dominds 1.7.1 → 1.7.2

package/dist/docs/context-health.md

@@ -116,9 +116,11 @@ Levels are derived from the two thresholds:
 ### Continuation package
 
 The remediation workflow centers around a _continuation package_ (a scannable, actionable bundle of
-context that survives a new course). When the agent is still clear-headed, this should preferably be
-kept as **one structured reminder**. But when the agent is already muddled in caution/critical,
-multiple rough reminders are acceptable as a bridge.
+context that survives a new course). The agent must not self-assess whether it is “still
+clear-headed” or “already muddled” because that judgment is not reliable or auditable. The only
+mechanical branch is whether the system has put the current course into `caution` / `critical`
+remediation. Outside remediation, prefer **one structured reminder**. During remediation, rough
+multi-reminder carry-over is acceptable as a bridge.
 
 Recommended structure (multi-line but compact; focus on details not covered in Taskdoc):
 
@@ -130,10 +132,13 @@ Recommended structure (multi-line but compact; focus on details not covered in T
 Rules:
 
 - Keep normal reminders concise and few.
-- When preparing a new course, prefer compressing into one continuation-package reminder if the
-  agent is still clear-headed.
-- If the agent is already muddled, prioritize preserving volatile information even via multiple rough
-  reminders; in the next course, the first action should be to review, merge, and delete redundancy.
+- Do not let the agent branch on subjective self-assessment such as “clear-headed” vs “muddled”.
+- When the current course is not under `caution` / `critical` remediation, prefer compressing into
+  one structured continuation-package reminder.
+- When the system has put the current course into `caution` / `critical` remediation, prioritize
+  preserving volatile information even via multiple rough reminders; in the current course, only
+  preserve info and `clear_mind`. Once the system actually starts the next course, the first step
+  is to review, merge, and delete redundancy.
 - Do not duplicate Taskdoc content except for a short bridge when strictly needed.
 - Do not paste long raw logs/tool outputs into the continuation package.
 
@@ -150,7 +155,7 @@ Current behavior:
   generations; configurable per model).
 - Each inserted prompt requires the agent to **curate reminders** (at least one call):
   - `update_reminder` (preferred) / `add_reminder`
-  - Prefer a structured continuation-package reminder; if already muddled, rough multi-reminder carry-over is acceptable
+  - Default to one structured continuation-package reminder; if the current course is already under remediation and one structured reminder cannot be produced directly from already observed facts, rough multi-reminder carry-over is acceptable
   - Then `clear_mind` when it becomes scannable/actionable
 
 ### Critical (red)
@@ -159,7 +164,7 @@ When `level === 'critical'`, the driver enters a **countdown remediation** (max
 
 - On each turn, the driver records a **role=user prompt** (persisted as a user message) that is
   visible in the UI as a user prompt. This prompt tells the agent to:
-  - curate reminders via `update_reminder` / `add_reminder`, preferably into a continuation-package reminder but allowing rough multi-reminder carry-over when already muddled, and
+  - curate reminders via `update_reminder` / `add_reminder`, preferably into a continuation-package reminder but allowing rough multi-reminder carry-over when the current course is already under remediation and a single structured reminder cannot be produced directly from already observed facts, and
   - then call `clear_mind` to start a new course.
 - The prompt includes a countdown: after **N** turns the system will automatically clear.
 - When the countdown reaches 0, the driver **automatically calls** `clear_mind` (with empty args; no
@@ -226,7 +231,7 @@ Additional constraints:
   - `caution`: driver inserts a persisted role=user prompt (UI-visible user instruction).
     On entering `caution` it inserts once; while still `caution` it reinserts on a cadence (default: every
     10 generations; configurable per model). Each time, the agent must call at least one of
-    `update_reminder` / `add_reminder` and preserve a continuation package. A single structured reminder is preferred, but multiple rough reminders are acceptable when already muddled; in the new course the agent should reconcile them first, then `clear_mind` when ready.
+    `update_reminder` / `add_reminder` and preserve a continuation package. A single structured reminder is preferred when the current course is not under remediation pressure; during remediation, multiple rough reminders are acceptable if they can be written directly from already observed facts without further reading/analysis. In the new course the agent should reconcile them first, then `clear_mind` when ready.
   - `critical`: driver runs a countdown remediation (max 5 turns) using **recorded role=user prompts**.
     Each prompt includes a countdown and instructs reminder curation + `clear_mind`. When the countdown
     reaches 0, the driver auto-executes `clear_mind` and starts a new course (no Q4H, no suspension).

package/dist/docs/context-health.zh.md

@@ -107,6 +107,15 @@ Dominds 计算比率：
 - 运行/验证（命令、端口、环境变量）
 - 容易丢失的临时细节（路径/ID/URL/示例输入）
 
+规则：
+
+- 普通提醒项保持简短且少量。
+- 不允许智能体靠主观感觉判断自己“头脑清楚”还是“已经发乱”；这两种说法没有可靠、可审计的客观判据。
+- 机制性分流规则只有一个：看**系统是否已将当前程置于上下文健康处置态**。若当前没有 `caution/critical` 处置指令，则默认优先压缩成一个结构化接续包提醒项。
+- 一旦系统已将当前程置于 `caution/critical` 处置态，当前程的目标就切换为“保住易丢信息并尽快 `clear_mind`”，此时允许多条粗略提醒项过桥；系统真正开启新一程后，第一步再复核、合并并删除冗余。
+- 除非确有必要，不要重复差遣牒已覆盖的内容。
+- 不要把长原始日志/大段 tool output 直接塞进接续包。
+
 ### 警告（黄色）
 
