OL-Memory

OL-Memory is a local-first office memory skill for Codex-style agents. It helps an agent remember confirmed workplace context, then use that context to draft practical office documents with clearer source grounding, missing-info reminders, and risk warnings.

OL means Office Life. The name is short, easy to say, and points to the real use case: keeping the useful parts of day-to-day work memory close at hand without turning them into a heavy database.

中文定位：OL-Memory 是你的本地职场记忆库。它不是自动监控工具，也不会替你编造信息；它只使用你确认录入的资料、你在聊天里补充的新信息，以及你要求它学习的文章/短视频/领导偏好信号。

What It Helps With

• Company introductions, project summaries, proposals, tender/bid materials, weekly/monthly reports, meeting notes, review drafts, and performance summaries.
• Reusing your confirmed profile, role, company, service scope, project cases, writing style, and forbidden claims.
• Learning from day-to-day input, including leader-shared articles, short videos, internal phrasing preferences, and recurring standards.
• Producing bilingual missing-info and risk reminders, so drafts are easier to review before sending.

Privacy Model

This repository contains the skill code only. Real user data should live outside the installed skill folder, usually in a local workspace named OL-Memory/.

Do not commit OL-Memory/, .env, logs, UI build folders, or exports containing private workplace information. The included .gitignore already excludes common local data paths.

Install

Clone or download this repository, then copy the skill folder into your Codex skills directory:

mkdir -p ~/.codex/skills
cp -R office-memory-assistant ~/.codex/skills/office-memory-assistant

Restart Codex if it was already open.

First Start

After installation, open the agent chat and say:

开始启动 OL

English trigger:

Start OL

The agent should not open the dashboard first. It should first ask a few required setup questions in chat, then initialize the local memory workspace and launch the page.

Required first-run questions:

1. What should OL-Memory call you, and what is your role?
2. What kinds of office documents do you write most often?
3. How would you describe your company/team and its service scope in one or two sentences?
4. What is one real project/case OL-Memory can remember? You may anonymize it.
5. What must OL-Memory avoid inventing, exaggerating, or mentioning?
6. What writing style, phrases, or leader preferences should OL-Memory follow or avoid?

After you answer, the agent initializes OL-Memory/ and opens a local dashboard, usually at http://127.0.0.1:<port>/.

Daily Use

Most daily use happens directly in the agent chat box. You can write short, natural instructions:

根据 OL-Memory 帮我写一份本周工作总结。

上级刚转发了一篇文章，重点是“客户体验”和“交付闭环”。帮我记录为学习素材，并结合这个方向改一下项目汇报。

这句话以后不要用：“赋能业务高质量发展”。帮我记住。

帮我根据公司介绍、我的岗位和最近项目，写一版投标用的团队能力介绍。

The dashboard is mainly for review, correction, and checking completeness. Chat is the main work entrance.

If The Page Is Not Running

127.0.0.1 is a local page. If the computer restarts, the terminal closes, or the server stops, the page will not keep running by itself.

The normal way to reopen it is to say in chat:

启动 OL

If you need to start it manually:

python3 ~/.codex/skills/office-memory-assistant/scripts/launch_ui.py --data-dir OL-Memory --open

If OL-Memory/ does not exist yet, initialize it first through the chat flow by saying 开始启动 OL.

Manual Initialization Fallback

The preferred path is agent-led setup. If you need a command-line fallback, use intake_profile.py:

python3 ~/.codex/skills/office-memory-assistant/scripts/intake_profile.py \
  --data-dir OL-Memory \
  --display-name "Your Name" \
  --role "Your Role" \
  --doc-types "weekly reports, project summaries, proposals" \
  --company-summary "Your company/team summary" \
  --project-case "One real or anonymized project case" \
  --forbidden-claims "Things the agent must not invent or mention" \
  --writing-preferences "Preferred style and disliked phrases"

Then launch the dashboard:

python3 ~/.codex/skills/office-memory-assistant/scripts/launch_ui.py --data-dir OL-Memory --open

Output Standard

When OL-Memory drafts something, it should provide:

• The usable draft.
• Which confirmed memory it used.
• Missing information reminders in Chinese and English.
• Risk warnings in Chinese and English when claims may be unverifiable, sensitive, or overconfident.
• Suggested memory updates when the user confirms new facts.

Developer Validation

From the repository root:

python3 office-memory-assistant/scripts/self_test.py

For the UI:

cd office-memory-assistant/assets/ui
npm ci
npm run build

For Codex skill validation, run the skill creator validator from your local Codex install if available:

python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py office-memory-assistant

Public Release Checklist

Before publishing:

• Confirm git status --short does not show local memory data.
• Confirm OL-Memory/, logs, .env, node_modules/, and dist/ are not tracked.
• Run the self test and UI build.
• Review README.md for any private company names, real people, or internal project details.