mocode-ai 1.1.7 → 1.1.9

package/dist/session/notes.js

@@ -0,0 +1,107 @@
+// session/notes.ts - Session Notepad(notes.md)的 plan 状态单一事实源。
+// 负责:canonical 文件路径、plan 结构化模型、markdown 渲染、以及"只替换活跃 ## Plan: 段、
+// 保留其它笔记段"的写入。被 plan_update 工具、agent core(事件驱动重同步/nag)、
+// config(notepad 索引 / compact 重注入)共用,避免路径与段解析逻辑散落多处。
+import fs from 'node:fs';
+import path from 'node:path';
+import { getSandboxRoot } from '../sandbox/root.js';
+import { getCurrentSessionId } from './state.js';
+/** 单个活跃 plan 允许的最大步骤数(防无限膨胀,对齐 Claude Code TodoWrite 的 20 上限)。 */
+export const PLAN_MAX_STEPS = 20;
+/** 当前会话 notes.md 的绝对路径;无会话时返 null。 */
+export function getNotesFilePath(sessionId = getCurrentSessionId()) {
+    if (!sessionId)
+        return null;
+    const root = getSandboxRoot() ?? process.cwd();
+    return path.join(root, '.mocode', 'sessions', sessionId, 'notes.md');
+}
+/** notes.md 的 mtime(ms),不存在或不可读返 null。供 core 判断"本步是否改动了 notes.md"。 */
+export function getNotesMtime(sessionId = getCurrentSessionId()) {
+    const p = getNotesFilePath(sessionId);
+    if (!p)
+        return null;
+    try {
+        return fs.statSync(p).mtimeMs;
+    }
+    catch {
+        return null;
+    }
+}
+/**
+ * 把 plan 渲染为 canonical markdown 段(不含首尾多余空行)。
+ * 复选框格式 `- [ ] N.` / `- [x] N.` 与 repl 状态栏 chip 的计数正则严格对齐;
+ * in_progress 步骤追加 ` ◀ 当前` 后缀(仍计为未完成),让 compact 重注入后模型能直接续上。
+ * 全部完成时标题写为 `## Done:`(自动结算),extractActivePlanSection 随之返回 null。
+ */
+export function renderPlanSection(plan) {
+    const allDone = plan.steps.length > 0 && plan.steps.every((s) => s.status === 'completed');
+    const header = allDone ? `## Done: ${plan.title}` : `## Plan: ${plan.title}`;
+    const lines = [header];
+    if (plan.goal)
+        lines.push(`Goal: ${plan.goal}`);
+    lines.push('', '### Steps');
+    plan.steps.forEach((s, i) => {
+        const box = s.status === 'completed' ? '[x]' : '[ ]';
+        const suffix = s.status === 'in_progress'
+            ? `  ◀ ${(s.activeForm ?? '').trim() || 'current'}`
+            : '';
+        lines.push(`- ${box} ${i + 1}. ${s.content}${suffix}`);
+    });
+    const done = plan.steps.filter((s) => s.status === 'completed').length;
+    lines.push('', '### Progress', `- ${done}/${plan.steps.length} steps complete`);
+    return lines.join('\n');
+}
+/** 读取当前活跃 `## Plan:` 段的标题;无活跃 plan 返 null。 */
+export function readActivePlanTitle(sessionId = getCurrentSessionId()) {
+    const p = getNotesFilePath(sessionId);
+    if (!p)
+        return null;
+    try {
+        const normalized = fs.readFileSync(p, 'utf8').replace(/\r\n?/g, '\n');
+        const headerLine = normalized.split('\n').find((l) => /^## Plan:\s*.+$/.test(l));
+        return headerLine?.match(/^## Plan:\s*(.+)$/)?.[1].trim() ?? null;
+    }
+    catch {
+        return null;
+    }
+}
