From 1705835c4304ef69ca6ec0e07279418028ed34d2 Mon Sep 17 00:00:00 2001
From: Kailigithub <12250313+Kailigithub@users.noreply.github.com>
Date: Sun, 26 Jul 2026 03:11:38 +0800
Subject: [PATCH] fix(memory): add Step 3.5 (codebase verification) to
 task_planning SOP
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Closes #560 — when generating TODOs in planning mode, agent should
first verify candidate features/modules do not already exist in the
codebase before writing 'evaluate/integrate' class TODOs.

Re-roll of #561 onto current main (the original was closed after 46
days without review activity per the v1.19 housekeeping protocol).

Diff: replaces the non-existent verify_custom_features helper with
concrete shell commands (find/grep) so the verification step is
actually executable in any environment.
---
 memory/autonomous_operation_sop/task_planning.md | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/memory/autonomous_operation_sop/task_planning.md b/memory/autonomous_operation_sop/task_planning.md
index 209d5184e..3e251b5bc 100644
--- a/memory/autonomous_operation_sop/task_planning.md
+++ b/memory/autonomous_operation_sop/task_planning.md
@@ -13,6 +13,9 @@
      - 避免和历史重复和低价值边际效应
   2. 反思：为什么这些任务低价值？如何设计才能高价值？
   3. **批判性**盘点已有报告和记忆（ls autonomous_reports/ + ../memory），考虑如何发挥更大价值或优化
+  3.5. **现状核实**：每条候选TODO涉及的功能/模块，先在代码库中确认是否已实现（`find . -name "<特征名>"` 或 `grep -rln "<特征名>"`），避免重复规划已存在的能力
+     - 产出类TODO（"评估X"/"集成X"）→ 先确认代码中不存在对应实现，再写入
+     - 已实现的功能只能写"优化/增强/测试效果"类TODO，禁止再写"评估/集成"
   4. 综合以上，产出5-7条TODO写入 `TODO.txt`，TODO已完成内容可压缩丢后面
   5. 每条格式：`[ ] 类型(产出/冲浪/环境) | 一句话目标 | 验收标准`
   6. **必须召唤subagent评审TODO**（保证可信三方）：input仅给TODO列表+"读记忆库和本sop自行判断，逐条评分1-10并简述理由"（不喂额外先验信息）