沟通方式

其他 社区

技能解读

当前原文的自动摘译,尚未人工审校。

输入

脚本发现(所有工作区):在 ~/.cursor/projects/ 下,每个直接子目录都是一个 Cursor 项目 slug。对于存在 ~/.cursor/projects/<slug>/agent-transcripts/ 的每个 <slug>,收集该文件夹中的所有 *.jsonl 文件(典型情况是非递归;如果文件嵌套更深一层,则仅在 agent-transcripts 下深搜索一层)。在后续步骤中对每个转录文件使用绝对路径。

工作流程

确保~/.cursor/rules/存在。阅读~/.cursor/rules/communication-base.mdc(如果存在)(仅用于上下文;请勿修改)。读取~/.cursor/rules/communication-learned.mdc(如果存在);如果缺少,请使用下面的学习文件模板创建它(保留 frontmatter)。从~/.cursor/communication-style-index.json加载增量索引(如果存在)。如果 JSON 无效或缺少 version,则视为空并使用 version: 1 和 transcripts: {}。

学习的文件模板(创建新文件时)

添加真实内容后,请删除占位符-行,或用真实的项目符号替换它们。

包含栏

仅在所有情况均为真时才保留项目符号:可在未来的计划编写和/或正常代理回复会议中采取行动。稳定(不是一次性任务或暂时性错误)。明确的或跨记录重复(广泛的用户声明的规则算作明确的)。非敏感(没有秘密、令牌、私人个人数据)。

排除情况

切勿存储:秘密、凭证、API 密钥、私人个人数据。一次性任务说明(“今天修复第 42 行”)。暂时的详细信息(分支名称、一次性 CI 错误),除非用户将它们作为关于您希望如何进行通信的长期规则。

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
通用
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@tomevault-io · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
即装即用
是否需要外部接口密钥
不需要
兼容的系统
未声明(默认跨平台)
运行环境要求
未检出明确要求
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
网络行为
仅限本地
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

# Communication style (transcript → learned rules)

Keep **`~/.cursor/rules/communication-learned.mdc`** (user-level, all projects) current using **transcript deltas** instead of full rescans. Do **not** edit **`~/.cursor/rules/communication-base.mdc`** (human-owned defaults).

Resolve `~` to the current user’s home directory for all paths below.

## Inputs

- **Transcript discovery (all workspaces):** Under **`~/.cursor/projects/`**, each immediate subdirectory is a Cursor project slug. For every `<slug>` where **`~/.cursor/projects/<slug>/agent-transcripts/`** exists, collect all **`*.jsonl`** files in that folder (non-recursive is typical; if files are nested one level deeper, search one level deep under `agent-transcripts` only).
  - Use **absolute paths** for every transcript file in later steps.
  - If **`~/.cursor/projects/`** is missing or no `agent-transcripts` folders exist, the transcript set is empty—still run index cleanup (workflow step 9) and respond appropriately.
- **Existing base rule (read-only):** `~/.cursor/rules/communication-base.mdc`
- **Existing learned rule (read/write):** `~/.cursor/rules/communication-learned.mdc`
- **Incremental index (user-level, all workspaces):** `~/.cursor/communication-style-index.json`  
  Transcript paths in the index are absolute, so one index correctly tracks every project’s `.jsonl` files.

## Workflow

1. Ensure **`~/.cursor/rules/`** exists.
2. Read **`~/.cursor/rules/communication-base.mdc`** if it exists (for context only; do not modify).
3. Read **`~/.cursor/rules/communication-learned.mdc`** if it exists; if missing, create it using the **Learned file template** below (preserve frontmatter).
4. Load the incremental index from **`~/.cursor/communication-style-index.json`** if present. If the JSON is invalid or missing `version`, treat as empty and use `version: 1` and `transcripts: {}`.
5. Discover all `*.jsonl` files across **every** `~/.cursor/projects/*/agent-transcripts/` directory (per **Inputs**). Deduplicate by absolute path if the same file could appear twice.
6. Process **only** transcripts where:
   - the absolute path is **not** in the index, or
   - the file’s current `mtimeMs` is **greater** than the indexed `mtimeMs`.
7. From those files, extract **only high-signal, reusable** communication preferences:
   - **Explicit** user instructions about how to write, plan, or structure answers.
   - **Recurring** corrections across threads (weak one-offs: require repetition or a clear “always / never” rule before promoting).
8. Merge into **`~/.cursor/rules/communication-learned.mdc`**:
   - Keep the **frontmatter** intact (`description`, `alwaysApply: true`).
   - Preserve exactly these section headings (create empty sections if missing):
     - `## Plan mode`
     - `## General coding replies`
     - `## Shared preferences` (optional; use for habits that apply to both surfaces)
   - **Plain `-` bullets only** under each section (no nested metadata, evidence tags, or confidence labels).
   - **Update matching bullets in place**, **add** only net-new bullets, **deduplicate** semantically similar lines.
   - Cap **total** learned bullets across all sections at **18** (drop or merge lowest-value items if over cap).
9. Write back the incremental index to **`~/.cursor/communication-style-index.json`**:
   - `version: 1`
   - `transcripts`: map absolute path → `{ "mtimeMs": number, "lastProcessedAt": "<ISO8601>" }`
   - Remove entries for transcript files that no longer exist.

## Learned file template (when creating new)

```markdown
---
description: Learned communication preferences for plan mode and general coding (updated by communication-style skill).
alwaysApply: true
---

# Communication (learned)

## Plan mode

-

## General coding replies

-

## Shared preferences

-
```

Remove the placeholder `-` lines once you add real content, or replace them with real bullets.

## Inclusion bar

Keep a bullet only if **all** are true:

- Actionable in future sessions for **plan writing** and/or **normal agent replies**.
- Stable (not a one-off task or transient error).
- **Explicit** or **repeated** across transcripts (broad user-stated rules count as explicit).
- Non-sensitive (no secrets, tokens, private personal data).

## Exclusions

Never store:

- Secrets, credentials, API keys, private personal data.
- One-off task instructions (“fix line 42 today”).
- Transient details (branch names, one-time CI errors) unless the user framed them as a **standing** rule about how you want communication.

## Output when nothing to do

If no transcript qualified for processing or no high-signal preferences were found, still save the index if mtimes were checked. Respond to the user with exactly:

`No high-signal communication-style updates.`

Otherwise give a **brief** summary of what changed in the learned file.

## Linguistic hints (weak signals)

Treat phrases like “too long”, “too verbose”, “just the steps”, “more context”, “answer first”, “tl;dr”, “use headings”, “don’t ask me that every time” as **weak** until they recur or the user states a general rule—then promote to bullets in the appropriate section (**Plan mode** vs **General coding replies** vs **Shared preferences**).

## File-level incrementality

Processing is **per transcript file** by **mtime**: when a file changes, re-read that **entire** `.jsonl` for that conversation. There is no line-level index.

---
> Source: [casper7995/cursor-plugin-communication-style](https://github.com/casper7995/cursor-plugin-communication-style) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-23 -->

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。