 当 `level === 'caution'` 时，驱动程序在**下一个** LLM 生成轮次自动插入一条 **role=user** 指导提示，并**将其持久化为普通用户消息**，以便 UI 将其渲染为正常的用户指令。

package/dist/minds/load.js

@@ -235,6 +235,16 @@ async function loadAgentMinds(agentId, dialog, options) {
     const agentHasPersonalMemoryTools = funcTools.some((t) => isPersonalMemoryToolName(t.name));
     const isSubdialog = dialog !== undefined && dialog.supdialog !== undefined;
     const taskdocMaintainerId = dialog && dialog instanceof dialog_1.SubDialog ? dialog.rootDialog.agentId : agent.id;
+    const contextHealthSnapshot = dialog?.getLastContextHealth();
+    const contextHealthPromptMode = contextHealthSnapshot &&
+        contextHealthSnapshot.kind === 'available' &&
+        contextHealthSnapshot.level === 'critical'
+        ? 'critical'
+        : contextHealthSnapshot &&
+            contextHealthSnapshot.kind === 'available' &&
+            contextHealthSnapshot.level === 'caution'
+            ? 'caution'
+            : 'normal';
     const promptdocContext = {
         language: workingLanguage,
         agentId: agent.id,
@@ -246,6 +256,7 @@ async function loadAgentMinds(agentId, dialog, options) {
         agentHasShellTools,
         agentHasReadonlyShell,
         shellSpecialistMemberIds,
+        contextHealthPromptMode,
     };
     const policyText = [
         (0, system_prompt_parts_1.buildShellPolicyPrompt)(promptdocContext),
@@ -260,6 +271,7 @@ async function loadAgentMinds(agentId, dialog, options) {
     const systemPrompt = (0, system_prompt_1.buildSystemPrompt)({
         language: workingLanguage,
         dialogScope: isSubdialog ? 'sideline' : 'mainline',
+        contextHealthPromptMode,
         agent,
         persona,
         knowledge,

