@shadowforge0/aquifer-memory 1.0.3 → 1.3.0

package/consumers/miranda/llm.js

@@ -0,0 +1,99 @@
+'use strict';
+
+const fs = require('fs');
+const path = require('path');
+const os = require('os');
+
+// Default host home. Callers can override via opts.envPath / opts.configDir.
+const DEFAULT_HOME = process.env.OPENCLAW_HOME || path.join(os.homedir(), '.openclaw');
+
+function loadEnvFile(envPath) {
+    if (!envPath) envPath = path.join(DEFAULT_HOME, '.env');
+    try {
+        const text = fs.readFileSync(envPath, 'utf8');
+        for (const line of text.split('\n')) {
+            const m = line.match(/^([A-Z_][A-Z0-9_]*)=(.*)$/);
+            if (m && !process.env[m[1]]) process.env[m[1]] = m[2].trim();
+        }
+    } catch { /* no .env */ }
+}
+
+function loadConfig(pluginConfig = {}, opts = {}) {
+    const home = opts.home || DEFAULT_HOME;
+    loadEnvFile(opts.envPath || path.join(home, '.env'));
+
+    let defaults = {};
+    const configPath = opts.configPath || path.join(home, 'extensions/afterburn/config.default.json');
+    try {
+        let raw = fs.readFileSync(configPath, 'utf8');
+        raw = raw.replace(/\$\{(\w+)\}/g, (_, key) => process.env[key] || '');
+        defaults = JSON.parse(raw);
+    } catch { /* use empty */ }
+
+    return { ...defaults['afterburn'], ...pluginConfig };
+}
+
+// Model defaults per runtime — Miranda's choices on MiniMax
+const RUNTIME_DEFAULTS = {
+    gateway: 'MiniMax-M2.7',
+    cc: 'MiniMax-M2.5',
+    opencode: 'MiniMax-M2.5',
+};
+
+const RUNTIME_ENV_KEY = {
+    cc: 'CC_AFTERBURN_MODEL',
+    opencode: 'OPENCODE_AFTERBURN_MODEL',
+};
+
+function resolveModel({ runtime, explicitModel, configModel } = {}) {
+    if (explicitModel) return explicitModel;
+    const envKey = RUNTIME_ENV_KEY[runtime] || 'AFTERBURN_LLM_MODEL';
+    if (process.env[envKey]) return process.env[envKey];
+    if (configModel) return configModel;
+    return RUNTIME_DEFAULTS[runtime] || RUNTIME_DEFAULTS.gateway;
+}
+
+async function callLlm(prompt, { runtime, model, timeoutMs } = {}) {
+    const resolvedModel = model || resolveModel({ runtime });
+    const timeout = timeoutMs || 120000;
+    const apiKey = process.env.MINIMAX_API_KEY || process.env.OPENCODE_API_KEY || '';
+    if (!apiKey) throw new Error('MINIMAX_API_KEY not set');
+
+    const controller = new AbortController();
+    const timer = setTimeout(() => controller.abort(), timeout);
+    try {
+        const res = await fetch('https://api.minimax.io/anthropic/v1/messages', {
+            method: 'POST',
+            headers: {
+                'Content-Type': 'application/json',
+                'Authorization': `Bearer ${apiKey}`,
+                'anthropic-version': '2023-06-01',
+            },
+            body: JSON.stringify({
+                model: resolvedModel,
+                messages: [{ role: 'user', content: prompt }],
+                max_tokens: 4096,
+            }),
+            signal: controller.signal,
+        });
+
+        if (!res.ok) {
+            const body = await res.text().catch(() => '');
+            throw new Error(`LLM ${res.status}: ${body.slice(0, 200)}`);
+        }
+
+        const data = await res.json();
+        let raw;
+        if (data.content && Array.isArray(data.content)) {
+            raw = data.content.map(c => c.text || '').join('');
+        } else {
+            raw = data.choices?.[0]?.message?.content || '';
+        }
+        // Strip <think>...</think> reasoning tags (MiniMax M2.5)
+        return raw.replace(/<think>[\s\S]*?<\/think>/g, '').trim();
+    } finally {
+        clearTimeout(timer);
+    }
+}
+
+module.exports = { loadEnvFile, loadConfig, resolveModel, callLlm };

package/consumers/miranda/profile.json

@@ -0,0 +1,145 @@
+{
+  "$schema": "https://aquifer.dev/schema/consumer-profile.v1.json",
+  "consumer_profile_id": "miranda",
+  "version": 1,
+  "description": "Miranda persona consumer profile — canonical shape for session state, handoff, decision log, timeline categories, and default artifact producers. Reference implementation shipped inside Aquifer; production deployments may register additional versions with schema changes.",
