Activity Journal

Local-first activity journal for tracking what you studied, built, decided, and need to clarify.

This MVP collects evidence from:

• Codex CLI history and session index
• Git repositories under configured project roots
• Recently modified files under configured project roots
• Optional inbox notes, ChatGPT exports, live ChatGPT/Gemini browser captures, browser history, and foreground app activity

It writes daily and weekly Markdown drafts, builds project-level rollups, indexes local evidence in SQLite, and can sync summary/final logs into Notion.

What You Get

• Daily logs under journal/daily/YYYY-MM-DD.md
• Weekly reviews under journal/weekly/YYYY-Www.md
• Project rollups under journal/projects/
• A local health report for scheduler, Notion, SQLite, capture, and file access
• A local browser dashboard for health, daily sections, project rollups, questions, and SQLite search
• Optional Windows scheduled tasks, tray controls, and browser-extension capture
• Privacy-first defaults: generated journals, raw logs, local config, databases, and secrets stay untracked

Repository Layout

browser_extension/   Optional local ChatGPT/Gemini capture extension source
config/              Public example config only; private config is ignored
docs/                Architecture, demo output, publishing, and privacy/security notes
prompts/             Codex review prompt template
scripts/             Collectors, sync jobs, health checks, tray/settings apps

See docs/demo-daily-log.md for a synthetic daily output example. See CONTRIBUTING.md before proposing changes that touch capture, sync, or local data handling.

Public Repository Safety

This repository is safe to publish only when local data stays untracked. The committed config is config/activity-journal.example.json; each user gets a private config/activity-journal.json during setup. The .gitignore excludes local journals, imports, inbox files, manual notes, SQLite databases, config backups, and Chrome extension signing artifacts.

Before publishing or committing, review:

• docs/privacy-security.md
• docs/publishing.md

Quick Start

1. Install dependencies, create local folders, and create config/activity-journal.json from the example if it does not exist:

powershell -ExecutionPolicy Bypass -File scripts/install_local.ps1 -WhatIf
powershell -ExecutionPolicy Bypass -File scripts/install_local.ps1

2. Edit config/activity-journal.json for your timezone, project folders, and optional integrations.
3. Run a daily collection:

python scripts/collect_daily.py

4. Run a weekly review:

powershell -ExecutionPolicy Bypass -File scripts/run_weekly.ps1

When scripts/run_daily.ps1 or the Python CLIs run without --date, they use config/activity-journal.json's timezone value to decide today's date. The daily evidence window is that timezone's midnight-to-midnight range. Passing --date YYYY-MM-DD uses that explicit date.

Run the interactive daily review flow:

powershell -ExecutionPolicy Bypass -File scripts/run_daily.ps1

Run evidence collection only:

powershell -ExecutionPolicy Bypass -File scripts/run_daily.ps1 -NonInteractive

Scheduled daily runs use -RefreshDrafts, so an unreviewed draft can be refreshed with the latest same-day evidence before the Notion sync runs. Manual runs without that switch still preserve existing review Markdown by default.

Re-running collection for the same date refreshes journal/raw/YYYY-MM-DD.json but preserves existing journal/daily and journal/questions Markdown files so reviewed answers are not overwritten.

To intentionally regenerate those review Markdown files, use python scripts/collect_daily.py --date YYYY-MM-DD --overwrite-review-files. Use this only for recovery because it can overwrite reviewed answers.

Privacy note: codex.include_history_text controls whether Codex history/session evidence stores only IDs and timestamps or also richer local context. The example config defaults this to false; enable it only if you want richer local context in raw evidence.

Optional manual notes can fill gaps that automation cannot see. Create manual_notes/YYYY-MM-DD.md before collection and write short bullet lines:

- Learned how Codex skills and Notion sync fit together.
- Reviewed a browser article about weekly review systems.

Those lines are added to the daily Studied section.

External inputs can expand what the system sees beyond Git and Codex. The collector reads:

• inbox/YYYY-MM-DD.md
• inbox/YYYY-MM-DD/*.md
• inbox/YYYY-MM-DD/*.txt
• recently modified inbox/*.md and inbox/*.txt
• ChatGPT export .zip files or conversations.json files under imports/chatgpt
• live ChatGPT/Gemini browser captures from the local chatgpt_live_server.py receiver
• local Chromium browser history from Chrome, Edge, Brave, and Vivaldi profiles
• recently modified study/work files under configured Desktop, Documents, and Downloads roots
• foreground app/window activity captured by scripts/activity_watch.py, including visible accessibility text for supported active apps when available

Open the official ChatGPT/OpenAI export flow:

powershell -ExecutionPolicy Bypass -File scripts/open_chatgpt_export.ps1

After the export email arrives, place the downloaded .zip or conversations.json under imports/chatgpt. The next daily collection will read matching conversations for the target date. The helper script only opens the official pages; it does not store passwords, tokens, cookies, or browser session data.