+/**
+ * 把 plan 写入 notes.md:替换已有活跃 `## Plan:` 段(原位),否则插到文件顶部;
+ * 其它所有笔记段(findings / Open Questions / 已结算 ## Done:)原样保留。
+ * 文件不存在则创建(含父目录)。返回 { path, settled } 或 { error }。
+ */
+export function writePlanToNotes(plan, sessionId = getCurrentSessionId()) {
+    const p = getNotesFilePath(sessionId);
+    if (!p)
+        return { error: 'no active session' };
+    const section = renderPlanSection(plan);
+    let existing = '';
+    try {
+        existing = fs.readFileSync(p, 'utf8').replace(/\r\n?/g, '\n');
+    }
+    catch {
+        existing = '';
+    }
+    let next;
+    const lines = existing.split('\n');
+    const start = lines.findIndex((l) => /^## Plan:\s*.+$/.test(l));
+    if (start >= 0) {
+        const endOffset = lines.slice(start + 1).findIndex((l) => /^##\s/.test(l));
+        const end = endOffset < 0 ? lines.length : start + 1 + endOffset;
+        const before = lines.slice(0, start).join('\n').replace(/\s+$/, '');
+        const after = lines.slice(end).join('\n').replace(/^\s+/, '');
+        next = [before, section, after].filter((s) => s.length > 0).join('\n\n') + '\n';
+    }
+    else {
+        const rest = existing.trim();
+        next = rest ? `${section}\n\n${rest}\n` : `${section}\n`;
+    }
+    try {
+        fs.mkdirSync(path.dirname(p), { recursive: true });
+        fs.writeFileSync(p, next, 'utf8');
+    }
+    catch (e) {
+        return { error: e instanceof Error ? e.message : String(e) };
+    }
+    return { path: p, settled: plan.steps.length > 0 && plan.steps.every((s) => s.status === 'completed') };
+}

package/dist/session/scheduler.js

@@ -1,159 +1,155 @@
-// Context Budget Scheduler(执行层):评估五区预算并执行可落地的动作。
-//
-//   agent/core.ts (步前)              repl /compact 命令(手动)
-//     ↓ runScheduler(history, step)    ↓ manualCompact(history, focus?)
-//   session/scheduler.ts(本文件)
-//     ├─ evaluateBudget(history, window, step) → BudgetReport
-//     ├─ scheduleActions(report) → warn | compact_history
-//     └─ 执行 actions:
-//          - warn:            仅写日志
-//          - compact_history: 调 maybeCompact(history, report)
-//
-// push-time cap / relevance / lifecycle / age-aware sweep 在进入 scheduler 前独立完成，
-// scheduler 不再生成无法执行的 Cold/Hot tool action。每次决策写入 actionLog，供 /context 调试。
-// manualCompact 与自动路径共享 scheduleActions，但用户显式触发时强制追加 compact_history。
-//
-// 开关:
-//   - config.contextBudget !== false(默认 true):agent 调 runScheduler
-//   - 关时 agent 仍走原 maybeCompact(history)无 report 路径,零行为变化
-//   - 手动 /compact 走 manualCompact;关时退化直接调 compactHistory(history, { focus })
-import { evaluateBudget, scheduleActions, formatReport, } from '../context/budget.js';
+// Context Budget Scheduler: normal operation is metadata-only. History content is
+// rewritten only after real occupancy reaches the shared pressure threshold.