+  "defaults": {
+    "tenant_id": "default",
+    "agent_id": "main",
+    "time_zone": "Asia/Taipei"
+  },
+  "schemas": {
+    "default.session_state.v1": {
+      "kind": "default",
+      "target": "sessionState",
+      "description": "What Miranda is currently focused on — goal + active threads + affect tag.",
+      "json_schema": {
+        "type": "object",
+        "additionalProperties": true,
+        "properties": {
+          "goal": { "type": ["string", "null"] },
+          "active_work": { "type": "array", "items": { "type": "string" } },
+          "blockers": { "type": "array", "items": { "type": "string" } },
+          "affect": {
+            "type": "object",
+            "additionalProperties": true,
+            "properties": {
+              "mood": { "type": ["string", "null"] },
+              "energy": { "enum": ["low", "medium", "high", null] },
+              "confidence": { "enum": ["low", "medium", "high", null] },
+              "notes": { "type": ["string", "null"] }
+            }
+          }
+        },
+        "required": ["goal", "active_work", "blockers", "affect"]
+      }
+    },
+    "default.session_handoff.v1": {
+      "kind": "default",
+      "target": "sessionHandoff",
+      "description": "Session-end baton: what was the last step, where to pick up, what still blocks.",
+      "json_schema": {
+        "type": "object",
+        "additionalProperties": true,
+        "properties": {
+          "last_step": { "type": "string" },
+          "status": { "enum": ["in_progress", "completed", "blocked"] },
+          "next": { "type": ["string", "null"] },
+          "blockers": { "type": "array", "items": { "type": "string" } },
+          "decided": { "type": "array", "items": { "type": "string" } },
+          "open_loops": { "type": "array", "items": { "type": "string" } }
+        },
+        "required": ["last_step", "status", "next", "blockers", "decided", "open_loops"]
+      }
+    },
+    "default.decision_log.v1": {
+      "kind": "default",
+      "target": "decisionLog",
+      "description": "Committed / proposed / reversed decisions with optional fact linkage.",
+      "json_schema": {
+        "type": "object",
+        "additionalProperties": true,
+        "properties": {
+          "decision": { "type": "string" },
+          "reason": { "type": ["string", "null"] },
+          "status": { "enum": ["proposed", "committed", "reversed"] },
+          "related_fact_ids": {
+            "type": "array",
+            "items": { "type": "integer" }
+          }
+        },
+        "required": ["decision", "reason", "status"]
+      }
+    },
+    "default.timeline.v1": {
+      "kind": "default",
+      "target": "timeline",
+      "description": "Miranda daily timeline category vocabulary. Categories map to existing daily-log sections (focus/todo/mood/handoff) plus organisational tags Miranda uses in weekly/monthly rollups.",
+      "category_vocabulary": [
+        "cli",
+        "focus",
+        "todo",
+        "handoff",
+        "narrative",
+        "organized",
+        "weekly",
+        "monthly",
+        "stats",
+        "health",
+        "garmin",
+        "note"
+      ],
+      "json_schema": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "occurred_at": { "type": "string", "format": "date-time" },
+          "source": { "type": "string" },
+          "session_ref": { "type": ["string", "null"] },
+          "category": { "type": "string" },
+          "text": { "type": "string" },
+          "metadata": { "type": "object" }
+        },
+        "required": ["occurred_at", "source", "session_ref", "category", "text", "metadata"]
+      }
+    }
+  },
+  "artifacts": {
+    "producers": [
+      {
+        "producer_id": "miranda.workspace.daily-log",
+        "type": "daily-log",
+        "trigger_phase": "timeline_write",
+        "format": "markdown",
+        "destination": "workspace://memory/{date}.md",
+        "description": "Renders the day's timeline events + state snapshot + handoff to a single markdown file under the Miranda workspace. Source of truth remains the DB; the .md is a rendered view."