package/dist/minds/system-prompt-parts.js

@@ -136,6 +136,26 @@ function buildShellPolicyPrompt(ctx) {
 }
 function getMemoryPromptCopy(ctx) {
     const runtimeMarkers = (0, inter_dialog_format_1.getRuntimeTransferMarkers)(ctx.language);
+    const contextHealthLineZh = ctx.contextHealthPromptMode === 'critical'
+        ? '当前上下文处于系统告急处置态：立即停止继续大实现/大阅读。本程只做保信息、提醒项提炼与 `clear_mind`。'
+        : ctx.contextHealthPromptMode === 'caution'
+            ? '当前上下文处于系统吃紧处置态：不要继续扩张上下文。本程优先提炼提醒项并尽快 `clear_mind`，不要继续大实现/大阅读。'
+            : '当前没有生效中的上下文健康处置指令，按正常工作流推进。';
+    const taskdocLogLineZh = ctx.contextHealthPromptMode === 'critical'
+        ? '不要把长日志/大段 tool output 直接塞进差遣牒；差遣牒只写结论+下一步；提醒项也只留可扫读摘录。当前是告急处置态：本程允许先保留多条粗略提醒项求稳，但不要在当前程提前做“新一程清醒复核”。系统真正开启新一程后，第一步再复核整理：删除冗余、纠正偏激/失真思路，再收敛成高质量提醒项。'
+        : ctx.contextHealthPromptMode === 'caution'
+            ? '不要把长日志/大段 tool output 直接塞进差遣牒；差遣牒只写结论+下一步；提醒项也只留可扫读摘录。当前是吃紧处置态：本程优先压缩/提炼提醒项并尽快 `clear_mind`；若一时来不及，可先保留多条粗略提醒项过桥，但不要在当前程提前做“新一程清醒复核”。系统真正开启新一程后，第一步再复核整理：删除冗余、纠正偏激/失真思路，再收敛成高质量提醒项。'
+            : '不要把长日志/大段 tool output 直接塞进差遣牒；差遣牒只写结论+下一步；提醒项也只留可扫读摘录。接续包提醒项默认应保持结构化、便于快速恢复。';
+    const contextHealthLineEn = ctx.contextHealthPromptMode === 'critical'
+        ? 'Current context is under system critical remediation: stop large implementations/reads immediately. In this course, do only preserve-info work, reminder distillation, and `clear_mind`.'
+        : ctx.contextHealthPromptMode === 'caution'
+            ? 'Current context is under system caution remediation: do not keep expanding context. In this course, prioritize reminder distillation and `clear_mind`; do not continue large implementations/reads.'
+            : 'There is no active context-health remediation instruction in effect right now; proceed with the normal workflow.';
+    const taskdocLogLineEn = ctx.contextHealthPromptMode === 'critical'
+        ? 'Do not paste long logs/tool outputs into Taskdoc; Taskdoc should record decisions + next steps; reminders should also keep only scannable excerpts. Current mode is critical remediation: in this course, rough multi-reminder bridge notes are acceptable, but do not perform the new-course “clear-headed review” early. Once the system actually starts the new course, the first step is to review/rewrite them: remove redundancy, correct biased or distorted bridge notes, then compress them into high-quality reminders.'
+        : ctx.contextHealthPromptMode === 'caution'
+            ? 'Do not paste long logs/tool outputs into Taskdoc; Taskdoc should record decisions + next steps; reminders should also keep only scannable excerpts. Current mode is caution remediation: in this course, prioritize compressing/distilling reminders and clearing soon; if needed, rough multi-reminder bridge notes are acceptable, but do not perform the new-course “clear-headed review” early. Once the system actually starts the new course, the first step is to review/rewrite them: remove redundancy, correct biased or distorted bridge notes, then compress them into high-quality reminders.'
+            : 'Do not paste long logs/tool outputs into Taskdoc; Taskdoc should record decisions + next steps; reminders should also keep only scannable excerpts. Keep continuation-package reminders structured and fast to resume from by default.';
     if (ctx.language === 'zh') {
         return {
             title: '### 记忆系统（重要）',
@@ -146,7 +166,7 @@ function getMemoryPromptCopy(ctx) {
             progressLine: '- 其中 `progress` 是全队共享公告牌：用于“阶段性进度快照”（关键决策/当前状态/下一步），不是流水账。',
             injectedTaskdocLine: '- 重要：差遣牒内容会被系统以内联形式注入到上下文中（本轮生成视角下即为最新；注入内容不包括全局约束）。需要回顾时请直接基于上下文里的差遣牒内容回顾与决策，不要试图用通用文件工具读取 `*.tsk/` 下的文件（会被拒绝）。',
             constraintsLine: '- 约定：`constraints` 只写任务特有的硬要求，不得写入系统提示/工具文档里已明确且由系统强制执行的通用规则（例如 `*.tsk/` 封装禁止通用文件工具）。一经发现重复，必须删除并告知用户。',
-            remindersLine: '- 提醒项（即编号提醒，工作集）：当前对话的高频工作记录/关键细节（偏私有，不作为全队公告）；默认保持少量（常见 1–3 条），优先 `update_reminder` 压缩/合并，不再需要就 `delete_reminder`。准备 `clear_mind` 开启新一程对话时，最好整理成“结构化接续包提醒项”；其中只保留差遣牒未覆盖、但恢复工作容易丢的细节（如第一步、关键定位、运行/验证信息、临时 ids/路径）。但若你已进入吃紧/告急、头脑不够清楚，允许先保留多条粗略提醒项把信息带过桥；新一程开始后的第一步，必须以清醒头脑复核并整理这些接续包/粗略提醒项：删除冗余、纠正偏激或失真的过桥思路、压缩成高质量提醒项。',
+            remindersLine: '- 提醒项（即编号提醒，工作集）：当前对话的高频工作记录/关键细节（偏私有，不作为全队公告）；默认保持少量（常见 1–3 条），优先 `update_reminder` 压缩/合并，不再需要就 `delete_reminder`。准备 `clear_mind` 开启新一程对话时，默认整理成“结构化接续包提醒项”；其中只保留差遣牒未覆盖、但恢复工作容易丢的细节（如第一步、关键定位、运行/验证信息、临时 ids/路径）。若系统已把当前程切到吃紧/告急处置态，则允许先保留多条粗略提醒项把信息带过桥；当前程只做保信息 + clear_mind；只有系统实际开启新一程后，第一步才是以清醒头脑复核并整理这些接续包/粗略提醒项：删除冗余、纠正偏激或失真的过桥思路、压缩成高质量提醒项。',
             teamMemoryLine: '- 团队记忆：稳定的团队约定/工程规约（跨任务共享）。',
             personalMemoryLine: '- 个人记忆：稳定的个人习惯/偏好与职责域知识；记忆会在每次生成时自动注入上下文，应保持少量且准确（关键文档/代码的精确路径 + 最小必要事实）。不要记录具体任务状态。',
             subdialogDutyLine: `你当前处于支线对话：此处不允许 \`change_mind\`。当你判断需要更新差遣牒（尤其是 progress 公告牌）时，请在合适时机直接诉请差遣牒维护人 \`@${ctx.taskdocMaintainerId}\` 执行更新，并给出你已合并好的“新全文/替换稿”（用于替换对应章节全文）。不要声称已更新，除非看到回执。`,
@@ -156,8 +176,8 @@ function getMemoryPromptCopy(ctx) {
             personalMemoryHintLine: `提示：你具备个人记忆工具（\`add_personal_memory\` / \`replace_personal_memory\` / \`drop_personal_memory\` / \`clear_personal_memory\`）。个人记忆仅对当前智能体可见，且系统会自动按成员隔离到 \`.minds/memory/individual/<member-id>/...\`；因此 \`path\` 不应包含你的成员 id（不要写 \`${ctx.agentId}/...\`）。首次创建时直接用 \`add_personal_memory\` 即可，目录会由系统自动创建。记忆会在每次生成时自动注入上下文：保持少量、保持准确、按“未来会一起更新的内容”合并；写稳定事实（关键路径 + 最小必要约定），不要写任务进度/当天状态；一旦你修改了相关文件或发现记忆过期/冲突，立刻用 \`replace_personal_memory\` 更新。`,
             subdialogWorkflowLine: `工作流：先做事 → 再提炼（\`update_reminder\`；必要时整理差遣牒更新提案并诉请 \`@${ctx.taskdocMaintainerId}\` 合并写入）→ 然后 \`clear_mind\` 清空噪音。`,
             mainlineWorkflowLine: '工作流：先做事 → 再提炼（`update_reminder` + `change_mind(progress)`）→ 然后 `clear_mind` 清空噪音。',
-            contextHealthLine: '系统会自动监控并提示上下文健康度：进入"吃紧"或"告急"状态时会插入用户可见提示。当收到此类提示时，立刻停止继续大实现/大阅读；先提炼，再 clear。',
-            taskdocLogLine: '不要把长日志/大段 tool output 直接塞进差遣牒；差遣牒只写结论+下一步；提醒项也只留可扫读摘录。若头脑清楚，接续包提醒项应尽量结构化、便于快速恢复；若已吃紧/告急，允许先保留多条粗略提醒项求稳，但进入新一程后必须先复核整理：删除冗余、纠正偏激/失真思路，再收敛成高质量提醒项。',
