Skip to content

0.1.20

Choose a tag to compare

View all tags
@github-actions github-actions released this 05 Sep 15:24
· 475 commits to dev since this release
v0.1.20
4ae4c7b

更新内容

✨ 新功能

  • batch stage-1 model calls with configurable concurrency
  • standalone memory-run script with prompt/response tracing
  • rewrite stage1/phase2 prompts with shared taxonomy and raise stage1 input budget to 12k tokens
  • feed prior MEMORY.md/summary into phase-2 consolidation
  • quota raw memory per artifact in phase-2 corpus build
  • show merge progress in the popover
  • run popover merge immediately
  • add workspace actions and secure Markdown preview
  • allow editing project memory from the popover
  • turn project memory into a structured popover
  • move project memory into a tab-bar popover
  • complete slash subcommands after a command chip
  • add /memory slash for local project memory
  • map memory://root and use title model for phase two
  • write optional managed skills from learn
  • sanitize learned lessons and consolidation text
  • run memory pipeline after parent spawn
  • run two-phase memory pipeline with injected model
  • scan session jsonl for memory threads
  • inject memory guidance on parent spawn
  • render OMP-style memory guidance text
  • add json sidecar leases for memory jobs
  • apply phase-two memory consolidation
  • select idle sessions for memory stage one
  • add learn tool for local memory lessons
  • store learned lessons under agent memory root
  • show assistant reviewing on the tool row
  • warn against cat/grep in bash tool description
  • give coworkers a peer message tool
  • add assistant review mode
  • rename paired inbound devices on the host
  • Windows local agent shell preference
  • write project memory to .enso/learned.md
  • coworker peer message via notifier
  • structured subagent yield and agent:// read
  • use official PowerShell on local Windows
  • allow per-server connect and tool-call timeouts
  • explore-fold tools and LLM-context collapse
  • intercept cat/grep/sed and PowerShell aliases
  • collapse unit prices and edit inline
  • 圆环与占用弹窗共用 oh-my-pi 上下文水位
  • allow custom model unit prices for cost estimates
  • 远程 server 无条件提供授权入口并提示新会话生效
  • main 缓存连接状态并提供快照通道
  • main 侧 OAuth 授权编排与 token 加密持久化
  • default 10-minute timeout for foreground bash
  • abort and retry when a run stalls with no output
  • 新增「加载项目内其它工具目录」开关，注入 .claude/.codex/.cursor 的 skills 与规则文件
  • 列表显示 skill/MCP/内置工具/指令的上下文占用
  • archive whole projects without touching per-session archived state
  • 指定模型时可设思考档，对齐 pi /thinking

🐛 问题修复

  • remove obsolete IPC capability coverage
  • follow streaming tool output
  • exclude archived empty conversations from reuse
  • prevent main agent tab from shrinking
  • keep conversation active while child agents run
  • surface stopReason=error as failure and leave failed stage-1 threads for retry
  • point cat >/>> and heredocs at write, not read
  • release phase-2 lease and keep dirty flag when consolidation fails
  • apply exclusion taxonomy to prior memory cleanup, keep superseded claims out of memory_summary
  • fix phase-2 corpus budgeting (raw-only candidate cap, full-block accounting, cap-not-skip newest)
  • drop redundant 'read summary first' rule, guide on-demand MEMORY.md reads
  • cap oversized tool results instead of dropping them
  • distinguish global job lifecycle status
  • restore baseline SDK types and async lifecycle checks
  • coalesce session persistence to prevent OOM (#51)
  • queue sends while compaction is in progress
  • match OMP local memory extraction filters
  • stop streaming code blocks from flickering
  • tighten the project memory popover
  • Enter on a slash subcommand sends it
  • reconnect once on stale transport errors
  • stop idle extract into project learned.md
  • remember last approval mode for new sessions
  • flush queued messages after compaction
  • do not stall-retry while tools or coworkers are live
  • download Electron binary during pnpm postinstall
  • allow retrying a failed session resume
  • restart exited worker before queueing commands
  • normalize stringified edit args before schema check
  • merge duplicate workspace identities
  • show pin mark on pinned sessions without hover
  • use stock pi edit tool instead of lenient wrapper
  • 仅对声明 thinkingLevelMap 的模型钳位档位
  • require model when agent_type is agent_pick
  • 调整语言与代理模式间距
  • 忽略零 Token 模型缺价提示
  • 补充 Gemini 3.8 目录单价
  • 补充 Gemini 3.8 模型路由
  • 主 agent 发送消息静默无响应（seq 守卫 / 僵尸轮 / 失败回流） (#49)
  • fill missing assistant usage before leaving Cursor
  • bump pi-cursor to 1.4.30 for Cursor 3.18.9 wire fields
  • 撤销授权后同步状态缓存并跨窗口推送
  • 授权取消时回滚旧凭据
  • 换 token 原地更新 provider，不再关闭正在使用的连接
  • token 白名单改为裁剪，避免 refresh 结果被整条丢弃
  • MCP server 条目要求非空 id
  • 授权成功后按 serverId 定向 warm，worker 未运行时标记待连接
  • 收窄 worker MCP 事件解析并复用 token 白名单
  • 重新授权前清除失效 token
  • 预挂回调 handler 避免无人接管的 promise rejection
  • 连接缓存按 token 指纹重建，失败时关闭 client
  • resync snapshot when optimistic echo masks a cold-evicted timeline
  • 未知支持集不回写档位，子代理可复位跟随
  • turn status dot green while ask_user is pending
  • merge worktree sessions into the host project
  • MCP occupancy probe times out and runs in parallel
  • 子代理模型复用主 Agent 思考交互
  • 思考档按模型真实能力过滤，Grok 不再出现极高/max
  • 保存时提交未点加号的模型，xAI 与 OpenAI 中转互不吞并
  • 地区不可用时不再阻断 Google 订阅登录
  • Custom 与官方同 URL 时保留独立条目并合并模型

📦 下载

各平台安装包见下方 Assets，或访问 Releases。

  • macOS: .dmg（Apple Silicon / Intel）
  • Windows: EnsoCode-Setup-0.1.20.exe
  • Linux: .AppImage / .deb

Full Changelog: v0.1.19...v0.1.20

Assets 17
Loading