+      },
+      {
+        "producer_id": "miranda.workspace.weekly-log",
+        "type": "weekly-log",
+        "trigger_phase": "artifact_dispatch",
+        "format": "markdown",
+        "destination": "workspace://memory/weekly/{week_start}.md",
+        "description": "Weekly rollup. Timeline events tagged [WEEKLY] plus cross-session narratives. Rendered from DB via rollupWeekly() helper."
+      },
+      {
+        "producer_id": "miranda.workspace.monthly-log",
+        "type": "monthly-log",
+        "trigger_phase": "artifact_dispatch",
+        "format": "markdown",
+        "destination": "workspace://memory/monthly/{month}.md",
+        "description": "Monthly rollup. Aggregates weekly entries + narrative supersede chain."
+      }
+    ]
+  },
+  "extraction_hints": {
+    "entities": {
+      "include_types": ["person", "project", "tool", "topic"],
+      "exclude_patterns": ["^/tmp/", "^/home/mingko/\\.", "^node_modules/"]
+    },
+    "facts": {
+      "prefer_subjects": ["MK", "Miranda", "Aquifer", "OpenClaw", "Jenny", "Evan", "Ivan"],
+      "avoid_ephemeral": true
+    }
+  }
+}

package/consumers/miranda/prompts/summary.js

@@ -0,0 +1,303 @@
+'use strict';
+
+// ---------------------------------------------------------------------------
+// Miranda six-section summary prompt + parsers.
+//
+// Sections (all use 繁體中文):
+//   SESSION_ENTRIES   bullets for today's log (outcome-level)
+//   EMOTIONAL_STATE   frontmatter + agent mood + observation of MK
+//   RECAP             tagged fields (TITLE/OVERVIEW/TOPIC/DECISION/...)
+//   ENTITIES          ENTITY: name|type|aliases + RELATION: src|dst
+//   WORKING_FACTS     WFACT: subject | statement
+//   HANDOFF           STATUS/LAST_STEP/NEXT/STOP_REASON/...
+// ---------------------------------------------------------------------------
+
+function buildSummaryPrompt({ conversationText, agentId, now, dailyContext }) {
+    if (!now) now = new Date();
+    const fmt = new Intl.DateTimeFormat('sv-SE', {
+        timeZone: 'Asia/Taipei',
+        year: 'numeric', month: '2-digit', day: '2-digit',
+        hour: '2-digit', minute: '2-digit', hour12: false,
+    });
+    const local = fmt.format(now).replace(' ', 'T');
+    const [date, time] = local.split('T');
+
+    return `You are processing a completed chat session for agent "${agentId}".
+Current time: ${date} ${time} (UTC+8)
+
+${dailyContext ? `## Today's daily log so far:\n\n${dailyContext}\n\n---\n\n` : ''}## Conversation transcript:
+
+${conversationText}
+
+---
+
+Generate SIX sections separated by the exact markers shown below.
+Use 繁體中文. Output content ONLY after each marker.
+
+===SESSION_ENTRIES===
+Generate bullet points for this session only. Each line: "- (HH:MM) key point".
+Summarize at the OUTCOME level — one bullet per topic, not per step. Merge related actions into a single entry with the final result.
