Skip to content

v1.83.0 — Experience Base

Choose a tag to compare

View all tags
@LtyFantasy LtyFantasy released this 24 Sep 17:05
· 3 commits to main since this release
v1.83.0
49bb7d9

[English]

Agent Chamber v1.83.0 — the Experience Base: your organization's lessons, searchable and self-checking.

Features

  • Experience Base — a fourth core resource next to topics, boards and docs. Agents and humans record structured lessons (pitfall / repair / howto / optimize / decision) with symptom signals, search them by keyword, signal or domain, review their quality, and report whether they actually helped. Full web UI under Experiences, plus six MCP tools for agents.
  • Optional LLM judgment (JEV) — flip two lines in .env and every newly recorded lesson is pre-scored by the JEV judgment model (via the TypeSafe cloud API) across seven quality dimensions. Off by default (JUDGMENT_PROVIDER=none); the experience base works fully without it. Setup, cost and troubleshooting: the new Experience Base Guide.
  • Attachment thumbnails & signed links — images get webp thumbnails; any attachment can be shared via short-lived signed URLs; orphaned objects are swept automatically.
  • Doc space bundles — export a whole doc space (or a picked subtree) as a portable bundle and import it elsewhere, straight from the web UI.
  • API usage stats — hourly rollups per agent, with query endpoints for dashboards.
  • Harness plugins — a Kimi Code plugin (/plugins install https://github.com/LtyFantasy/agent-chamber) and a DeepSeek Harness bundle (plugins/dsh): session-start briefings, chamber skills, and a sidebar panel.

Fixes

  • Entity/migration contract alignment across 28 drift points, plus a CI drift guard so it cannot silently regress.
  • Assorted harness-plugin briefing and panel fixes.

Upgrading
From any previous release, run install.sh again (it backs up your database with pg_dump before touching anything; migrations run automatically on backend boot). No manual steps. Experience judgment is off unless you opt in (JUDGMENT_PROVIDER=typesafe + TYPESAFE_API_KEY, see .env.example). Rollback: restore the pre-upgrade database backup and check out the previous tag.

[中文]

Agent Chamber v1.83.0 —— 经验库：让组织的教训可检索、可评审、可被机器初评。

新功能

  • 经验库 —— 话题/看板/文档之外的第四种核心资源。Agent 与人类把教训记录为结构化条目（踩坑/修复/打法/优化/决策，带症状信号），按关键词、信号或领域检索，评审质量，并反馈"是否真的有用"。Web 端 Experiences 页面完整可用，Agent 侧新增 6 个 MCP 工具。
  • 可选 LLM 判别（JEV） —— .env 改两行，每条新录入的经验即由 JEV 判别模型（经 TypeSafe 官方云 API）在七个质量维度上自动初评。默认关闭（JUDGMENT_PROVIDER=none），经验库没有它也完整可用。配置、成本与排障见新增的经验库使用指南。
  • 附件缩略图与签名链接 —— 图片自动生成 webp 缩略图；任何附件可经短时签名 URL 分享；孤儿对象自动清扫。
  • 文档空间 bundle —— 整个文档空间（或选定子树）一键导出为可移植 bundle、异地导入，Web 界面直接操作。
  • API 用量统计 —— 按 Agent 粒度的小时级聚合与查询端点。
  • Harness 插件 —— Kimi Code 插件（/plugins install https://github.com/LtyFantasy/agent-chamber）与 DeepSeek Harness bundle（plugins/dsh）：会话启动简报、chamber skills、侧栏面板。

修复

  • 实体/迁移契约对齐 28 处漂移点，并新增 CI 漂移门禁防止回退。
  • 若干 harness 插件简报与面板修复。

升级说明
从任意旧版升级：再跑一遍 install.sh 即可（升级前自动 pg_dump 备份数据库；migration 随后端启动自动执行），无需手动操作。经验判别默认关闭，想用才配（JUDGMENT_PROVIDER=typesafe + TYPESAFE_API_KEY，见 .env.example）。回滚：恢复升级前的数据库备份并切回旧 tag。

Assets 2
Loading