Agent Journal
CherryLover/agent-journal
Gerador de journal de trabalho que captura suas sessões de IA localmente
Ver no GitHubAgent que lê logs de sessão de Codex, Claude Code e Kiro CLI, agrupa trabalho por diretório e gera automaticamente um resumo diário e semanal. Funciona local-first, preservando privacidade dos dados de desenvolvimento.
Instalação
npx convez-ai CherryLover/agent-journal1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar relatórios diários de atividades de desenvolvimento
- Criar resumos semanais para retrospectivas
- Rastrear uso de tokens e créditos entre ferramentas de IA
- Extrair highlights shareable de sessões de trabalho
Veredito

Ferramenta niche mas bem executada para quem trabalha intensivamente com múltiplos CLIs de IA e precisa documentar output. A abordagem local-first é prudente; destaque para o highlight-picking semi-automático que respeita decisão humana.
README
A tradução para o português deste README ainda está na fila.
Agent Journal
A local-first tool that reads session logs from Codex, Claude Code, and Kiro CLI, groups work by directory, and optionally asks an AI model to generate a daily outline and a weekly-ready summary.
Supported tools
| Tool | Log location |
|---|---|
| Codex | ~/.codex/state_5.sqlite + ~/.codex/sessions/**/*.jsonl |
| Claude Code | ~/.claude/projects/<encoded-cwd>/<sessionId>.jsonl (CLI + desktop app's in-app Code; desktop titles enriched from ~/Library/Application Support/Claude/claude-code-sessions/) |
| Kiro CLI | ~/.kiro/sessions/cli/*.json + ~/.kiro/sessions/cli/*.jsonl |
Codex and Claude Code store token counts; Kiro uses a credit system — both are shown in the report.
What it does
- reads real user / assistant chat records from all three tools
- groups sessions by working directory (
cwd) - tracks tool usage, approximate token counts, and Kiro credits
- optionally asks
claudeorcodexto generate:- a compressed daily outline (from real chat logs)
- a weekly-ready summary (directory-grouped, condensed)
- writes 2 final Markdown files per day
- highlight picking with human decision (
highlights.py): scans recent sessions every ~4 hours during the day, extracts shareable highlights with a suggested channel and angle, and shows them on the desktop app's default "亮点" page for you to adopt or dismiss; adopted items are appended to a topic-bank Markdown — no final copy is auto-written
Usage
Launch the local desktop UI:
python3 ui_app.py
The UI uses Python's standard Tkinter library and provides a left-side navigation shell for generating journals, reviewing history, creating share drafts, and viewing run logs. If your Python build does not include Tk support, install the matching python-tk package first.
Without AI summarization (just collect and export):
python3 journal_mvp.py --date 2026-03-11
With AI summarization:
python3 journal_mvp.py --date 2026-03-11 --summarizer claude
# or
python3 journal_mvp.py --date 2026-03-11 --summarizer codex
All options:
--date Target date in YYYY-MM-DD format (required)
--summarizer none | claude | codex (default: none)
--summary-model Optional model name for the summarizer
--result-dir Output directory for final Markdown files
--timezone IANA timezone string (default: Asia/Shanghai)
--home Home directory to read logs from (default: ~)
Output
By default the script writes to ~/Documents/ObCc/01_Journal/Agent_Journal/. To use a different directory, pass --result-dir:
python3 journal_mvp.py --date 2026-03-11 --summarizer claude \
--result-dir ~/my-journal
Two files are written per day:
<date>_outline.md— daily outline<date>_weekly-read.md— weekly review material