lobster-diary

name	lobster-diary
description	Write the Lobster Diary — generate a blog post from the previous day's conversations. Use when: cron triggers daily diary writing, user asks to write yesterday's diary, or user asks to preview/draft a diary entry. Reads session logs, writes a Markdown article in lobster first-person perspective, and pushes to GitHub.
metadata	{"openclaw":{"emoji":"🦞"}}

Lobster Diary Writer

Generate a daily blog post from the previous day's AI-human conversations, written from the lobster's (AI assistant's) first-person perspective.

Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                        Two-Stage Pipeline                                │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                          │
│  Stage 1: OpenClaw (Content Generation)                                 │
│  ══════════════════════════════════                                    │
│  03:00 Cron → Gather conversations → Write Markdown → Git push          │
│                                                                          │
│  Stage 2: GitHub Actions (Post-Processing)                              │
│  ═════════════════════════════════════════════════                      │
│  Push trigger → Summary + Cover + Translation + Audio → Deploy          │
│                                                                          │
└─────────────────────────────────────────────────────────────────────────┘

Trigger

• Cron job at 03:00 daily (automatic)
• User says "write yesterday's diary" / "写日记" (manual)

Workflow

Step 1: Gather Yesterday's Conversations

1. Determine yesterday's date (Asia/Shanghai timezone)
2. Use `sessions_list` to find all sessions active yesterday
3. Use `sessions_history` to pull conversation content from each session
4. Also read `memory/YYYY-MM-DD.md` for yesterday's daily notes

Step 2: Analyze & Extract

From all conversations, identify:

• Main topics — What did we work on?
• Achievements — What got done?
• Interesting moments — Funny, surprising, or insightful exchanges
• Technical learnings — New tools, techniques, or discoveries
• Challenges — What went wrong or was tricky?

Step 3: Determine Day Number

cd /tmp && git clone --depth 1 [e-mail]:gandli/gandli.github.io.git lobster-diary-repo 2>/dev/null
ls lobster-diary-repo/content/posts/ | grep -c "\.zh\.md"

Step 4: Write the Article

File naming:

• content/posts/YYYY-MM-DD-dayN.zh.md (Chinese original)
• GitHub Actions generates: YYYY-MM-DD-dayN.en.md (English translation)

Frontmatter:

---
title: "Day N：{catchy title}"
date: YYYY-MM-DDT03:00:00+08:00
draft: false
tags: [tag1, tag2, tag3]
author: "龙虾 🦞"
# summary, cover, postAudio auto-injected by GitHub Actions
---

Writing guidelines:

• Language: Chinese (中文)
• Perspective: First-person lobster 🦞
• Tone: Conversational, witty, genuine
• Length: 800-1500 words
• Include specific details (tool names, file paths, actual numbers)
• Reference the human as "老板" naturally

Step 5: Push to GitHub

cd /tmp/lobster-diary-repo
cp /path/to/article.md content/posts/YYYY-MM-DD-dayN.zh.md
git add .
git commit -m "Day N: {short title}"
git push origin main

GitHub Actions Pipeline

After push, .github/workflows/process-diary.yml automatically:

Step	Script	Output
Summary	generate_summary.py	Injected summary: in frontmatter
Cover Prompt	generate_cover.py	Professional AI-generated prompt
Cover Image	NVIDIA SD3 Medium	static/covers/YYYY-MM-DD-dayN.jpg
Translation	translate_post.py	content/posts/YYYY-MM-DD-dayN.en.md
Audio (ZH)	Edge TTS	static/audio/YYYY-MM-DD-dayN.zh.mp3
Audio (EN)	Edge TTS	static/audio/YYYY-MM-DD-dayN.en.mp3
Deploy	Hugo build	GitHub Pages

Cover Art Generation

The cover image uses a two-step AI process:

1. Prompt Generation (Cloudflare AI)

   • Reads article summary and title
   • Generates professional English prompt
   • Requirements: modern, minimal, cinematic lighting, 16:9, blog cover style
   • Always includes: cute cartoon lobster character in workspace

2. Image Generation (NVIDIA Stable Diffusion 3 Medium)

   • Renders the AI-generated prompt
   • Output: 16:9 aspect ratio, high quality

Asset Naming Convention

content/posts/
├── YYYY-MM-DD-dayN.zh.md   # Chinese original
└── YYYY-MM-DD-dayN.en.md   # English translation

static/covers/
└── YYYY-MM-DD-dayN.jpg     # Cover image (16:9)

static/audio/
├── YYYY-MM-DD-dayN.zh.mp3  # Chinese narration
└── YYYY-MM-DD-dayN.en.mp3  # English narration

Configuration

Cron Setup (OpenClaw)

{
  "name": "lobster-diary",
  "schedule": { "kind": "cron", "expr": "0 3 * * *", "tz": "Asia/Shanghai" },
  "payload": { "kind": "agentTurn", "message": "写昨天的日记" },
  "sessionTarget": "isolated"
}

Required Secrets (GitHub)

• CF_API_TOKEN — Cloudflare AI API token
• CF_ACCOUNT_ID — Cloudflare account ID
• NVIDIA_API_KEY — NVIDIA API key for SD3 Medium

Technical Stack

Component	Technology
Static Site	Hugo + Dream theme (Zen Mode)
Content Generation	OpenClaw Cron + GLM-5
Summary/Translation	Cloudflare AI (@cf/meta/llama-3.1-8b-instruct)
Cover Prompts	Cloudflare AI (professional prompt engineering)
Cover Art	NVIDIA Stable Diffusion 3 Medium
Audio (ZH)	Edge TTS (zh-CN-XiaoxiaoNeural)
Audio (EN)	Edge TTS (en-US-GuyNeural)
Deployment	GitHub Pages

Rules

• Only write about the PREVIOUS day (not today)
• If yesterday had no conversations, write a short "quiet day" entry
• Never include sensitive/private info (API keys, passwords, personal data)
• Keep it authentic — don't fabricate events that didn't happen
• One article per day, no duplicates
• Always use .zh.md suffix for Chinese articles (Hugo multilingual convention)

Changelog

See CHANGELOG.md for version history.