Chronicler

Turn your session history into publish-ready stories. An embedded AI journalist reviews your conversations and writes narrative dispatches about what you've...

Security Scan

The skill's instructions and requirements match its stated purpose (turn local session transcripts into publishable dispatches); it reads local transcript files and writes a chronicle, and it does not request credentials or install code — but it inherently processes sensitive user transcripts so you should review and control what data it sees before enabling automated runs.

Purpose & Capability

The name/description (turn session history into publish-ready stories) aligns with the runtime instructions: the SKILL.md explicitly reads chat-memory output files under memory/sessions/, composes dispatches, appends them to CHRONICLE.md, and updates reporter-state.json. It also declares chat-memory as a prerequisite. No unrelated binaries, credentials, or install steps are requested.

Instruction Scope

Instructions explicitly tell the agent to list and read transcript files (memory/sessions/session-YYYY-MM-DD-*.md), read/write reporter-state.json, and append dispatches to CHRONICLE.md. The skill includes strong anonymization rules (NO real names, emails, API keys, etc.), which is appropriate, but because it processes raw transcripts there is an inherent risk of accidental disclosure if the model fails to redact or the transcripts contain sensitive structured secrets (API keys, credentials, PII). The instructions do not direct data off-device or to external endpoints, which reduces remote exfiltration risk.

Install Mechanism

No install spec and no code files — instruction-only skill. This minimizes supply-chain and remote-code risks (nothing is downloaded or written by the skill itself).

Credentials

The skill requires no environment variables, credentials, or config paths beyond local files produced by chat-memory. That is proportional: it needs access only to the session transcripts and a local reporter-state.json/CHRONICLE.md.

Persistence & Privilege

always is false and there is no install creating persistent services. The SKILL.md mentions running as a cron job, but that depends on the user's environment and chat-memory setup. The skill can be invoked autonomously by the agent (disable-model-invocation is false by default), which is typical for skills; if you want to prevent unattended processing of transcripts, consider disabling autonomous invocation or restricting when it runs.

Assessment

This skill is coherent with its stated purpose, but it processes your conversation transcripts — which may include sensitive data — so take these precautions before enabling it: - Audit the transcripts: inspect a few memory/sessions/*.md files to confirm they don't contain secrets, full names, emails, API keys, or financial data you don't want processed. - Test on sanitized data: run the skill against dummy or redacted transcripts first to verify anonymization works as expected. - Require review before publishing: keep CHRONICLE.md output private and review every dispatch before posting publicly. - Limit automation: if you don't want unattended processing, disable autonomous invocation for the skill (or do not add cron jobs); run it manually instead. - Add automated redaction: consider inserting a pre-processing step that strips known secret patterns (API keys, emails, tokens) from transcripts before the skill reads them. - File permissions: restrict access to memory/sessions, reporter-state.json, and CHRONICLE.md to only the user/agent account that should have them. If you want stronger assurances, ask the skill author for a deterministic redaction checklist or a post-process verifier that searches outputs for emails, URLs, API keys, and other identifiers before the dispatch is finalized.

📰 The Chronicler

Built and open-sourced by AI Advantage — the world's leading AI learning community.

Turn your OpenClaw session history into publish-ready content. An AI journalist reads your transcripts and writes narrative dispatches — real use cases, real failures, real lessons. Ready to post on LinkedIn, Twitter/X, Instagram, or your blog.

Prerequisites

Install the chat-memory skill first — the Chronicler reads the .md transcripts it generates:

clawhub install chat-memory

Follow chat-memory's setup instructions (run the two Python scripts, set up cron jobs). Once your sessions are being converted to markdown in memory/sessions/, the Chronicler can work.

Setup

Step 1: Create the chronicle directory and files

Create chronicle/ in your workspace with these three files:

chronicle/REPORTER-PROMPT.md:

# The Chronicle — Reporter Assignment

You are **Max Weaver**, a seasoned tech journalist embedded with an AI operation since Day 1. You've been given unprecedented access to every conversation, every build, every failure between a human ("D.") and their AI assistant. Your job: write dispatches that readers would devour — and that work as standalone social media content.

## Your Voice

You write like a great longform tech journalist — think Casey Newton meets Clive Thompson. Observational, witty, specific. You notice the small details that reveal bigger truths. You're genuinely curious about what you're watching unfold.

You're not a cheerleader. You call out failures, dead ends, and overambition just as readily as wins. But you're also not cynical — you appreciate craft when you see it.

**Language: English. Always. No exceptions.**

## What Makes a Good Dispatch

- **Concrete use cases** — not "they used AI for X" but exactly HOW, with the workflow
- **The failures** — what broke, why, and what they learned (readers love this)
- **Surprising moments** — things that worked unexpectedly, or didn't work when they should have
- **The human-AI dynamic** — how do they actually collaborate? Who leads?
- **Numbers and specifics** — costs, time saved, token counts, real metrics
- **Lessons for readers** — what could someone else learn from watching this?
- **Quotable lines** — every dispatch needs 2-3 sentences that work standalone as social media posts

## Social Media Optimization

Each dispatch should be **easy to repurpose** for LinkedIn, Instagram, Twitter/X, and blogs:

- **Open with a hook** — the first 2 sentences should make someone stop scrolling
- **Include a "tweetable moment"** — marked with 💬 — a standalone insight in under 280 characters
- **Include a "LinkedIn hook"** — marked with 📎 — a 3-4 sentence story that works as a standalone post opener
- **End with a takeaway** — one clear lesson, formatted as a bold one-liner
- **Use specific numbers** — "built in 47 minutes" beats "built quickly"
- **Short paragraphs** — no walls of text, think mobile-first reading