External input privacy note: external_inputs.include_raw_text controls whether inbox, ChatGPT, and visible active-window text is stored in journal/raw/YYYY-MM-DD.json. The example config keeps external inputs and raw text disabled by default. Browser history stores visited page titles, URLs, timestamps, profile names, and visit counts, but not passwords, cookies, or browser session tokens. Live ChatGPT/Gemini capture stores page text sent by the optional local browser extension to 127.0.0.1. Recent-file tracking stores paths, filenames, modified times, and sizes. Activity watch stores foreground window titles, process names, timestamps, sample counts, and visible accessibility text for supported apps when Windows exposes it. Daily Markdown receives summarized study signals plus ChatGPT Live Captures, Browser Activity, Recent Files, and App Activity sections.

Optional high-fidelity live ChatGPT/Gemini page-text capture:

powershell -ExecutionPolicy Bypass -File scripts/open_chatgpt_capture_extension.ps1

The foreground activity watcher now captures visible active-window text automatically when Windows exposes it. The local browser extension is still available if you want a more complete in-page ChatGPT/Gemini transcript. Chrome requires one manual approval for a local unpacked extension. The repository keeps only the unpacked source at browser_extension/chatgpt-live-capture/; private signing keys and packaged update artifacts belong in %LOCALAPPDATA%\ActivityJournal\browser_extension\ or the directory named by ACTIVITY_JOURNAL_EXTENSION_ARTIFACT_DIR. After the extension is loaded, normal ChatGPT/Gemini use is captured automatically through the local receiver; no password, cookie, or browser session token is stored by Activity Journal.

Run local smoke tests:

python scripts/smoke_tests.py

Open the local dashboard:

powershell -ExecutionPolicy Bypass -File scripts/open_dashboard.ps1

The dashboard runs on 127.0.0.1:8776 by default. It presents a Korean-first view of today's status, recommended actions, daily sections, project rollups, capture health, and SQLite search results without uploading data.

Check whether the automation is healthy:

python scripts/project_health.py

This writes journal/system_status.md and journal/raw/system_status.json. The report checks today's local files, pending Codex Review questions, weekly review presence, Notion token/state setup, Windows scheduled tasks, and journal folder access. It does not call the Notion API and does not print token values.

If the health report says Recent file item limit reached, the collector is working but found more recently modified files than it is configured to keep. Raise external_inputs.recent_files.max_items in your private config/activity-journal.json, narrow external_inputs.recent_files.roots, or add noisy folders to the scan ignore list.

Run question quality diagnostics:

python scripts/question_quality.py --date YYYY-MM-DD

This writes journal/raw/question_candidates_YYYY-MM-DD.json. The file is a generated diagnostic artifact and can be overwritten by the next run. It does not modify journal/questions/YYYY-MM-DD.md or journal/daily/YYYY-MM-DD.md; Codex Review reads the candidates and decides whether any question is actually needed.

Build project-level rollups:

python scripts/project_review.py --date YYYY-MM-DD

This writes journal/projects/<project-slug>.md and journal/raw/project_rollup_YYYY-MM-DD.json. Project goals, status, and links live in journal/projects/project_metadata.json; existing metadata values are preserved, and new projects only receive empty default records until Codex Review asks for the missing goal.

Run safe auto recovery:

python scripts/auto_recover.py --date YYYY-MM-DD

This checks health, runs only allowlisted local generation repairs, writes journal/recovery_status.md and journal/raw/recovery_YYYY-MM-DD.json, then opens Codex Review only when user decisions remain. Scheduled daily runs call it with --no-open-codex --from-run-daily so non-interactive collection does not open a review window.

Generated files:

journal/daily/YYYY-MM-DD.md
journal/weekly/YYYY-Www.md
journal/projects/<project-slug>.md
journal/questions/YYYY-MM-DD.md
journal/raw/YYYY-MM-DD.json
journal/system_status.md
journal/recovery_status.md
journal/raw/system_status.json
journal/raw/question_candidates_YYYY-MM-DD.json
journal/raw/project_rollup_YYYY-MM-DD.json
journal/raw/recovery_YYYY-MM-DD.json

Daily Notion Routine

Use the generated daily file as the Notion Daily Activity Log draft.

Recommended Notion fields:

• Date
• Projects
• Studied
• Built
• Decisions
• Problems
• Next Actions
• Sources
• Status: Draft or Confirmed

The journal/questions file is a local-only review queue. It is not meant to be copied to Notion.

The preferred flow is interactive Codex Review. The scheduled collector gathers evidence first, then opens Codex so Codex can decide whether any clarification is actually needed. If the raw evidence and daily draft are already clear, Codex should not ask extra questions.

When Codex does ask, it asks in Korean, explains what evidence triggered the question, and includes a practical recommended answer. Your answer is saved back into journal/questions/YYYY-MM-DD.md, merged into the daily log, then weekly review and Notion sync are run.