+            contextHealthLine: contextHealthLineZh,
+            taskdocLogLine: taskdocLogLineZh,
         };
     }
     return {
@@ -169,7 +189,7 @@ function getMemoryPromptCopy(ctx) {
         progressLine: '- Taskdoc `progress` is the team’s shared bulletin board: distilled milestone snapshots (key decisions/current status/next steps), not raw logs.',
         injectedTaskdocLine: '- Important: the Taskdoc content is injected inline into the context (the latest as of this generation; injected content excludes global constraints). Review the injected Taskdoc instead of trying to read files under `*.tsk/` via general file tools (they will be rejected).',
         constraintsLine: '- Convention: Taskdoc `constraints` must contain task-specific requirements only; do not include global, system-enforced rules already stated in system prompt/tool docs (e.g. `.tsk/` encapsulation bans general file tools). If duplication is found, you MUST remove it and notify the user.',
-        remindersLine: '- Reminders (i.e. numbered reminders, working set): your high-frequency per-dialog worklog + critical details (not a team bulletin board); keep it small by default (often 1–3 items), prefer `update_reminder` to compress/merge, and delete when obsolete. When preparing `clear_mind` to start a new course, prefer a structured continuation-package reminder that keeps only details not already covered by Taskdoc but easy to lose during resume (first step, key pointers, run/verify info, volatile ids/paths). But if you are already in caution/critical and not clear-headed enough, you may carry multiple rough reminders across the course boundary first; at the start of the new course, your first step must be to review and rewrite those continuation/rough reminders with a clear head: remove redundancy, correct biased or distorted bridge notes, and compress them into high-quality reminders.',
+        remindersLine: '- Reminders (i.e. numbered reminders, working set): your high-frequency per-dialog worklog + critical details (not a team bulletin board); keep it small by default (often 1–3 items), prefer `update_reminder` to compress/merge, and delete when obsolete. When preparing `clear_mind` to start a new course, default to a structured continuation-package reminder that keeps only details not already covered by Taskdoc but easy to lose during resume (first step, key pointers, run/verify info, volatile ids/paths). If the system has already put the current course into caution/critical remediation, rough multi-reminder bridge notes are acceptable; in the current course, only preserve volatile information and clear_mind. Only after the system actually starts the new course does the mandatory first step become reviewing and rewriting those continuation/rough reminders with a clear head: remove redundancy, correct biased or distorted bridge notes, and compress them into high-quality reminders.',
         teamMemoryLine: '- Team memory: stable shared conventions (cross-task).',
         personalMemoryLine: '- Personal memory: stable personal habits/preferences and responsibility-scope knowledge. Memory is automatically injected into context on each generation: keep it small and accurate (exact key doc/code paths + minimal key facts); do not store per-task state.',
         subdialogDutyLine: `You are currently in a subdialog: \`change_mind\` is not allowed here. When Taskdoc should be updated (especially the shared progress bulletin board), tellask the Taskdoc maintainer \`@${ctx.taskdocMaintainerId}\` with a fully merged replacement draft (full-section replacement). Do not claim it is updated until you see a receipt.`,
@@ -179,8 +199,8 @@ function getMemoryPromptCopy(ctx) {
         personalMemoryHintLine: `Hint: you have personal-memory tools (\`add_personal_memory\` / \`replace_personal_memory\` / \`drop_personal_memory\` / \`clear_personal_memory\`). Personal memory is private to the current agent and is automatically isolated under \`.minds/memory/individual/<member-id>/...\`; therefore \`path\` MUST NOT include your member id (do not write \`${ctx.agentId}/...\`). For first-time setup, just call \`add_personal_memory\`—the directory will be created automatically. Memory is automatically injected into context on each generation: keep it small, keep it accurate, and group facts that are updated together. Store stable facts (exact key paths + minimal contracts), not daily state/progress. If you changed those files or detect staleness/conflicts, immediately \`replace_personal_memory\` to keep it accurate.`,
         subdialogWorkflowLine: `Workflow: do work → distill (\`update_reminder\`; when Taskdoc needs updates, draft a merged replacement and ask \`@${ctx.taskdocMaintainerId}\`) → then \`clear_mind\` to drop noise.`,
         mainlineWorkflowLine: 'Workflow: do work → distill (`update_reminder` + `change_mind(progress)`) → then `clear_mind` to drop noise.',
-        contextHealthLine: 'System will automatically monitor and alert on context health: yellow (caution/"Caution") or red (critical/"Critical") will insert a user-visible prompt. When you receive such alerts, immediately stop large implementations/reads; distill first, then clear.',
-        taskdocLogLine: 'Do not paste long logs/tool outputs into Taskdoc; Taskdoc should record decisions + next steps; reminders should also keep only scannable excerpts. When clear-headed, keep continuation-package reminders structured and fast to resume from; when already in caution/critical, multiple rough reminders are acceptable as a bridge, but in the new course you must first review and rewrite them: remove redundancy, correct biased or distorted bridge notes, then compress them into high-quality reminders.',
+        contextHealthLine: contextHealthLineEn,
+        taskdocLogLine: taskdocLogLineEn,
     };
 }
 function buildMemorySystemPrompt(ctx) {

