v0.3.1

v0.3.1

中文

本版本新增 Codex 自动压缩阈值配置，统一 Codex 并发设置，并优化思维图的 Agent 检索与更新流程。同时改进移动端新建卡片和列表浏览体验，修复历史清理、Codex 限流后重新执行及会话用量记录展示问题。

Codex 配置

• 全局设置的 Codex 页新增自动压缩阈值，可按 Token 数配置上下文自动压缩的触发时机。
• 阈值留空时跟随 Codex 默认值；自定义值必须为正整数，配置会持久化并在服务启动时恢复。
• 自动压缩阈值变更沿用上下文长度配置的恢复机制：重启共享 Codex App Server，并将活动卡片转入恢复队列，使用新配置继续执行。
• 将 Agent 并发数量从通用设置移至 Codex 设置，与上下文长度和自动压缩阈值集中管理；单独修改并发数量会触发队列调度，无需重启 Codex App Server。
• GraphQL 设置接口同步调整：agentMaxConcurrent 从 GeneralSettings / UpdateGeneralSettingsInput 移至 CodexSettings / UpdateCodexSettingsInput；更新 Codex 设置时须提交该字段，并可通过 autoCompactTokenLimit 设置自动压缩阈值。

思维图检索与更新

• Agent 的 mind_map_search 支持 AND、OR 和括号组合查询，AND 优先于 OR，例如 workflow AND (runner OR retry)；相邻查询词必须显式使用操作符连接。
• 将节点 Tag 纳入检索和匹配字段，并作为命中节点的元数据返回，减少独立 Tag 节点和 Tag 关系对结果的干扰。
• Agent 搜索默认返回数量由 20 降至 5；优先返回命中节点详情与代码位置，关联节点仅返回 ID 和标题摘要。
• 为搜索输出增加大小预算，超出预算时省略详情或后续条目，并通过返回数量和截断标记说明结果是否完整。
• 搜索结果直接返回可用于 mind_map_update 的 tagRevision；mind_map_tags 精简为标签名称列表和版本号，减少重复获取标签的开销。
• 调整异步整理指令，引导 Agent 围绕当前会话进行必要的窄范围检索、复用已有节点并集中更新；没有稳定的图变化时跳过更新，失效内容清理也限定在当前会话涉及的节点及关系内。

卡片创建与列表浏览

• 移动端新建卡片改用完整页面布局，移除页面内重复的对话框标题和容器样式，并使用统一页面工具栏。
• 新建卡片页面由页面本身负责滚动，移除移动端提示词输入框的额外高度限制，改善长内容输入和滚动体验。
• 卡片列表默认每页显示 20 条，并同步调整共享分页默认值及列表分页选项，减少翻页操作。

历史记录与工作树清理

• 清理已关闭卡片的历史记录前，主动完成待执行或已到重试时间的工作树清理，避免仅因清理仍处于待处理状态而拒绝历史清理。
• 在卡片锁内重新确认关闭状态和清理结果，确认工作树清理完成后再继续删除会话文件及历史数据。
• 工作树清理失败或暂不满足重试条件时，保留历史记录并返回清理状态及可用的错误代码，便于定位问题。

Codex 限流恢复

• 将 Codex 返回的 HTTP 429 明确识别为限流失败，并保留对应错误信息。
• 限流失败后清除不可继续使用的 Codex 会话关联；再次执行卡片时创建新的 Codex 会话，避免继续恢复原会话。
• 工作流节点重试同样使用新会话；限流退出时跳过基于原会话的问题回答续跑路径，避免重新进入失效会话。

会话用量记录

• 解析 Codex 会话日志时过滤 token_usage_record，避免该类底层用量记录进入会话时间线。
• 继续以 token_count 读取用量信息，保留累计 Token、本次用量和上下文长度等统计数据。

English

This release adds configurable Codex auto-compaction thresholds, brings concurrency controls into Codex settings, and improves Agent search and update workflows for mind maps. It also improves mobile card creation and list browsing, and fixes history cleanup, restarting after Codex rate limits, and usage-record handling in session transcripts.

