Session Digest

自动总结当天对话到 memory/YYYY-MM-DD.md。cron 23:00 自动运行，提取对话让 agent 自己总结。

Security Scan

The skill does what it says (extracts session messages for daily summarization) but it reads session files under ~/.openclaw/agents (a config path not declared), writes plaintext to /tmp, and has some modest privacy/scope concerns that you should review before installing.

Purpose & Capability

The script's behavior (reading agent session .jsonl files and producing a /tmp text file) matches the skill's stated purpose of extracting conversations for daily summaries. There are no unexpected external services or binaries required.

Instruction Scope

SKILL.md instructs running scripts/extract.js and having the agent read the produced /tmp file and summarize it — this is within scope. However the script reads all live agent sessions (main, claude, gemini, etc.) and will include any message content present in those session files; the provided simple sanitization (regex removals) may not remove all sensitive or system-level content. The skill also documents a cron trigger but provides no install steps to set that up.

Install Mechanism

No install spec is present (instruction-only plus a small script). Nothing is downloaded or written to system locations beyond the script writing a temporary output file; low install risk.

Credentials

Registry metadata declared no required config paths or credentials, yet the script reads from ~/.openclaw/agents (or OPENCLAW_DIR) — i.e., it accesses user-local agent session files. The skill does not request credentials, but it implicitly requires read access to that config/data path; this mismatch should have been declared. Also the output is written to /tmp which may be world-readable on some systems.

Persistence & Privilege

The skill is not always-enabled and does not modify other skills or system config. It only writes a temporary file and logs; no elevated persistence or privilege escalation is requested.

What to consider before installing

This skill appears to do what it says: it scans your local OpenClaw agent session files and writes a plain-text /tmp/session-digest-YYYY-MM-DD.txt for the agent to summarize. Before installing or enabling it, consider: 1) The script reads ~/.openclaw/agents/*/sessions/*.jsonl (it did not declare this config path in metadata) — confirm you are comfortable granting it read access to any session data stored there. 2) The temporary output is plaintext in /tmp and could contain sensitive system messages or secrets that the basic regex cleaning may not remove; inspect outputs before having the agent post them into memory. 3) If you want tighter scope, modify the script to limit which agents/sessions are scanned or add stronger redaction, and set up cron manually only after testing. 4) Because metadata omitted the config-path dependency, prefer to run the script manually once and review results before allowing automated or scheduled runs.

Session Digest - 每日对话总结

自动总结当天所有 session 对话，生成精简的每日记忆。

工作方式

1. cron 23:00 触发
2. 运行 extract.js 提取当天对话到临时文件
3. agent 读临时文件，自己总结
4. 写入 memory/YYYY-MM-DD.md

手动使用

# 提取对话
node ~/.openclaw/workspace/skills/session-digest/scripts/extract.js [YYYY-MM-DD]

# 然后让 agent 读 /tmp/session-digest-YYYY-MM-DD.txt 并总结

输出格式

# YYYY-MM-DD Weekday

### 做成了什么
- xxx

### 改了什么
- xxx

### 学到了什么
- xxx

### 待办
- [ ] xxx

---
N 个 session，M 条消息

隐私 & 安全

• 不调外部 API：extract.js 只读本地文件，不联网
• 读取所有 agents：main、claude、gemini 等所有存活的 session
• 数据不离开本地：所有数据都在 ~/.openclaw/ 目录内

文件

• scripts/extract.js - 提取对话到临时文件
• SKILL.md - 本文件