+import { evaluateBudget, scheduleActions, formatReport, DEFAULT_BUDGET_POLICY, } from '../context/budget.js';
 import { chatTools } from '../llm/index.js';
 import { config } from '../config/index.js';
 import { maybeCompact, contextState } from './compact.js';
 import * as layout from '../ui/layout.js';
 import { ui } from '../ui/theme.js';
 import { pruneStaleArtifacts, refreshArtifactFreshness } from '../context/artifacts.js';
-/** 创建 runAgentCore 闭包持有的 scheduler(每次 agent 启动一个新实例)。 */
+import { pruneSuperseded } from '../context/relevance.js';
+import { createAgeAwareEncodingState } from '../context/age-aware.js';
+function atPressure(report) {
+    return Math.max(report.rawTotal, report.total)
+        >= report.window * DEFAULT_BUDGET_POLICY.pressureTriggerRatio;
+}
+function emptyPressure(report) {
+    return {
+        triggered: false,
+        before: report.total,
+        after: report.total,
+        superseded: 0,
+        staleArtifacts: 0,
+        encodedLogsAndSearches: 0,
+    };
+}
+/** One scheduler instance is owned by one agent run. */
 export function createBudgetScheduler(state = contextState) {
-    const obs = {
+    const evaluate = (history, step, activeTools) => evaluateBudget(history, config.contextWindowTokens, step, state.correction, activeTools);
+    const scheduler = {
         lastRunLog: null,
         async runStep(history, step, activeTools = chatTools) {
-            // Re-check file-backed hashes first, then discard stale/rebuildable facts before budgeting.
+            // External file changes and mutations only update artifact metadata here.
             refreshArtifactFreshness(state, history);
-            pruneStaleArtifacts(state, history);
-            const report = evaluateBudget(history, config.contextWindowTokens, step, state.correction, activeTools);
+            const report = evaluate(history, step, activeTools);
+            const pressure = emptyPressure(report);
+            pressure.triggered = atPressure(report);
+            if (pressure.triggered) {
+                // A single 80% pressure event owns every history rewrite. Run all enabled
+                // low-cost cleanup first, then always compact; do not introduce per-stage
+                // thresholds or stop early when one stage happens to cross below 80%.
+                if (config.contextRelprune) {
+                    pressure.superseded = pruneSuperseded(history, report.hotBoundary);
+                }
+                pressure.staleArtifacts = pruneStaleArtifacts(state, history, report.hotBoundary);
+                if (config.contextOptimize) {
+                    const ageAware = createAgeAwareEncodingState(history);
+                    pressure.encodedLogsAndSearches = ageAware.sweepPressure(history, report.hotBoundary);
+                }
+                pressure.after = evaluate(history, step, activeTools).total;
+            }
+            // Use the trigger report intentionally: cleanup may reduce the current estimate,
+            // but crossing 80% commits this step to compacting for maximum token savings.
             const actions = scheduleActions(report);
             let compactHistoryCalled = false;
             let historyRebuilt = false;
-            for (const a of actions) {
-                if (a.kind === 'warn') {
-                    // system 超:写一行提示(配置漂移应由用户处理,不是调度器压)
-                    layout.contentWrite(`  ${ui.accent}●${ui.reset} ${ui.yellow}调度器警告 [${a.layer}] ${a.reason}${ui.reset}\n`);
-                }
-                else if (a.kind === 'compact_history') {
-                    // 路由到 maybeCompact；把结构重建信号传回 core，使 lifecycle 按新 index 恢复。
-                    const result = await maybeCompact(history, report, undefined, state, activeTools);
-                    compactHistoryCalled = true;
-                    historyRebuilt ||= result?.historyRebuilt === true;
+            for (const action of actions) {
+                if (action.kind === 'warn') {
+                    layout.contentWrite(`  ${ui.accent}●${ui.reset} ${ui.yellow}调度器警告 [${action.layer}] ${action.reason}${ui.reset}\n`);