package/dist/minds/system-prompt.js

@@ -38,16 +38,36 @@ function formatTeamIntro(team, selfAgentId, language) {
 function pickLocalized(language, localized) {
     return language === 'zh' ? localized.zh : localized.en;
 }
-function buildFbrContextHealthScopeRule(scope, language) {
+function buildFbrContextHealthScopeRule(scope, language, contextHealthPromptMode) {
+    if (contextHealthPromptMode === 'critical') {
+        return pickLocalized(language, {
+            zh: scope === 'mainline'
+                ? '- 当前上下文处于系统告急处置态：本程禁止发起 FBR。先按处置要求保信息、维护提醒项，并立即 `clear_mind`；系统真正开启新一程并完成接续包复核后，再恢复 FBR 与差遣牒更新。'
+                : '- 当前上下文处于系统告急处置态：本程禁止发起 FBR。先按处置要求保信息、维护提醒项，并立即 `clear_mind`；系统真正开启新一程并完成接续包复核后，再恢复 FBR。',
+            en: scope === 'mainline'
+                ? '- Current context is under system critical remediation: do not start FBR in this course. Preserve volatile information, maintain reminders, and clear immediately; resume FBR and Taskdoc updates only after the system starts the new course and the continuation package has been reviewed.'
+                : '- Current context is under system critical remediation: do not start FBR in this course. Preserve volatile information, maintain reminders, and clear immediately; resume FBR only after the system starts the new course and the continuation package has been reviewed.',
+        });
+    }
+    if (contextHealthPromptMode === 'caution') {
+        return pickLocalized(language, {
+            zh: scope === 'mainline'
+                ? '- 当前上下文处于系统吃紧处置态：本程不要发起 FBR。先按处置要求提炼提醒项并尽快 `clear_mind`；系统真正开启新一程并完成接续包复核后，再恢复 FBR 与差遣牒更新。'
+                : '- 当前上下文处于系统吃紧处置态：本程不要发起 FBR。先按处置要求提炼提醒项并尽快 `clear_mind`；系统真正开启新一程并完成接续包复核后，再恢复 FBR。',
+            en: scope === 'mainline'
+                ? '- Current context is under system caution remediation: do not start FBR in this course. Distill reminders and clear soon; resume FBR and Taskdoc updates only after the system starts the new course and the continuation package has been reviewed.'
+                : '- Current context is under system caution remediation: do not start FBR in this course. Distill reminders and clear soon; resume FBR only after the system starts the new course and the continuation package has been reviewed.',
+        });
+    }
     if (scope === 'mainline') {
         return pickLocalized(language, {
-            zh: '- 系统会持续自动监控上下文健康度：没有吃紧/告急提示时，可以安全进行 FBR。如果有吃紧/告急提示，系统会提醒你，应先处理（提炼 + clear_mind）。随后基于当前可观测事实调用 \\`change_mind\\` 更新差遣牒，体现任务最新进展情况。FBR 自诉请正文不要冗余包含差遣牒已有信息。',
-            en: '- Before every FBR, the system will automatically alert on context health status: if there are no yellow/red alerts, you can safely proceed with FBR. If there are yellow/red alerts, handle them first (distill + clear_mind). Then call \\`change_mind\\` based on currently observable facts to update the Taskdoc with the latest task progress; do not redundantly include information already present in the Taskdoc in the FBR FBR body.',
+            zh: '- 当前没有生效中的上下文健康处置指令，可以按正常流程进行 FBR。完成 FBR 后，基于当前可观测事实调用 \\`change_mind\\` 更新差遣牒，体现任务最新进展情况。FBR 自诉请正文不要冗余包含差遣牒已有信息。',
+            en: '- There is no active context-health remediation instruction in effect, so FBR may proceed normally. After FBR, call \\`change_mind\\` based on currently observable facts to update the Taskdoc with the latest progress; do not redundantly include information already present in the Taskdoc in the FBR body.',
         });
     }
     return pickLocalized(language, {
-        zh: '- 系统会持续自动监控上下文健康度：没有吃紧/告急提示时，可以安全进行 FBR。如果有吃紧/告急提示，系统会提醒你，应先处理。随后基于当前可观测事实分析是否与差遣牒内容存在差异，并将发现的情况包含在 FBR 自诉请正文中。',
-        en: '- Before every FBR, the system will automatically alert on context health status: if there are no yellow/red alerts, you can safely proceed with FBR. If there are yellow/red alerts, handle them first. Then analyze whether currently observable facts differ from the Taskdoc, and include the findings in the FBR FBR body.',
+        zh: '- 当前没有生效中的上下文健康处置指令，可以按正常流程进行 FBR。随后基于当前可观测事实分析是否与差遣牒内容存在差异，并将发现的情况包含在 FBR 自诉请正文中。',
+        en: '- There is no active context-health remediation instruction in effect, so FBR may proceed normally. Then analyze whether currently observable facts differ from the Taskdoc, and include the findings in the FBR body.',
     });
 }
 function buildFbrPhaseContract(language) {
@@ -181,8 +201,8 @@ function buildTellaskCollaborationProtocol(language, dialogScope) {
     }
     return lines.join('\n');
 }
-function buildFbrGuidelines(language, dialogScope) {
-    const fbrContextHealthRule = buildFbrContextHealthScopeRule(dialogScope, language);
+function buildFbrGuidelines(language, dialogScope, contextHealthPromptMode) {
+    const fbrContextHealthRule = buildFbrContextHealthScopeRule(dialogScope, language, contextHealthPromptMode);
     const fbrPhaseContract = buildFbrPhaseContract(language);
     const lines = pickLocalized(language, {
         zh: [
@@ -258,7 +278,7 @@ function buildFunctionToolRules(language, funcToolRulesText) {
 }
 function buildSystemPrompt(input) {
     const collaborationProtocol = buildTellaskCollaborationProtocol(input.language, input.dialogScope);
-    const fbrGuidelines = buildFbrGuidelines(input.language, input.dialogScope);
+    const fbrGuidelines = buildFbrGuidelines(input.language, input.dialogScope, input.contextHealthPromptMode);
     const tellaskInteractionRules = buildTellaskInteractionRules(input.language);
     const functionToolRules = buildFunctionToolRules(input.language, input.funcToolRulesText);
     if (input.language === 'zh') {