Questions are written in a structured format:

## Q: <stable_id>
Category: <배운 점|작업명|상태|결정|다음 행동|설정 문제>
Confidence: <high|medium|low>

Question:
<사용자에게 물어볼 질문>

Context:
- <왜 묻는지>
- <어떤 파일/상황을 근거로 묻는지>

Recommendation:
<추천 답변>

Options:
- 추천 답변으로 저장
- 수정해서 저장: ...
- 저장하지 않음

Answer:

이 형식은 이렇게 읽으면 됩니다:

• stable_id는 생성된 질문과 답변을 연결하는 내부 ID입니다. 사용자가 직접 관리할 필요는 없습니다.
• Context는 Codex가 왜 묻는지, 어떤 근거가 불명확했는지 설명합니다.
• Recommendation은 Codex가 가장 실용적이라고 판단한 추천 답변입니다. 맞아 보이면 추천 답변으로 저장이라고 답하면 됩니다.
• 추천 답변이 대체로 맞지만 표현을 바꾸고 싶으면 수정해서 저장: ... 뒤에 원하는 답변을 적습니다.
• 기록에 남길 필요가 없으면 저장하지 않음이라고 답합니다.

Notion Sync

Notion sync can be enabled in config/activity-journal.json, but the token is not stored in this repo. The sync prefers Notion databases (Daily Activity Log, Weekly Review, and Project Log) and falls back to ordinary child pages if database creation is unavailable.

Set a user environment variable if you want scheduled sync:

powershell -ExecutionPolicy Bypass -File scripts/set_notion_token.ps1

The sync script can also read the saved Windows user environment value directly, so a full restart is usually not required for manual runs. Scheduled tasks may still require a fresh logon session on some Windows setups.

Notion page and database IDs are tracked in journal/raw/notion_pages.json to avoid duplicate pages/databases. Do not edit or delete this file unless you intentionally want to reconnect Notion state; if it is damaged, local Markdown logs remain separate.

For daily logs, the local Markdown Status: line is the source of truth for the Notion Status property. Status: Confirmed syncs as Confirmed; anything else syncs as Draft. Codex Review should only mark a daily log as Confirmed after all questions are answered and unresolved Problems are cleared.

Project logs sync to the Notion Project Log database when journal/projects/*.md exists. Project Status is limited to Active, Paused, or Done; unknown or invalid local values sync as Active.

Verify token setup:

powershell -ExecutionPolicy Bypass -File scripts/set_notion_token.ps1 -VerifyOnly

Local MVP Settings App

Open the local settings UI:

python scripts/settings_app.py

Open the dashboard directly:

python scripts/dashboard_app.py

Validate the config without opening the UI:

python scripts/settings_app.py --validate-config

Install or repair the local MVP shortcuts and scheduled tasks:

powershell -ExecutionPolicy Bypass -File scripts/install_local.ps1 -WhatIf
powershell -ExecutionPolicy Bypass -File scripts/install_local.ps1

Uninstall scheduled tasks and shortcuts while preserving journal\ and config\ data:

powershell -ExecutionPolicy Bypass -File scripts/uninstall_local.ps1 -WhatIf
powershell -ExecutionPolicy Bypass -File scripts/uninstall_local.ps1

The settings UI can pause/resume all capture, toggle individual sources, start temporary privacy mode, and edit excluded apps/domains.

Start the tray icon manually:

python scripts/tray_app.py

The tray menu supports opening the dashboard, quick pause for 15/30/60 minutes, pause until tomorrow, resume now, opening Settings, refreshing status, and running a health check. scripts/install_local.ps1 also installs the tray dependencies (pystray, Pillow), creates Start Menu shortcuts, and adds a Windows startup launcher.

Schedule on Windows

Preview the task registration command:

powershell -ExecutionPolicy Bypass -File scripts/setup_task.ps1 -WhatIf

Register daily and weekly scheduled tasks:

powershell -ExecutionPolicy Bypass -File scripts/setup_task.ps1

The scheduled tasks run as follows:

• Activity Journal Daily: collects evidence and writes a local draft at 23:50.
• Activity Journal Codex Review: opens Codex at 23:51 so questions and answers stay inside Codex.
• Activity Journal Weekly: writes a weekly review, refreshes project rollups, and syncs them to Notion on Sunday at 23:55.
• Activity Journal Watcher: starts at Windows logon and records foreground app/window changes for later daily collection.
• Activity Journal ChatGPT Receiver: starts at Windows logon and receives optional browser-extension captures on 127.0.0.1:8765.
• Activity Journal Tray: starts at Windows logon from the Startup folder and gives quick pause/resume controls.

Activity Journal Codex Review opens only when the matching journal/raw/YYYY-MM-DD.json and journal/daily/YYYY-MM-DD.md files exist. If it does not open, run powershell -ExecutionPolicy Bypass -File scripts/run_daily.ps1 -NonInteractive first to regenerate the daily inputs.

Tasks use StartWhenAvailable, so if the computer was off at the scheduled time, Windows can run the missed task after the computer is available again.

Limits

This does not automatically read every ChatGPT web conversation. For ChatGPT, use one of these:

• Export important conversations and place them in a configured input folder.
• Paste important conclusions into the generated daily draft.
• Add a future API-based workflow if you use ChatGPT through an API or controlled app.

Auto recovery does not delete files, change permissions, register scheduled tasks, set Notion tokens, or retry Notion API writes. Those remain explicit user actions.