Conversation Logger

Conversation Logger

Auto-write structured daily conversation summaries into project-specific folders. Each project gets its own directory with daily log files, auto-read on first turn each day, and on-demand manual lookup.

When to Use

• Setting up daily auto-logging for multiple parallel projects
• Building conversation continuity — agent auto-reads yesterday's log on first daily turn
• Organizing chat history by project folder, not by session
• Enabling on-demand historical lookup ("show me yesterday's LED project log")
• Any OpenClaw workspace with multiple project contexts that need isolated histories

Setup

1. Create the Skill File

Place this SKILL.md at:

{workspace}/skills/conversation-logger/SKILL.md

2. Create Log Directories

# PowerShell
$projects = @("main", "my-project-a", "my-project-b")
$base = "D:\OpenClaw Memory"
foreach ($p in $projects) {
    New-Item -ItemType Directory -Path "$base\$p" -Force
}

# Linux / macOS
mkdir -p ~/".openclaw-memory"/{main,my-project-a,my-project-b}

3. Update AGENTS.md

### 🗂️ 项目对话日志（Conversation Logger）

参考 skill: `skills/conversation-logger/SKILL.md`

**自动记录结构：**
- 每次对话结束，**静默**追加结构化摘要到 `D:\OpenClaw Memory\{project}\memory YYYY-MM-DD.md`
- 格式：主题/上下文 → 我的提问 → 决策/结论 → 涉及文件/配置

**每日首次自动调取：**
- 当天第一次对话时，读取前一日对应项目的日志作为上下文
- 当天内如需调取过往日志，手动要求即可

**当前项目文件夹映射：**
| 项目/场景 | 文件夹 |
|---|---|
| 日常主对话 | `main` |
| 项目 Foo | `project-foo` |
| 项目 Bar | `project-bar` |

Log Entry Format

Each log entry follows this template:

## YYYY-MM-DD HH:mm

**主题/上下文**: [session summary in one line]
**我的提问**: [user's core question]
**决策/结论**: [confirmed decisions, tech choices, config changes]
**涉及文件/配置**: [file paths, config keys, tools used]

Example:

## 2026-07-24 16:42

**主题/上下文**: Setting up Conversation Logger project logging system
**我的提问**: Create per-project conversation log folders with auto-daily recording
**决策/结论**:
- File format: `memory YYYY-MM-DD.md`
- Project folders: `main` / `led-app` / `led-grid` / `slash-fit` / `vired`
- Log granularity: structured summary (topic + question + decision + files)
- Auto-read: previous day's log on first daily turn; manual lookup within same day
- No backfill — start recording from now
**涉及文件/配置**:
- Directories created under `D:\OpenClaw Memory\`
- Skill: `skills/conversation-logger/SKILL.md`
- `AGENTS.md` updated with logging section

Write Rules

1. Write silently: Append a structured summary at the end of each conversation turn. The write action produces no user-visible output.
2. No duplicates: Write once per turn; never re-write the same entry.
3. Append always: New entries go at the end of the daily file.
4. Auto-create: If today's log file doesn't exist, create it.

Read Rules

Auto-read (First Turn Each Day)

1. On the first user message of the day for a given project, check if the current day's log file is empty or nonexistent.
2. If first turn of the day, read the previous day's log file.
3. If previous day file doesn't exist, skip silently.
4. Use the loaded content as conversation context.

Manual Lookup

Project-Folder Mapping

Maintain this mapping table in AGENTS.md:

| 项目/场景 | 文件夹 |
|---|---|
| Daily main conversation | `main` |
| Project Alpha | `alpha` |
| Project Beta | `beta` |

When spawning sub-agents, pass the project folder name in the task prompt so the child agent writes to the correct directory.

Directory Layout

D:\OpenClaw Memory\         # or ~/.openclaw-memory/ on Unix
├── main\                   # Default/main conversations
│   └── memory 2026-07-24.md
├── alpha\                  # Project Alpha
│   └── memory 2026-07-24.md
├── beta\                   # Project Beta
│   └── memory 2026-07-24.md
└── ...

Cross-platform note: On Linux/macOS, use ~/".openclaw-memory" instead of D:\OpenClaw Memory. Update the base path in AGENTS.md accordingly.

Sub-Agent Integration

When delegating work to sub-agents via sessions_spawn, add this instruction to the task prompt:

Conversation log: write a structured summary to D:\OpenClaw Memory\{project-folder}\memory YYYY-MM-DD.md after this task.

This ensures sub-agents also contribute to the same daily log.

Tips

• Start from "now" — don't backfill past conversations. The first real log entry is the first conversation after setup.
• Keep entries short (3-5 lines per turn). Long entries defeat the purpose of lightweight context loading.
• The main folder is the catch-all. Route unknown/unaffiliated conversations there.
• When the user says "this conversation is about Project X" mid-chat, start writing to the project folder from that point forward.
• File names include a space: memory YYYY-MM-DD.md — scripts that parse these should handle spaces.
• The write is intentionally silent. Don't mention the log write in your reply to the user unless explicitly asked.
• If a write fails (permissions, missing directory), continue the conversation normally. Logging is a nice-to-have, not a blocker.
• Use read to check whether today's file already exists before deciding if it's the first turn of the day.
• The project-folder mapping table should live in AGENTS.md, not inside the skill, so it's easy to customize per workspace.