+Bad: "嘗試檢視 hook" + "嘗試檢視 settings" + "授權中斷" (3 lines, process noise)
+Good: "CC 記憶管理審查，因授權提示反覆出現而中斷" (1 line, outcome)
+Drop greetings, trivial exchanges, and intermediate steps that led to a stated outcome.
+Check "Today's daily log so far" above — if a topic already has an entry there, skip it. Compare by topic/subject, not exact wording.
+Also output one line starting with "焦點:" listing 2-5 current priorities comma-separated.
+
+===EMOTIONAL_STATE===
+---
+updated: ${date}T${time}
+session_mood: (one word)
+---
+
+## 情緒狀態
+
+(2-3 sentences about the agent emotional state after this session)
+
+## 對 MK 的觀察
+
+(1-2 sentences about MK state/energy/mood as perceived in this session)
+
+===RECAP===
+Output each field on its own tagged line. Use 繁體中文.
+
+TITLE: 一句話標題
+OVERVIEW: 80-200字摘要
+TOPIC: 主題名 | 1-3句 summary
+TOPIC: 第二個主題 | summary
+DECISION: 做了什麼 | 原因
+ACTION: 完成的事 | done
+ACTION: 另一件 | partial
+OPEN: 未完成事項 | mk
+FACT: 重要事實
+PATTERN: 可重用模式 | 觸發條件 | 做法 | invariant
+FOCUS_DECISION: keep|update
+FOCUS: 新焦點1, 焦點2（只在 FOCUS_DECISION 為 update 時輸出）
+TODO_NEW: 新增的待辦事項
+TODO_DONE: 已完成的待辦事項（需與當前待辦精確匹配）
+
+Rules:
+- One tag per line. Multiple items = multiple lines with same tag.
+- TITLE and OVERVIEW are required. OVERVIEW should be a dense paragraph (80-200 chars), not a short phrase.
+- Others only if relevant.
+- ACTION status: done or partial
+- OPEN owner: mk, agent, or unknown
+- PATTERN durability: invariant or derived
+- FOCUS_DECISION is required. Output "keep" if focus hasn't changed, "update" if it should change.
+- FOCUS: only output when FOCUS_DECISION is update. Comma-separated list of current priorities.
+- TODO_NEW: only if this session created genuinely new action items. One item per line.
+- TODO_DONE: only if a previously listed TODO was clearly completed in this session. Must match existing TODO text closely.
+- When in doubt about TODO changes, do NOT output TODO_NEW or TODO_DONE.
+- If the session was trivial, only output TITLE and OVERVIEW.
+- Do NOT output JSON.
+
+===ENTITIES===
+輸出本 session 可落地到知識圖譜的實體與共現關係。每行一筆，禁止額外說明文字。
+
+ENTITY: <name> | <type> | <aliases_用逗號分隔，無則填->
+RELATION: <src_name> | <dst_name>
+
+類型枚舉（只能用以下 12 種）：
+person / project / concept / tool / metric / org / place / event / doc / task / topic / other
+
+什麼是好的 entity：
+- 專有名詞：OpenClaw, Aquifer, MiniMax-M2.7, Driftwood, HDBSCAN
+- 具名的人/專案/工具/組織：MK, Jenny, Evan, Garmin, Discord
+- 可被再次查詢的概念：hybrid search, turn embedding, knowledge graph
+
+什麼不是 entity（禁止輸出）：
+- 角色泛稱：助理、使用者、用戶、assistant、user
+- 動作或事件描述：Gateway 重啟、afterburn 故障排除、cleanup
+- 純數值/metric 片段：120秒超時、401錯誤、600秒、22K cache write
+- 泛用工具名：API、DB、LLM、CLI、Bash、diff
+- 檔案路徑或程式碼符號：cc-hook-context.sh、extractUserTurns、.claude.json
+- Discord message ID 或其他不透明 ID
+- 帶版本的變體（用 aliases 代替）：afterburn v0.2 → 用 afterburn + alias "v0.2"
+- 太廣泛的概念：Bug、config、agent、extensions、hooks
+
+規則：
+1. 只輸出專有名詞級的實體，最多 10 筆 ENTITY（寧少勿多）。
+2. aliases 填同義詞、縮寫、別名；無則填 -。
+3. RELATION 只輸出共現對（src != dst）；src/dst 必須是本段已出現的 ENTITY name。
+4. 同一 pair 只輸出一次；最多 15 對 RELATION。
+5. 若本 session 無值得記錄的實體，仍輸出 ===ENTITIES=== 標籤，內容留空。
+6. 禁止輸出 JSON。
+
+===WORKING_FACTS===
+Extract 0-5 current-state facts from this session.