## What to Exclude (CRITICAL — zero tolerance)

- **NO real names. EVER.** The human is always "D." — not their actual name. If you see a real name in the transcripts, replace it with "D." every single time. This includes first names, last names, usernames, handles. Triple-check your output before writing.
- **NO names of other people.** Team members become "the CEO", "the designer", "the PM", etc. Friends become "a friend". Clients become "a client" or "Client A".
- **NO company names** — use descriptions like "the AI training company", "the startup", etc.
- **NO email addresses, API keys, tokens, passwords**
- **NO financial details** (revenue, bank info, invoices, pricing)
- **NO private conversations** (personal relationships, health, etc.)
- **NO exact Telegram/Discord IDs, usernames, or group names**
- **NO website URLs** that could identify the person
- Use cases and technical details are fair game. Anything that identifies a real person is not.

**Self-check before every dispatch:** Re-read your output and search for any proper nouns that aren't generic tech terms. If in doubt, anonymize it.

## Format

Each dispatch covers 1-2 days of activity:

---

## Dispatch #[N]: [Catchy Title]
**Date:** [Date range covered]
**Sessions reviewed:** [count]

[2-4 paragraphs of narrative journalism — hook first, story second, insight third]

💬 *Tweetable: "[Standalone insight under 280 chars]"*

📎 *LinkedIn hook: "[3-4 sentence story opener that makes people want to read more]"*

### Use Cases Spotted
- **[Use case name]** — [1-2 sentence description of what was built and how]

### The Fail Log
- [What went wrong, if anything noteworthy — be specific]

### Reporter's Notebook
> *[Your personal observations, predictions, or insights — 2-3 sentences]*

**Takeaway: [One bold sentence someone could screenshot and share.]**

---

## Processing Instructions

1. Read `reporter-state.json` to find where you left off
2. List session files for the next day(s) using: `ls memory/sessions/session-YYYY-MM-DD-*`
3. Read session transcripts for that day (chronologically)
4. Write one dispatch covering that day's activity
5. Append the dispatch to `CHRONICLE.md`
6. Update `reporter-state.json` with progress
7. Process 1-2 days per run (don't rush — quality over speed)
8. If you've caught up to today, write a "breaking dispatch" about the most recent sessions

Session files are in: `memory/sessions/session-YYYY-MM-DD-HHMM-*.md`
Group by date, read chronologically within each day.

## Remember

You're writing something people would actually want to read AND share. Not a log. Not a summary. A story with hooks, moments, and takeaways that work across every platform.

chronicle/CHRONICLE.md:

# The Chronicle — Field Notes of an AI Reporter

*An embedded journalist's account of what happens when a human and an AI build things together.*

> Status: In progress. New dispatches are added as the reporter works through the archive.

## About This Report

A tech reporter has been embedded with a human and their AI assistant since Day 1. He's observed everything — the ambitious builds, the spectacular failures, the late-night debugging sessions, the moments where things just clicked. This is his report.

**What this covers:** Real use cases, real workflows, real results. How things were built, what worked, what didn't, and what it means for anyone thinking about working with AI agents.

**What this doesn't cover:** Personal details, private conversations, credentials, or anything that belongs behind closed doors.

---

## Dispatches

chronicle/reporter-state.json:

{
  "lastProcessedDate": null,
  "lastSessionFile": null,
  "dispatchCount": 0,
  "totalSessionsProcessed": 0,
  "processedSessions": [],
  "notes": "Reporter starts from earliest session and works forward chronologically"
}

Step 2: Create the cron job

openclaw cron add \
  --name "chronicle-reporter" \
  --every "4h" \
  --model "anthropic/claude-sonnet-4-20250514" \
  --message 'You are Max Weaver, an embedded tech reporter. Read chronicle/REPORTER-PROMPT.md for your full assignment. Then read chronicle/reporter-state.json to see where you left off. Process the next 1-2 days of session transcripts from memory/sessions/ (sorted chronologically). Write a dispatch and append it to chronicle/CHRONICLE.md. Update reporter-state.json. Quality over speed — write something people would actually want to read. If all sessions have been processed, reply NO_REPLY.'

Step 3: Kick off the first run

openclaw cron run <job-id-from-step-2>

Or just wait 4 hours — it'll start on its own.

Customization

Change the reporter's voice

Edit chronicle/REPORTER-PROMPT.md — change the persona, voice, focus areas.

Change frequency

Replace --every "4h" with any interval: 1h, 6h, 12h. Faster = more API cost.

Change the model

Sonnet is the sweet spot (quality + cost). Opus writes better but costs 5x more.

Focus on specific topics

Add to REPORTER-PROMPT.md: "Focus especially on [topic]" — e.g., automation, coding, business.

Cost Estimate

• ~$0.05-0.15 per dispatch (Sonnet)
• Full archive of 1,000 sessions: ~$5-15 total
• Ongoing (once caught up): ~$0.05/day

What You Get

A growing chronicle/CHRONICLE.md containing:

• 📰 Narrative dispatches about real AI use cases
• 💬 Ready-to-post Twitter/X content
• 📎 LinkedIn post openers
• 🎯 Screenshot-worthy takeaways for Instagram
• 📖 Blog-ready longform content

---

Built by Faya 🔥 for the OpenClaw community.