+                    continue;
                 }
+                const result = await maybeCompact(history, report, undefined, state, activeTools);
+                compactHistoryCalled = true;
+                historyRebuilt ||= result?.historyRebuilt === true;
             }
             const log = {
                 step,
                 report,
                 actions,
+                pressure,
                 compactHistoryCalled,
                 ts: Date.now(),
             };
-            obs.lastRunLog = log;
-            // 暴露给 /context 共享读(repl / context 命令)
+            scheduler.lastRunLog = log;
             state.schedulerLog = log;
             return historyRebuilt;
         },
     };
-    return obs;
+    return scheduler;
 }
-/** 便捷:agent/core.ts 不需要每次 createBudgetScheduler,直接 runScheduler(history, step)。 */
 export async function runScheduler(history, step, state = contextState, activeTools = chatTools) {
-    const s = createBudgetScheduler(state);
-    return s.runStep(history, step, activeTools);
+    return createBudgetScheduler(state).runStep(history, step, activeTools);
 }
-/** 手动 /compact 入口(repl):与自动路径共享预算评估和可执行 action，但强制执行 history 摘要。
- *  即便 layers.history.overBudget=false 或 totalOver=false，manual 仍追加 compact_history，
- * 并把 focus 透传给 LLM 摘要 prompt。
- *
- *  关系:runScheduler 是「自动触发」,manualCompact 是「用户显式触发」,二者共享 scheduleActions。
- *
- *  force=true:即便 oldGroups 空(history 全在保护区)也强行把早期消息降级压一次。
- *  适合"history 太长,自动阈值从未触发,但用户想强制压"的场景。
- *
- *  退化:config.contextBudget === false 时直接调 compactHistory(history, { focus }),与改造前等价。
- *  返回 SchedulerRunLog + compactDetail 字段,供 repl 文案展示"为什么没压"。 */
+/** User-requested compaction bypasses automatic pressure gating. */
 export async function manualCompact(history, focus, opts) {
     if (config.contextBudget === false) {
-        // 退化:不经调度器,直压 history(等价于改造前,但带 manual/force 透传)
-        const r = await import('./compact.js').then(({ compactHistory }) => compactHistory(history, {
+        const result = await import('./compact.js').then(({ compactHistory }) => compactHistory(history, {
             window: config.contextWindowTokens,
-            threshold: config.compactThreshold,
+            threshold: DEFAULT_BUDGET_POLICY.pressureTriggerRatio,
             focus,
             manual: true,
             force: opts?.force,
         }));
-        return {
+        const report = evaluateBudget(history, config.contextWindowTokens, -1, contextState.correction);
+        const log = {
             step: -1,
-            report: evaluateBudget(history, config.contextWindowTokens, -1, contextState.correction),
+            report,
             actions: [{ kind: 'compact_history', focus }],
+            pressure: emptyPressure(report),
             compactHistoryCalled: true,
             ts: Date.now(),
             compactDetail: {
-                reason: r.reason,
-                estimateBefore: r.estimateBefore,
-                estimateAfter: r.estimateAfter,
-                protectedRatio: r.protectedRatio,
-                oldGroupCount: r.oldGroupCount,
+                reason: result.reason,
+                estimateBefore: result.estimateBefore,
+                estimateAfter: result.estimateAfter,
+                protectedRatio: result.protectedRatio,
+                oldGroupCount: result.oldGroupCount,
                 focus,
             },
         };
+        contextState.schedulerLog = log;
+        return log;
     }
     const report = evaluateBudget(history, config.contextWindowTokens, -1, contextState.correction);
     let actions = scheduleActions(report);
-    // 用户显式说「要压」:即使 report 不含 history 触发,仍追加 compact_history
-    const hasCompact = actions.some(a => a.kind === 'compact_history');
-    if (!hasCompact) {
+    if (!actions.some((action) => action.kind === 'compact_history')) {
         actions = [...actions, { kind: 'compact_history', focus }];
     }
     else if (focus) {
-        // 已有 compact_history(action 由 scheduleActions 产,无 focus)— 注入 focus