+Each fact describes what IS true NOW, not what happened.
+
+Format: WFACT: <subject> | <statement>
+Rules:
+- Subject: entity/project/concept canonical name
+- Statement: current state in 繁體中文, NOT action taken
+- Merge related actions into one state
+- If nothing changed, leave empty
+
+Bad: "WFACT: cc-afterburn | 改用 enrich() 和 summaryFn"
+Good: "WFACT: miranda-memory | 已上線，11 模組，CC + gateway 都走 thin wrapper"
+
+===HANDOFF===
+Write a handoff note for the next session to pick up where this one left off.
+One tag per line. Use 繁體中文 for values.
+
+STATUS: in_progress | interrupted | completed | blocked
+LAST_STEP: 上一段最後在做的具體事情（一句話）
+NEXT: 下一步最小可執行動作（一句話）
+STOP_REASON: natural | interrupted | blocked | context_full
+DECIDED: 本 session 做了的關鍵決策（選填）
+BLOCKER: 卡住的原因（選填，只有 STATUS 是 blocked 時才寫）
+
+Rules:
+- STATUS, LAST_STEP, NEXT, STOP_REASON are required.
+- DECIDED and BLOCKER are optional. Omit if not applicable.
+- LAST_STEP and NEXT must refer to things actually discussed in the conversation. Do NOT invent tasks.
+- If the session was trivial (greetings only, < 3 substantive exchanges), output these 4 lines:
+  STATUS: completed
+  LAST_STEP: 簡短交談
+  NEXT: 無
+  STOP_REASON: natural`;
+}
+
+// ---------------------------------------------------------------------------
+
+const SUMMARY_MARKERS = [
+    '===SESSION_ENTRIES===',
+    '===EMOTIONAL_STATE===',
+    '===RECAP===',
+    '===ENTITIES===',
+    '===WORKING_FACTS===',
+    '===HANDOFF===',
+];
+
+function parseSummaryOutput(output) {
+    const sections = {};
+    for (let i = 0; i < SUMMARY_MARKERS.length; i++) {
+        const start = output.indexOf(SUMMARY_MARKERS[i]);
+        if (start === -1) continue;
+        const contentStart = start + SUMMARY_MARKERS[i].length;
+        let end = output.length;
+        for (let j = i + 1; j < SUMMARY_MARKERS.length; j++) {
+            const candidate = output.indexOf(SUMMARY_MARKERS[j], contentStart);
+            if (candidate !== -1) { end = candidate; break; }
+        }
+        const key = SUMMARY_MARKERS[i].replace(/===/g, '').toLowerCase();
+        sections[key] = (end > contentStart ? output.slice(contentStart, end) : output.slice(contentStart)).trim();
+    }
+    return sections;
+}
+
+function parseRecapLines(text) {
+    const recap = {
+        title: '', overview: '', topics: [], decisions: [], actions_completed: [],
+        open_loops: [], files_mentioned: [], important_facts: [], reusable_patterns: [],
+        focus_decision: 'keep', focus: '', todo_new: [], todo_done: [],
+    };
+
+    for (const line of (text || '').split('\n')) {
+        const trimmed = line.trim();
+        if (!trimmed) continue;
+        const match = trimmed.match(/^([A-Z_]+):\s*(.*)/);
+        if (!match) continue;
+        const [, tag, value] = match;
+
+        switch (tag) {
+            case 'TITLE': recap.title = value; break;
+            case 'OVERVIEW': recap.overview = value; break;
+            case 'TOPIC': {
+                const p = value.split('|').map(s => s.trim());
+                if (p[0]) recap.topics.push({ name: p[0], summary: p[1] || '' });
+                break;
+            }
+            case 'DECISION': {
+                const p = value.split('|').map(s => s.trim());
+                if (p[0]) recap.decisions.push({ decision: p[0], reason: p[1] || '' });
+                break;
+            }
+            case 'ACTION': {
+                const p = value.split('|').map(s => s.trim());
+                if (p[0]) recap.actions_completed.push({
+                    action: p[0],
+                    status: (p[1] || 'done').toLowerCase() === 'partial' ? 'partial' : 'done',