Codex configuration

• Add an auto-compaction threshold to the Codex section of global settings, allowing the context-compaction trigger to be configured by token count.
• Leaving the threshold empty follows the Codex default; custom values must be positive integers. Persist the setting and restore it when the service starts.
• Apply threshold changes through the existing context-window recovery mechanism: restart the shared Codex App Server and queue active cards to continue with the new configuration.
• Move Agent concurrency from general settings to Codex settings alongside the context window and auto-compaction threshold. Changing concurrency alone triggers queue scheduling without restarting the Codex App Server.
• Update the GraphQL settings API accordingly: move agentMaxConcurrent from GeneralSettings / UpdateGeneralSettingsInput to CodexSettings / UpdateCodexSettingsInput. Codex settings updates now require this field and accept autoCompactTokenLimit for the compaction threshold.

Mind map search and updates

• Support AND, OR, and parentheses in Agent mind_map_search queries, with AND taking precedence over OR, for example workflow AND (runner OR retry). Adjacent search terms must be joined by an explicit operator.
• Include node tags in search matching and matched-field reporting, and return them as metadata on matching nodes, reducing noise from standalone Tag nodes and Tag relationships.
• Reduce the default Agent search limit from 20 to 5. Prioritize matching-node details and code locations, while returning related nodes as ID-and-title summaries.
• Add an output-size budget for search results. Omit details or later entries when the budget is exceeded, and report result counts and truncation status.
• Return tagRevision directly from search for use with mind_map_update. Simplify mind_map_tags to tag names and the revision, reducing redundant tag lookups.
• Refine asynchronous maintenance instructions to guide the Agent toward focused searches for the current session, reuse of existing nodes, and consolidated updates. Skip updates when there are no stable graph changes, and limit obsolete-content cleanup to nodes and relationships involved in the current session.

Card creation and list browsing

• Use a full-page layout for mobile card creation, remove duplicate dialog headings and container styling, and adopt the shared page toolbar.
• Let the page own scrolling and remove the additional mobile prompt-input height restriction, improving long-prompt editing and scrolling.
• Show 20 cards per page by default and align shared pagination defaults and list page-size options, reducing pagination.

History and worktree cleanup

• Before purging closed-card history, finish pending worktree cleanup or retries that are due, instead of rejecting history cleanup solely because worktree cleanup is still pending.
• Recheck the card's closed state and cleanup outcome under the card lock, and confirm worktree cleanup has completed before deleting session files and history.
• Preserve history when worktree cleanup fails or is not yet eligible for retry, and return the cleanup status and available error code to help diagnose the issue.

Recovery after Codex rate limits

• Classify HTTP 429 responses from Codex as rate-limit failures and preserve the corresponding error information.
• Clear the unusable Codex session association after a rate-limit failure. Executing the card again creates a new Codex session instead of resuming the previous one.
• Use a fresh session for workflow-node retries as well. On rate-limited exits, skip the question-answer continuation path tied to the previous session to avoid re-entering it.

Session usage records

• Filter token_usage_record entries when parsing Codex session logs so these low-level usage records do not appear in the session timeline.
• Continue reading usage from token_count, preserving cumulative tokens, current usage, and context-window statistics.

Complete commit list / 完整提交列表

• 66e9a05 — Use a full-page layout and page-owned scrolling for mobile card creation.

  移动端新建卡片使用完整页面布局和页面滚动，统一工具栏并修正输入区域样式。

• 578fef0 — Add persistent Codex auto-compaction settings and move Agent concurrency controls into Codex settings, with corresponding GraphQL API changes.

  新增持久化 Codex 自动压缩阈值配置，将 Agent 并发控制移入 Codex 设置，并同步调整 GraphQL 接口。

• 934aeb2 — Add Boolean Agent mind map queries, tag-aware matching, budgeted results, and focused search/update guidance.

  优化 Agent 思维图检索，支持布尔查询、Tag 匹配、输出预算及更聚焦的检索更新流程。

• e99c630 — Finish eligible worktree cleanup before purging closed-card history, and set default card-list pagination to 20 items.

  清理已关闭卡片历史前先完成符合条件的工作树清理，并将卡片列表默认分页调整为每页 20 条。

• 9bc8bc5 — Classify Codex HTTP 429 failures and start fresh Codex sessions on subsequent card executions or workflow retries.

  识别 Codex HTTP 429 限流失败，并在后续卡片执行或工作流重试时使用新的 Codex 会话。

• d0d89f5 — Filter token usage records from transcripts while preserving token-count statistics.

  过滤会话日志中的底层用量记录，同时保留基于 token_count 的用量统计。

查看完整变更 / Full changelog: v0.3.0...v0.3.1