package/dist/shared/i18n/driver-messages.js

@@ -31,13 +31,13 @@ function formatNewCourseStartPrompt(language, args) {
             ? `你刚清理头脑，开启了第 ${args.nextCourse} 程对话。`
             : `系统因上下文已告急（critical）而自动开启了第 ${args.nextCourse} 程对话。`;
         return (`${prefix} ` +
-            '第一步先复核并整理接续包提醒项：以清醒头脑删除冗余、纠正偏激或失真的过桥思路、压缩成高质量提醒项；再继续推进任务。');
+            '现在已经进入新一程：第一步先复核并整理接续包提醒项，以清醒头脑删除冗余、纠正偏激或失真的过桥思路、压缩成高质量提醒项；再继续推进任务。');
     }
     const prefix = args.source === 'clear_mind'
         ? `This is course #${args.nextCourse} of the dialog. You just cleared your mind.`
         : `System auto-started course #${args.nextCourse} of the dialog because context health is critical.`;
     return (`${prefix} ` +
-        'Your first step is to review and rewrite any continuation-package reminders with a clear head: remove redundancy, correct biased or distorted bridge notes, compress them into high-quality reminders, and then continue the task.');
+        'You are now in a new course: your first step is to review and rewrite any continuation-package reminders with a clear head, remove redundancy, correct biased or distorted bridge notes, compress them into high-quality reminders, and then continue the task.');
 }
 function formatReminderItemGuide(language, index, content, options) {
     function isRecord(value) {
@@ -195,9 +195,9 @@ function formatAgentFacingContextHealthV3RemediationGuide(language, args) {
                 '',
                 '影响：对话历史中的工具调用/结果信息很多已经过时，成为你的思考负担。',
                 '',
-                '行动：先尽量保住易丢信息；如头脑还清楚，优先整理成结构化接续包提醒项（下一步行动 + 关键定位信息 + 运行/验证信息 + 容易丢的临时细节）。',
+                '行动：先尽量保住易丢信息，并优先把已经掌握的事实压缩成结构化接续包提醒项（下一步行动 + 关键定位信息 + 运行/验证信息 + 容易丢的临时细节）；若不额外继续阅读/分析就做不到，允许先记成多条粗略提醒项。',
                 '',
-                '若你已经发乱，允许先记成多条粗略提醒项带到新一程；求稳比求整洁更重要。进入新一程后，第一步就是复核并整理提醒项：删除冗余、纠正偏激/失真思路、压缩成高质量提醒项，然后再继续工作。接续包只保留差遣牒未覆盖、但恢复工作会丢的信息。然后主动 clear_mind，开启新一程对话继续工作。',
+                '当前已处于吃紧处置阶段：不要继续扩张上下文，也不要提前进入“按接续包做清醒复核”的模式；那是系统真正开启新一程后的第一步。接续包只保留差遣牒未覆盖、但恢复工作会丢的信息。然后主动 clear_mind，开启新一程对话继续工作。',
                 '',
                 '操作：',
                 '- update_reminder({ "reminder_no": 1, "content": "..." })（推荐）',
@@ -209,14 +209,14 @@ function formatAgentFacingContextHealthV3RemediationGuide(language, args) {
             '',
             `系统最多再提醒你 ${args.promptsRemainingAfterThis} 次，之后将自动清理头脑开启新一程对话。`,
             '',
-            '行动：尽快保住易丢信息，然后 clear_mind。能整理就整理成结构化接续包提醒项；若已经发乱，先保留多条粗略提醒项也可以。',
+            '行动：尽快保住易丢信息，然后 clear_mind。若能直接基于已掌握事实压缩成结构化接续包提醒项，就这样做；否则先保留多条粗略提醒项也可以。',
             '',
             '操作：',
             '- update_reminder({ "reminder_no": 1, "content": "..." })',
             '- add_reminder({ "content": "..." })',
             '- clear_mind({})',
             '',
-            '接续包要点：下一步行动 + 关键定位信息 + 运行验证方式 + 容易丢的临时细节；不要重复差遣牒已有内容。若先带多条粗略提醒项过桥，新一程第一步就要重新审视并整理：删除冗余、纠正偏激/失真思路、压缩成高质量提醒项。',
+            '接续包要点：下一步行动 + 关键定位信息 + 运行验证方式 + 容易丢的临时细节；不要重复差遣牒已有内容。当前处于告急处置阶段时，不要提前做“新一程清醒复核”；系统真正开启新一程后，第一步才是重新审视并整理：删除冗余、纠正偏激/失真思路、压缩成高质量提醒项。',
         ].join('\n');
     }
     if (args.kind === 'caution' && args.mode === 'soft') {
@@ -225,9 +225,9 @@ function formatAgentFacingContextHealthV3RemediationGuide(language, args) {
             '',
             'Impact: stale call/results in dialog history are creating cognitive noise.',
             '',
-            'Action: first preserve easy-to-lose information; if you are still clear-headed, prefer a structured continuation-package reminder (next step + key pointers + run/verify info + easy-to-lose volatile details).',
+            'Action: first preserve easy-to-lose information, and prefer compressing already observed facts into a structured continuation-package reminder (next step + key pointers + run/verify info + easy-to-lose volatile details). If that cannot be done without more reading/analysis, rough multi-reminder carry-over is acceptable.',
             '',
-            'If you are already muddled, multiple rough reminders are acceptable as a bridge; survival matters more than neatness. In the new course, your first step is to review and rewrite reminders: remove redundancy, correct biased or distorted bridge notes, compress them into high-quality reminders, and only then continue. Keep only details not already covered by Taskdoc. Then proactively clear_mind to start a new dialog course.',
+            'You are already in caution remediation for the current course, so do not keep expanding context and do not switch early into “clear-headed continuation-package review” mode; that is the first step only after the system actually starts the new course. Keep only details not already covered by Taskdoc. Then proactively clear_mind to start a new dialog course.',
             '',
             'Operations:',
             '- update_reminder({ "reminder_no": 1, "content": "..." })',
@@ -239,14 +239,14 @@ function formatAgentFacingContextHealthV3RemediationGuide(language, args) {
         '',
         `System will remind you ${args.promptsRemainingAfterThis} more time(s), then automatically clear mind.`,
         '',
-        'Action: preserve easy-to-lose information, then clear_mind. Prefer a structured continuation-package reminder if you can still think clearly; otherwise multiple rough reminders are acceptable as a bridge.',
+        'Action: preserve easy-to-lose information, then clear_mind. If you can compress already observed facts into one structured continuation-package reminder without more reading/analysis, do that; otherwise multiple rough reminders are acceptable as a bridge.',
         '',
         'Operations:',
         '- update_reminder({ "reminder_no": 1, "content": "..." })',
         '- add_reminder({ "content": "..." })',
         '- clear_mind({})',
         '',
-        'Continuation package: next step + key pointers + run/verify info + easy-to-lose volatile details. Do not duplicate Taskdoc content. If you bridge with multiple rough reminders first, reconcile them at the start of the new course by removing redundancy, correcting biased or distorted bridge notes, and compressing them into high-quality reminders.',
+        'Continuation package: next step + key pointers + run/verify info + easy-to-lose volatile details. Do not duplicate Taskdoc content. During critical remediation in the current course, do not start the new-course cleanup early; once the system actually starts the new course, the first step is to reconcile rough bridge reminders by removing redundancy, correcting biased or distorted bridge notes, and compressing them into high-quality reminders.',
     ].join('\n');
 }
 function formatDomindsNoteDirectSelfCall(language) {