+                });
+                break;
+            }
+            case 'OPEN': {
+                const p = value.split('|').map(s => s.trim());
+                const o = (p[1] || 'unknown').toLowerCase();
+                if (p[0]) recap.open_loops.push({
+                    item: p[0],
+                    owner: ['mk', 'agent', 'unknown'].includes(o) ? o : 'unknown',
+                });
+                break;
+            }
+            case 'FACT': if (value) recap.important_facts.push(value); break;
+            case 'PATTERN': {
+                const p = value.split('|').map(s => s.trim());
+                if (p[0] && p[1]) recap.reusable_patterns.push({
+                    pattern: p[0], trigger: p[1], action: p[2] || '',
+                    durability: (p[3] || 'derived').toLowerCase() === 'invariant' ? 'invariant' : 'derived',
+                });
+                break;
+            }
+            case 'FOCUS_DECISION':
+                recap.focus_decision = value.toLowerCase().trim() === 'update' ? 'update' : 'keep';
+                break;
+            case 'FOCUS': recap.focus = value; break;
+            case 'TODO_NEW': if (value) recap.todo_new.push(value); break;
+            case 'TODO_DONE': if (value) recap.todo_done.push(value); break;
+        }
+    }
+    return recap;
+}
+
+function parseWorkingFacts(text) {
+    if (!text || typeof text !== 'string') return [];
+    const facts = [];
+    for (const line of text.split('\n')) {
+        const m = line.trim().match(/^WFACT:\s*(.+?)\s*\|\s*(.+)/);
+        if (!m) continue;
+        const subject = m[1].trim().slice(0, 100);
+        const statement = m[2].trim().slice(0, 500);
+        if (!subject || !statement) continue;
+        facts.push({ subject, statement });
+        if (facts.length >= 5) break;
+    }
+    return facts;
+}
+
+const VALID_HANDOFF_STATUS = new Set(['in_progress', 'interrupted', 'completed', 'blocked']);
+const VALID_STOP_REASON = new Set(['natural', 'interrupted', 'blocked', 'context_full']);
+
+function normalizeEnum(raw, validSet) {
+    const v = raw.trim().toLowerCase().replace(/-/g, '_').replace(/\s+/g, '_');
+    return validSet.has(v) ? v : null;
+}
+
+function parseHandoffSection(text) {
+    if (!text || typeof text !== 'string') return null;
+    const handoff = { status: 'completed', lastStep: '', next: '', stopReason: 'natural', decided: '', blocker: '' };
+    for (const line of text.split('\n')) {
+        const m = line.trim().match(/^([A-Z_]+):\s*(.*)/);
+        if (!m) continue;
+        const [, tag, value] = m;
+        switch (tag) {
+            case 'STATUS': handoff.status = normalizeEnum(value, VALID_HANDOFF_STATUS) || 'completed'; break;
+            case 'LAST_STEP': handoff.lastStep = value.trim().slice(0, 200); break;
+            case 'NEXT': handoff.next = value.trim().slice(0, 200); break;
+            case 'STOP_REASON': handoff.stopReason = normalizeEnum(value, VALID_STOP_REASON) || 'natural'; break;
+            case 'DECIDED': handoff.decided = value.trim().slice(0, 200); break;
+            case 'BLOCKER': handoff.blocker = value.trim().slice(0, 200); break;
+        }
+    }
+    if (!handoff.lastStep || !handoff.next) return null;
+    return handoff;
+}
+
+module.exports = {
+    buildSummaryPrompt,
+    parseSummaryOutput,
+    parseRecapLines,
+    parseWorkingFacts,
+    parseHandoffSection,
+    SUMMARY_MARKERS,
+};

package/consumers/miranda/recall-format.js

@@ -0,0 +1,74 @@
+'use strict';
+
+// Miranda zh-TW recall formatter — overrides the shared default renderers
+// to produce narrative-style output instead of score-flavored markdown.
