agent-recall-mcp 3.3.6 → 3.3.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/README.md +135 -0
  2. package/package.json +2 -2
package/README.md CHANGED
@@ -58,6 +58,61 @@ curl -o ~/.claude/commands/arsave.md https://raw.githubusercontent.com/Goldentri
58
58
  curl -o ~/.claude/commands/arstart.md https://raw.githubusercontent.com/Goldentrii/AgentRecall/main/commands/arstart.md
59
59
  ```
60
60
 
61
+ ### The Difference
62
+
63
+ ```
64
+ WITHOUT AgentRecall WITH AgentRecall
65
+ ────────────────── ────────────────
66
+
67
+ Day 1: Build monorepo Day 1: /arstart → /arsave
68
+ Day 2: "What monorepo?" Day 2: /arstart
69
+ → 20 min re-explaining → 2 sec: loads all decisions
70
+ → Agent repeats same mistakes → Knows "no version inflation"
71
+ → Forgets your priorities → Knows "arsave = hero section"
72
+ → Misses half the tasks → Pushes to both repos
73
+ ```
74
+
75
+ ### Real Session Flow
76
+
77
+ This is from an actual multi-day project where a human gave scattered, non-linear instructions. The agent used AgentRecall throughout:
78
+
79
+ ```
80
+ Session 1 (Tuesday) Session 2 (Wednesday, different agent)
81
+ ───────────────── ─────────────────────────────────────
82
+
83
+ /arstart /arstart
84
+ │ │
85
+ ├─ palace_walk ──→ "monorepo project, ├─ palace_walk ──→ loads Tuesday's
86
+ │ SDK + CLI planned" │ architecture decisions
87
+ │ │ in 200 tokens
88
+ ├─ recall_insight ──→ 3 prior lessons ├─ recall_insight
89
+ │ • "structurize scattered input" │ • "no version inflation"
90
+ │ • "search before building" │ • "arsave/arstart = hero section"
91
+ │ • "ask when ambiguous" │
92
+ │ ├─ Continues exactly where
93
+ ▼ │ Session 1 left off
94
+ Human: "we need SDK, CLI, │
95
+ update README, fix versions" ▼
96
+ │ Human: "publish to npm,
97
+ ├─ alignment_check update both GitHub repos"
98
+ │ confidence: medium │
99
+ │ 4 tasks detected ├─ No re-explanation needed.
100
+ │ → present plan → human confirms │ Agent already knows the
101
+ │ │ monorepo structure, package
102
+ ├─ Execute in order: │ names, and version policy.
103
+ │ 1. Core extraction ✓ │
104
+ │ 2. Tool logic split ✓ └─ Done in 2 minutes
105
+ │ 3. MCP wrappers ✓ (vs 20 min cold start
106
+ │ 4. SDK + CLI ✓ without AgentRecall)
107
+ │
108
+ /arsave
109
+ │
110
+ ├─ journal_write ──→ decisions + tasks saved
111
+ ├─ awareness_update ──→ "scattered input →
112
+ │ structurize → confirm"
113
+ └─ palace_write ──→ architecture room updated
114
+ ```
115
+
61
116
  ---
62
117
 
63
118
  ## Why Choose AgentRecall
@@ -511,6 +566,86 @@ MIT License.
511
566
 
512
567
  ---
513
568
 
569
+ <p align="center">
570
+ <a href="#arsave-and-arstart"><img src="https://img.shields.io/badge/%2Farsave-保存会话-FF6B6B?style=for-the-badge" alt="/arsave"></a>
571
+ <a href="#arsave-and-arstart"><img src="https://img.shields.io/badge/%2Farstart-加载上下文-4ECDC4?style=for-the-badge" alt="/arstart"></a>
572
+ </p>
573
+
574
+ ## `/arsave` 和 `/arstart`
575
+
576
+ > **两个命令，就这么简单。**
577
+
578
+ | 命令 | 时机 | 功能 |
579
+ |------|------|------|
580
+ | **`/arsave`** | 会话结束时 | 写入日志 + 整合到宫殿 + 更新感知 + 可选 git 推送 |
581
+ | **`/arstart`** | 会话开始时 | 召回跨项目洞察 + 遍历宫殿 + 加载上下文 |
582
+
583
+ 会话结束时输入 `/arsave`，所有内容自动保存。下次开始时输入 `/arstart`，所有上下文自动恢复。
584
+
585
+ ```bash
586
+ # 安装命令（一次性，仅 Claude Code）
587
+ mkdir -p ~/.claude/commands
588
+ curl -o ~/.claude/commands/arsave.md https://raw.githubusercontent.com/Goldentrii/AgentRecall/main/commands/arsave.md
589
+ curl -o ~/.claude/commands/arstart.md https://raw.githubusercontent.com/Goldentrii/AgentRecall/main/commands/arstart.md
590
+ ```
591
+
592
+ ### 效果对比
593
+
594
+ ```
595
+ 没有 AgentRecall 有 AgentRecall
596
+ ────────────── ────────────
597
+
598
+ 第 1 天：构建单仓 第 1 天：/arstart → /arsave
599
+ 第 2 天："什么单仓？" 第 2 天：/arstart
600
+ → 20 分钟重新解释 → 2 秒：加载所有决策
601
+ → 智能体重复同样的错误 → 知道"不要版本膨胀"
602
+ → 忘记你的优先级 → 知道"arsave 要放首位"
603
+ → 遗漏一半的任务 → 自动推送两个仓库
604
+ ```
605
+
606
+ ### 真实会话流程
607
+
608
+ 以下来自一个真实的多日项目，人类给出了分散、非线性的指令。智能体全程使用 AgentRecall：
609
+
610
+ ```
611
+ 会话 1（周二） 会话 2（周三，不同的智能体）
612
+ ────────── ────────────────────────
613
+
614
+ /arstart /arstart
615
+ │ │
616
+ ├─ palace_walk ──→ "单仓项目， ├─ palace_walk ──→ 200 token
617
+ │ 计划构建 SDK + CLI" │ 加载周二的架构决策
618
+ │ │
619
+ ├─ recall_insight ──→ 3 条历史教训 ├─ recall_insight
620
+ │ • "结构化分散的输入" │ • "不要版本膨胀"
621
+ │ • "先搜索再构建" │ • "arsave 放在最显眼的位置"
622
+ │ • "模糊时询问，明确时执行" │
623
+ │ ├─ 从会话 1 离开的地方
624
+ ▼ │ 无缝继续
625
+ 人类："我们需要 SDK、CLI， │
626
+ 更新 README，修复版本号" ▼
627
+ │ 人类："发布到 npm，
628
+ ├─ alignment_check 更新两个 GitHub 仓库"
629
+ │ confidence: medium │
630
+ │ 检测到 4 个任务项 ├─ 无需重新解释。
631
+ │ → 呈现方案 → 人类确认 │ 智能体已经知道单仓结构、
632
+ │ │ 包名和版本策略。
633
+ ├─ 按顺序执行： │
634
+ │ 1. 核心提取 ✓ └─ 2 分钟完成
635
+ │ 2. 工具逻辑拆分 ✓ （没有 AgentRecall
636
+ │ 3. MCP 封装 ✓ 需要 20 分钟冷启动）
637
+ │ 4. SDK + CLI ✓
638
+ │
639
+ /arsave
640
+ │
641
+ ├─ journal_write ──→ 决策 + 任务已保存
642
+ ├─ awareness_update ──→ "分散输入 →
643
+ │ 结构化 → 确认 → 执行"
644
+ └─ palace_write ──→ architecture 房间已更新
645
+ ```
646
+
647
+ ---
648
+
514
649
  ## 为什么选择 AgentRecall
515
650
 
516
651
  **你的智能体在会话之间遗忘一切。** 决策蒸发，错误重复，上下文每次从零构建。AgentRecall 用持久记忆修复这个问题 — 记忆会复合增长，而不是无限膨胀。
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "agent-recall-mcp",
3
- "version": "3.3.6",
3
+ "version": "3.3.7",
4
4
  "description": "Memory Palace MCP server for AI agents",
5
5
  "type": "module",
6
6
  "main": "dist/index.js",
@@ -11,7 +11,7 @@
11
11
  "test": "node --test test/*.test.mjs"
12
12
  },
13
13
  "dependencies": {
14
- "agent-recall-core": "^3.3.6",
14
+ "agent-recall-core": "^3.3.7",
15
15
  "@modelcontextprotocol/sdk": "^1.28.0",
16
16
  "zod": "^3.24.0"
17
17
  },