package/dist/tools/ctrl.js

@@ -427,8 +427,8 @@ exports.clearMindTool = {
     name: 'clear_mind',
     description: 'Clear dialog mind and start a new course, optionally adding a reminder.',
     descriptionI18n: {
-        en: 'Clear dialog mind and start a new course. If needed, pass reminder_content so you can resume quickly. Prefer a structured "Continuation Package" when still clear-headed; if context is already degraded, rough bridge notes are acceptable and can be reorganized in the new course.',
-        zh: '清理头脑并开启新一程对话。若需要，可传入 reminder_content 以便快速接续。头脑还清楚时，优先整理成结构化"接续包"；若上下文已吃紧到发乱，粗略过桥笔记也可以，进入新一程后再重新整理。',
+        en: 'Clear dialog mind and start a new course. If needed, pass reminder_content so you can resume quickly. Default to a structured "Continuation Package"; if the current course is already under system caution/critical remediation, rough bridge notes are acceptable and can be reorganized in the new course.',
+        zh: '清理头脑并开启新一程对话。若需要，可传入 reminder_content 以便快速接续。默认整理成结构化"接续包"；若当前程已被系统置于吃紧/告急处置态，粗略过桥笔记也可以，进入新一程后再重新整理。',
     },
     parameters: {
         type: 'object',

package/dist/tools/prompts/control/en/principles.md

@@ -35,7 +35,7 @@ Reminders are **session-level** temporary working-set information for tracking p
 - Automatically cleared after dialog ends
 - Can be added, modified, or deleted at any time
 - Should stay compact, scannable, and directly actionable by default
-- Before `clear_mind`, prefer a structured continuation-package reminder when still clear-headed; if already degraded, rough multi-reminder carry-over is acceptable
+- Before `clear_mind`, default to a structured continuation-package reminder; if the current course is already under system remediation, rough multi-reminder carry-over is acceptable
 
 **Difference from memory:**
 | Feature | reminder | memory |
@@ -88,7 +88,7 @@ Taskdoc is a **task contract** defining goals, constraints, and progress.
 