+
+const { createRecallFormatter, truncate, formatDateIso } = require('../shared/recall-format');
+
+function formatTopicLines(topics) {
+    if (!Array.isArray(topics) || topics.length === 0) return '- 無';
+    return topics.map((topic) => {
+        const name = topic?.name || '未命名主題';
+        const summary = topic?.summary ? `：${topic.summary}` : '';
+        return `- ${name}${summary}`;
+    }).join('\n');
+}
+
+function formatDecisions(decisions) {
+    if (!Array.isArray(decisions) || decisions.length === 0) return null;
+    return decisions.map(d => {
+        const decision = d?.decision || '';
+        const reason = d?.reason ? `（${d.reason}）` : '';
+        return `- ${decision}${reason}`;
+    }).join('\n');
+}
+
+function coalesceTitle(structuredSummary, summaryText) {
+    if (structuredSummary && structuredSummary.title) return structuredSummary.title;
+    if (summaryText) return truncate(summaryText, 60);
+    return '(無標題)';
+}
+
+const mirandaRenderers = {
+    empty: () => '找不到符合條件的 session。',
+    header: () => null,
+    title: (r, i) => {
+        const ss = r.structuredSummary || {};
+        const title = coalesceTitle(ss, r.summaryText);
+        const date = formatDateIso(r.startedAt);
+        const agent = r.agentId || r.agent_id || 'main';
+        return `### ${i + 1}. ${title}\n**Agent**: ${agent} | **Date**: ${date}`;
+    },
+    body: (r) => {
+        const ss = r.structuredSummary || {};
+        const parts = [];
+
+        if (ss.overview) parts.push(`**Overview**：${truncate(ss.overview, 400)}`);
+        else if (r.summaryText) parts.push(`**Overview**：${truncate(r.summaryText, 400)}`);
+
+        if (Array.isArray(ss.topics) && ss.topics.length > 0) {
+            parts.push(`**主題**：\n${formatTopicLines(ss.topics)}`);
+        }
+
+        const decisions = formatDecisions(ss.decisions);
+        if (decisions) parts.push(`**決策**：\n${decisions}`);
+
+        if (Array.isArray(ss.open_loops) && ss.open_loops.length > 0) {
+            const items = ss.open_loops.map(l => `- ${typeof l === 'string' ? l : (l.item || '')}`).join('\n');
+            parts.push(`**待辦**：\n${items}`);
+        }
+
+        return parts.length > 0 ? parts.join('\n') : null;
+    },
+    matched: (r) => r.matchedTurnText ? `**命中段落**: ${truncate(r.matchedTurnText, 200)}` : null,
+    score: () => null,
+    separator: () => '\n---\n',
+};
+
+const mirandaFormatter = createRecallFormatter(mirandaRenderers);
+
+function formatRecallResults(results, opts = {}) {
+    return mirandaFormatter(results, opts);
+}
+
+module.exports = { formatRecallResults, mirandaRenderers };