Worklog System 2.0: Turn Claude Code Features into a Single CLI Command

Why a CLI tool?

The previous worklog system required interacting with Claude Code, which introduced two pain points: the need to write natural‑language sentences and a dependence on an AI assistant. The worklog CLI eliminates both.

Zero AI dependency : works with rule‑based keyword classification.

One‑command recording : e.g., worklog log dev 完成了登录API.

Full compatibility with the original Claude Code data directory ( kb/).

Cross‑platform : macOS, Linux, Windows.

Quick installation

Requires Node.js ≥ 18. Install with: npm install -g worklog-kb Verify the installation:

worklog --version
# → 1.1.0

Practical example

Using the wechat‑agents project as a demo, initialize the knowledge base:

cd ~/Documents/core-file/wechat-agents
worklog init
# Accept defaults (knowledge‑base path ./kb, LLM provider none)

The command creates the following directory structure under kb/:

kb/
├── inbox/
├── daily/
├── weekly/
├── monthly/
├── tags/
│   ├── index.md
│   ├── dictionary.md
│   └── synonyms.md
└── templates/
    ├── daily-log.md
    ├── weekly-summary.md
    ├── monthly-rollup.md
    └── type‑guide.md

Recording work

Structured input (most common):

worklog log dev 完成文章评分Agent的主编终审逻辑开发
worklog log review 评审了内容策略模块的选题规划方案
worklog log design 完成多Agent协作流水线的架构设计
worklog log bugfix 修复热点响应工作流timeout超时问题
worklog log learn 学习OpenClaw自动化工作流配置

Each command prints a confirmation and creates markdown files in kb/inbox and kb/daily.

Tagging

Add tags with --tags:

worklog log dev 实现chief-editor的发布闸门逻辑 --tags #主编
worklog log dev 完成build.py多平台构建脚本 --tags #构建,#脚本
worklog log review 评审harness合规检查流程 --tags #Harness,#质量闸门
worklog log design 设计文章0-10分评分体系 --tags #项目/微信Agent,#评分

Git integration

When a Git repository is present, worklog git extracts today’s commits and lets you select which ones to record. Selected commits are stored with the dev type and the project name is inferred from the repository.

Optional AI classification

If an LLM is configured, the --ai flag accepts free‑form Chinese sentences, which the model splits into multiple structured entries.

Status and reports

Check the current state: worklog status Generate a weekly report: worklog weekly Generate a monthly report:

worklog monthly

Configuration

Show current settings: worklog config --show Switch LLM providers (DeepSeek, OpenAI, Gemini, Ollama, or none) and change the knowledge‑base path with worklog config --provider … or worklog config --path ….

Best‑practice tag taxonomy

Example tags for the wechat‑agents project include module tags (#主编, #文案), functional tags (#构建, #合规), technology tags (#Python, #TypeScript), and hierarchical tags (#项目/微信Agent, #技术/Python).

FAQ

No LLM? Rule‑based classification covers >90 % of cases.

Data safety: all entries are local markdown files under kb/.

Backup: copy the directory or commit it to Git.

Migration: the CLI can read/write the same kb/ used by Claude Code.

Collaboration: push the repository and let teammates run worklog log.

Weekly range follows ISO week (Monday‑Sunday).

The CLI alone is sufficient for daily logging and report generation; AI is optional for richer notes.

Conclusion

The worklog‑cli provides a lightweight, zero‑dependency way to record development activities, extract Git commits, and produce weekly or monthly reports, while remaining fully compatible with the original Claude Code workflow.