 - Keep concise: reminders are often 1-3 items; prefer `update_reminder` to compress/merge
 - Separate carriers: shared decisions/status belong in Taskdoc; reminders keep local resume details
-- Collapse before clearing: if you are still clear-headed, rewrite into a structured continuation-package reminder; if already degraded, rough multi-reminder carry-over is acceptable but must be reconciled first in the new course
+- Collapse before clearing: default to a structured continuation-package reminder; if the current course is already under system remediation, rough multi-reminder carry-over is acceptable but must be reconciled first in the new course
 - Avoid raw-material dumps: do not paste long logs or large tool outputs into reminders
 
 ## Limitations and Notes

package/dist/tools/prompts/control/en/tools.md

@@ -199,7 +199,7 @@ message: <error message>
 ## Reminder Content Guidance
 
 - Normal reminders should stay concise, fresh, and directly actionable; often 1-3 items total
-- For a continuation package, prefer structured notes when clear-headed: next step, key pointers, run/verify, easy-to-lose volatile details
-- If already in caution/critical and muddled, rough multi-reminder bridge notes are acceptable; reconcile them first in the new course
+- For a continuation package, use structured notes by default: next step, key pointers, run/verify, easy-to-lose volatile details
+- If the current course is already under caution/critical remediation, rough multi-reminder bridge notes are acceptable; in the current course only preserve info + `clear_mind`, and reconcile them as the first step only after the system actually starts the new course
 - Keep only details not already covered by Taskdoc; do not repeat team-shared status
 - Do not paste long logs, large tool outputs, or raw material into reminders

package/dist/tools/prompts/control/zh/index.md

@@ -55,7 +55,7 @@ control 是 Dominds 的**对话控制工具集**，用于管理对话状态、
 - 标记待处理事项
 - 追踪当前下一步/阻塞
 - 记录阻塞问题
-- 在 `clear_mind` 前承载接续包；若已发乱，可先带多条粗略提醒项过桥
+- 在 `clear_mind` 前承载接续包；若当前程已被系统置于吃紧/告急处置态，可先带多条粗略提醒项过桥
 
 ### 差遣牒
 

package/dist/tools/prompts/control/zh/principles.md

@@ -35,7 +35,7 @@
 - 对话结束后自动清除
 - 可随时添加、修改、删除
 - 默认应保持少量、可扫读、可直接指导下一步
-- 准备 `clear_mind` 时，头脑清楚则优先压缩成结构化接续包提醒项；若已吃紧/告急，也可先保留多条粗略提醒项过桥
+- 准备 `clear_mind` 时，默认优先压缩成结构化接续包提醒项；若当前程已被系统置于吃紧/告急处置态，也可先保留多条粗略提醒项过桥
 
 **与 memory 的区别：**
 | 特性 | reminder | memory |
@@ -76,7 +76,7 @@
 
 - **当前工作集**：当前下一步、阻塞、关键定位信息
 - **易丢细节**：临时路径、id、命令、样例输入
-- **换程接续**：在 `clear_mind` 前整理接续包；若已发乱，可先保留多条粗略提醒项
+- **换程接续**：在 `clear_mind` 前整理接续包；若当前程已被系统置于吃紧/告急处置态，可先保留多条粗略提醒项
 
 ### 2. 差遣牒更新时机
 
@@ -88,7 +88,7 @@
 
 - 保持简洁：默认提醒项常见 1–3 条，优先 `update_reminder` 压缩/合并
 - 区分载体：团队共享决策/状态写入差遣牒，提醒项只留本地恢复所需细节
-- 换程前收束：若头脑清楚，优先整理成结构化接续包提醒项；若已发乱，先保留多条粗略提醒项过桥也可以，但新一程第一步就要收敛
+- 换程前收束：默认优先整理成结构化接续包提醒项；若系统已把当前程切到吃紧/告急处置态，则先保留多条粗略提醒项过桥也可以；当前程只做保信息 + `clear_mind`，系统真正开启新一程后第一步再收敛
 - 拒绝原料堆积：不要把长日志/大段 tool output 直接塞进提醒项
 
 ## 限制与注意事项

package/dist/tools/prompts/control/zh/tools.md

@@ -26,7 +26,7 @@
 **适用：**
 
 - 添加新的临时工作集条目
-- 若准备 `clear_mind`，也可新建接续包提醒项；若已发乱，先记粗略过桥信息也可以
+- 若准备 `clear_mind`，也可新建接续包提醒项；若当前程已被系统置于吃紧/告急处置态，先记粗略过桥信息也可以
 
 **参数：**
 
@@ -70,7 +70,7 @@ deleted_at: <删除时间戳>
 **适用：**
 
 - 压缩/合并现有提醒项
-- 把换程前需要保留的信息整理成接续包；若已发乱，也可先保留多条粗略提醒项
+- 把换程前需要保留的信息整理成接续包；若当前程已被系统置于吃紧/告急处置态，也可先保留多条粗略提醒项
 - 删除已写入差遣牒、无需在提醒项重复保留的内容
 
 **参数：**
@@ -199,7 +199,7 @@ message: <错误消息>
 ## 提醒项内容建议
 
 - 默认提醒项应保持短、新、能直接指导下一步，常见 1–3 条
-- 若用于接续包，头脑清楚时优先结构化内容：下一步行动、关键定位、运行/验证、容易丢的临时细节
-- 若已吃紧/告急，先保留多条粗略提醒项也可以；进入新一程后第一步就收敛整理
+- 若用于接续包，默认优先结构化内容：下一步行动、关键定位、运行/验证、容易丢的临时细节
+- 若已吃紧/告急，先保留多条粗略提醒项也可以；当前程只做保信息 + `clear_mind`，系统真正开启新一程后第一步再收敛整理
 - 接续包只保留差遣牒未覆盖的细节；不要重复团队共享状态
 - 不要把长日志、大段 tool output、原始材料直接塞进提醒项

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dominds",
-  "version": "1.7.1",
+  "version": "1.7.2",
   "description": "DevOps Mindsets — Sustainable Agentic Product Lifecycle",
   "type": "commonjs",
   "private": false,