CommandCodeAI / CommandCodeAI/command-code

Toggle Taste learnings displays

未关闭
#782 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

主要语言
没有语言数据
星标
4k
派生
350
PR 合并指标
30 天内没有已合并 PR

描述

Summary

I've had extreme amounts of output from taste.md learnings, so much so that sometimes the output from taste.md is much more that the actually needed output. In such cases it would be handy to have a toggle option to keep toggle taste learning display.

Here is an eg
output is interleaved with taste info + 2 blocks of taste op for each block of actual info

❯ ok then give me a better way 
  explain approach 1 to me     

✻ Thought for 1 second [ctrl+o to expand]

⠶ Approach 1 — sizeof works because the compiler knows the array size at compile time in
  the same scope where it was declared:

  char *a[] = {"hello", "world"};
    size_t n = sizeof(a) / sizeof(a[0]);   // 2

  sizeof(a) gives the total bytes — 2 pointers × 8 bytes = 16 on 64-bit. sizeof(a[0])
  gives one pointer = 8 bytes. 16/8 = 2. It's computed at compile time, no runtime cost,
   no sentinel needed.

 TASTE  Reinforced
 └ Asks conceptual "why" questions to understand the underlying logic/reasoning behind
   a suggested fix before applying it, rather than blindly following instructions 
   [Confidence 70% → 80%]
   taste.md

 TASTE  Reinforced
 └ Asks conceptual "why" questions to understand the underlying logic/reasoning behind
   a suggested fix before applying it, rather than blindly following instructions 
   [Confidence 70% → 80%]
   --prefers-the-assistant-to-first-read-and-understand-the-problem-spec-(e.g.,-lab-pdf
   )-and-existing-code-progress,-then-wait-for-explicit-instructions-before-making-chan
   ges.-confidence:-0.7/taste.md

  But it only works when a is the actual array. Pass it to a function, it decays to
  char** and sizeof(a) becomes just the pointer size (8), so the trick fails.

  The better way for runtime use: pass the size along with the array.

  void f(char **a, size_t count);

  That's what every real C API does (argc/argv style). It's simple, explicit, never
  reads out of bounds, and you know the count the moment you built the array — no
  guessing, no sentinel, no UB.

 TASTE  Moved
 └ - Prefers the assistant to first read and understand the problem spec (e.g., lab
   PDF) and existing code progress, then wait for explicit instructions before making
   changes. Confidence: 0.7
   --prefers-the-assistant-to-first-read-and-understand-the-problem-spec-(e.g.,-lab-pdf
   )-and-existing-code-progress,-then-wait-for-explicit-instructions-before-making-chan
   ges.-confidence:-0.7/taste.md

 TASTE  Moved
 └ - Prefers the assistant to first read and understand the problem spec (e.g., lab
   PDF) and existing code progress, then wait for explicit instructions before making
   changes. Confidence: 0.7
   --prefers-the-assistant-to-first-read-and-understand-the-problem-spec-(e.g.,-lab-pdf
   )-and-existing-code-progress,-then-wait-for-explicit-instructions-before-making-chan
   ges.-confidence:-0.7/taste.md

 ✻ Worked for 7s

So just to clarify I want taste learning, but I do not want to annoyed by its "learning" each and every time

Expected Behavior

Silent updates to taste.md

Actual Behavior

I asked it to stop showing me these outputs and this is the response.

 TASTE  Learned
 └ Does not want taste-learning updates/pop-ups surfaced or referenced during the
   conversation; still wants the assistant to keep learning their taste silently in the
   background [Confidence 95%]
   taste.md
Steps to reproduce the issue

Do anything

Command Code Version

1.39.2

Operating System

Linux

Terminal/IDE

VSCode

Shell

zsh

Session file (optional)

No response

Fix prompt (optional)

No response

Additional context

No response

贡献指南

这个仓库没有索引到贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

调研方向

首先定位负责渲染 TASTE Learned、Reinforced 和 Moved 消息并处理 taste.md 更新的 CLI 代码;该 issue 未指定具体的源文件或测试。添加一个由用户控制的开关,在抑制这些显示的同时保留向 taste.md 的学习,并验证静默更新仍会写入 taste.md,而正常输出不受影响。

由索引模型根据 Issue 内容生成。

评估

领域
ai, cli
Issue 类型
功能
难度
3/5
预计耗时
1-2 天
活跃度
活跃
描述清晰度
基本清楚
新手友好度
52/100

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。