用户)、`start_next`、`cancel`、`stop`,格式错误只修复一次。Worker 每轮以隐藏 `GORI_WORKER_RESULT_V2` envelope 收尾:仅 `PENDING`(`summary` 必填,可选 `question`、`workspaceDirty`、`attachments`),同样只修复一次。envelope 缺闭合标签(模型截断)时先按花括号配平 salvage(字符串/转义感知,配平点必须在文本末尾),失败才进入修复流程。
确认语义: