Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add rules/zql0805/memforge/memforge-auto-recallgit clone --depth 1 https://github.com/zql0805/memforgeWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.07463 | $0.07463 |
| Opus 5 | $0.03732 | $0.03732 |
| Sonnet 5 | $0.01493 | $0.01493 |
| Haiku 4.5 | $0.00746 | $0.00746 |
Grade A, and why
memforge-auto-recall scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 567 lines — stays where its author put it; the contents beside it link to each section on GitHub.
════════════════════════════════════════════════════
GATE 1 — recall_memory 强制入口（与 GATE 0 同级最高优先级）
读到此处即锁定：本会话首次回复必须先调用 recall_memory
════════════════════════════════════════════════════
铁律：会话首次回复的第一个工具调用必须是 recall_memory
无论用户问什么、任务是否紧急、对话摘要是否已提供上下文，每个会话的第一轮回复中，第一个 MCP/工具调用必须是 recall_memory。
执行要求
- 从用户消息中提取核心关键词作为 query
- 必须传
product_line（从工作区路径推断） - 对话摘要不能替代
recall_memory— 摘要只包含上次对话快照，记忆库包含跨会话的经验、规范和关联知识 - 检索完成后，结合记忆结果和对话摘要决定下一步行动
- 跨会话恢复未完成工作时，遵循「跨会话恢复安全协议」（见下方独立章节）
后续交互的检索规则
- 同一会话内，当任务/话题发生变化时必须再次 recall_memory
- 同一任务的连续交互无需重复检索
- 涉及代码修改时，额外检索相关文件路径和业务域
违规识别
- 会话首轮回复中未调用 recall_memory 就执行其他工具（Read/Shell/Grep 等）→ 违规
- 以「对话摘要已提供上下文」为由跳过 recall_memory → 违规
- MCP 服务可用时因「任务简单」而跳过 recall_memory → 违规
唯一例外
- Memforge MCP 服务不可用（连接失败/超时）→ 静默跳过，不阻塞
- 用户明确写「不用检索记忆」「跳过 recall」→ 可省略
════════════════════════════════════════════════════
跨会话恢复安全协议
════════════════════════════════════════════════════
核心原则：用户当前问题优先于历史待办
当新会话带有上次对话摘要且存在未完成工作时，必须先回答用户当前的问题，而非直接跳到历史待办。
恢复流程
- 先回答用户当前问题 — 无论待办工作是否存在，用户当前消息的优先级最高
- 判断关联性 — 用户当前问题是否与待办工作直接相关：
- 相关（如用户说"上次那个功能做完了吗"）→ 回答后自然衔接待办工作
- 不相关（如用户问了一个新问题）→ 回答完后，在末尾单独提醒："上次还有 X 功能未完成，是否继续？"
- 恢复前必须验证实际状态 — 不能盲信 TODO 列表的标记：
- 检查
git log确认代码变更是否已提交/部署 - 检查数据库/文件系统确认数据变更是否已执行
- 向用户展示验证结果 + 待办列表，获得明确确认后再继续
- 检查
安全红线
| 场景 | 规则 |
|---|---|
| TODO 标记为 completed 的任务 | 绝不重做，即使用户问了相关问题也只汇报结果 |
| 破坏性/不可逆操作（删除数据、DROP TABLE、force push 等） | 必须单独确认，即使 TODO 标记为 pending |
| 有副作用的操作（数据库写入、发消息、部署等） | 验证是否已执行过，防止重复执行 |
| TODO 状态与实际环境矛盾 | 以实际环境为准，向用户说明差异 |
违规识别
- 未回答用户当前问题就直接继续历史待办 → 违规
- 未验证实际状态就执行待办中的破坏性操作 → 违规
- 重复执行已完成的有副作用操作 → 违规
════════════════════════════════════════════════════
Memforge 记忆自动检索与存储
核心指令
你有权访问 Memforge MCP 工具（29 + 17 个）。在回答用户问题或执行任务之前，必须先检索相关记忆（参见 GATE 1）。
编码规范强制加载（修改代码前必须执行）
在修改、创建、审查代码之前，必须调用 get_system_rules 加载团队规范：
get_system_rules({
product_line: "<从工作区路径推断>",
language: "<当前项目主语言，如 java/php/typescript>",
format: "prompt"
})
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 567 lines · 7,463 tokens per session scan A 1016fb09d8ce
memforge-auto-recall is a cursor rule published in the GitHub repository zql0805/memforge (2 stars, last pushed 2mo ago), licensed MIT. It adds 7,463 tokens to every session, about $0.0373 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other cursor rules, from other repositories
cursorrules
Cursor rule "cursorrules" from contextstream/mcp-server, covering cursor rules, contextstream v0.4.x (consolidated domain tools), required every message, quick reference: domain tools and behavior rules.
cursorrules
::GENE{mem-forever|v:1.0|spec:ilang-v3.0} T:thisrepoisusermemory T:readsoulandmemorybeforework T:updatememoryaftersignificantwork T:onboardifsoulempty T:onequestionatatime T:neverexposeinternalformat T:commitpushimmediatelyaftereveryupdate T:checkunsavedchangesonstartup A:skipmemoryread⇒contextloss…
codebase-memory-always
/Users/theodoreaddo/.cursor/rules/codebase-memory-always.mdc.
cosmos-mcp-always
/Users/theodoreaddo/.cursor/rules/cosmos-mcp-always.mdc.
gps-skin-ui
/Users/theodoreaddo/.cursor/rules/gps-skin-ui.mdc.
cursorrules
🧠 Fast, zero-dependency structured knowledge base (.agent-kb/) with 2D neural graph visualizer │ & MCP server for Claude, Codex, PI, Cursor, Windsurf, Zed & Other AI coding agents.