-        actions = actions.map(a => a.kind === 'compact_history' ? { ...a, focus } : a);
+        actions = actions.map((action) => action.kind === 'compact_history' ? { ...action, focus } : action);
     }
-    const s = createBudgetScheduler();
-    // 直接驱动 runStep 的执行逻辑(action list 我们已自己生成)
     let compactHistoryCalled = false;
     let compactDetail;
-    for (const a of actions) {
-        if (a.kind === 'warn') {
-            // 复用现有 contentWrite 路径(经 scheduler.runStep);这里略,只记录到 log
-        }
-        else if (a.kind === 'compact_history') {
-            const r = await maybeCompact(history, report, {
-                manual: true,
-                force: opts?.force,
-                focus: a.focus,
-            });
-            compactHistoryCalled = true;
-            if (r) {
-                compactDetail = {
-                    reason: r.reason,
-                    estimateBefore: r.estimateBefore,
-                    estimateAfter: r.estimateAfter,
-                    protectedRatio: r.protectedRatio,
-                    oldGroupCount: r.oldGroupCount,
-                    focus: a.focus,
-                };
-            }
+    for (const action of actions) {
+        if (action.kind !== 'compact_history')
+            continue;
+        const result = await maybeCompact(history, report, {
+            manual: true,
+            force: opts?.force,
+            focus: action.focus,
+        });
+        compactHistoryCalled = true;
+        if (result) {
+            compactDetail = {
+                reason: result.reason,
+                estimateBefore: result.estimateBefore,
+                estimateAfter: result.estimateAfter,
+                protectedRatio: result.protectedRatio,
+                oldGroupCount: result.oldGroupCount,
+                focus: action.focus,
+            };
         }
     }
     const log = {
         step: -1,
         report,
         actions,
+        pressure: emptyPressure(report),
         compactHistoryCalled,
         ts: Date.now(),
         compactDetail,
     };
-    s.lastRunLog = log;
     contextState.schedulerLog = log;
     return log;
 }
-// ── 调试导出:用于 scripts/check-budget.ts 把报告打到 stdout 调试 ────────
 export { evaluateBudget, scheduleActions, formatReport };

package/dist/session/trace-metrics.js

@@ -1,99 +1,16 @@
 import { readFileSync } from 'node:fs';
-/** 把 history 中所有 user/tool/assistant 消息的文本/参数汇总到一个大字符串,便于
- * 用单一正则扫描多个 marker。比逐条消息 join 更稳:LLM 流式 contentPart 数组
- * 与多模态混合时,joinStringField 内部把每个 part 展平成字符串。 */
-function joinHistoryForMarkerScan(history) {
-    if (!history)
-        return '';
-    const parts = [];
-    for (const message of history) {
-        if (typeof message.content === 'string') {
-            parts.push(message.content);
-        }
-        else if (Array.isArray(message.content)) {
-            for (const part of message.content) {
-                if (typeof part === 'string')
-                    parts.push(part);
-                else if (part && typeof part === 'object' && 'text' in part && typeof part.text === 'string') {
-                    parts.push(part.text);
-                }
-            }
-        }
-        if ('tool_call_id' in message && typeof message.tool_call_id === 'string') {
-            parts.push(message.tool_call_id);
-        }
-    }
-    return parts.join('\n');
-}
-/** QUAL-01 质量维度。
- *
- * 优先用硬事件(retry_reflection / checklist_triggered / ask_human_call),
- * 这是 core.ts 在接缝处显式 emit 的,可信度高。
- * 如果 events 里没有对应硬事件(老 trace / 旧 fixture 走 history 文本回放),fallback
- * 到 history 文本扫描 + tool_call_end(name === 'ask_human', status === 'success')。
- *
- * - `retry_reflection` → 反思重试注入次数(RETRY-01)
- * - `checklist_triggered` → PROMPT-02 触发次数
- * - `ask_human_call` (status === 'success') → ASK-01 触发次数
- *
- * 三类都给出 fallback 0,确保报告维度永远是 number,不会被 NaN 污染。 */
-function reduceQualityDimensions(events, history) {
-    // 1. 硬事件计数(优先级最高)。
-    let reflectionRounds = 0;
-    let checklistTriggered = 0;
-    let askHumanCount = 0;
-    let hasHardReflection = false;
-    let hasHardChecklist = false;
-    let hasHardAskHuman = false;
-    for (const event of events) {
-        if (event.type === 'retry_reflection') {
-            reflectionRounds += 1;
-            hasHardReflection = true;
-        }
-        else if (event.type === 'checklist_triggered') {
-            checklistTriggered += 1;
-            hasHardChecklist = true;
-        }
-        else if (event.type === 'ask_human_call' && event.data.status === 'success') {
-            askHumanCount += 1;
-            hasHardAskHuman = true;
-        }
-    }
-    // 2. fallback: 历史事件没有硬信号时,从 history 文本 + tool_call_end 推断。
-    //    保留旧 fixture / 回放 JSONL 的兼容能力。
-    if (!hasHardReflection || !hasHardChecklist || !hasHardAskHuman) {
-        if (!hasHardReflection) {
-            const historyText = joinHistoryForMarkerScan(history);
-            reflectionRounds += (historyText.match(/\[retry reflection:/g) ?? []).length;
-        }
-        if (!hasHardChecklist) {
-            const historyText = joinHistoryForMarkerScan(history);
-            checklistTriggered += (historyText.match(/\[checklist\]/g) ?? []).length;
-        }
-        if (!hasHardAskHuman) {
-            for (const event of events) {
-                if (event.type !== 'tool_call_end')
-                    continue;
-                if (event.data.name === 'ask_human' && event.data.status === 'success') {
-                    askHumanCount += 1;
-                }
-            }
-        }
-    }
-    return { reflectionRounds, askHumanCount, checklistTriggered };
+function countAskHumanCalls(events) {
+    return events.filter((event) => event.type === 'ask_human_call' && event.data.status === 'success').length;
 }
-export function reduceTraceMetrics(events, history) {
+export function reduceTraceMetrics(events) {
     const ends = events.filter((event) => event.type === 'tool_call_end');
     let recovered = false;
     let hadFailure = false;
     let successes = 0;
-    let toolRetries = 0;
     let tokens = 0;
     let hasTokens = false;
     for (const event of ends) {
         const status = String(event.data.status ?? 'error');
-        const retry = Number(event.data.retry ?? 0);
-        toolRetries += Number.isFinite(retry) ? retry : 0;
         if (status === 'success') {
             successes++;
             if (hadFailure)
@@ -113,21 +30,17 @@ export function reduceTraceMetrics(events, history) {
         }
     }
     const modelRetries = events.filter((event) => event.type === 'model_retry').length;
-    const firstValidation = events.find((event) => event.type === 'validation_end');
     const turnEnd = [...events].reverse().find((event) => event.type === 'turn_end');
-    const quality = reduceQualityDimensions(events, history);
     return {
         toolCalls: events.filter((event) => event.type === 'tool_call_start').length,
         toolFailures: ends.length - successes,
         toolRecovery: recovered,
         firstSuccessRate: ends.length ? successes / ends.length : 1,
         modelRetries,
-        toolRetries,
-        retries: modelRetries + toolRetries,
+        retries: modelRetries,
         tokens: hasTokens ? tokens : null,
         durationMs: Number(turnEnd?.data.durationMs ?? 0),
-        firstValidationPassed: firstValidation?.data.status === 'passed',
-        ...quality,
+        askHumanCount: countAskHumanCalls(events),
     };
 }
 /** Reads event JSONL; malformed/legacy summary lines are ignored. */

package/dist/session/trace.js

@@ -41,14 +41,5 @@ export function appendCurrentSessionTrace(trace) {
     const sessionId = getCurrentSessionId();
     if (!sessionId)
         return;
-    const validation = trace.validation
-        ? {
-            status: trace.validation.status,
-            level: trace.validation.level,
-            durationMs: trace.validation.durationMs,
-            verificationComplete: trace.validation.verificationComplete,
-            fingerprint: trace.validation.fingerprint,
-        }
-        : undefined;
-    appendTraceLine(sessionId, { ...trace, sessionId, validation });
+    appendTraceLine(sessionId, { ...trace, sessionId });
 }