Lesson 15 — OpenClaw Proactive Agent: Upgrading AI from "Waiting for Instructions" to "Proactively Reaching Out" (2026)

Goal: Install the Proactive Agent Skill so OpenClaw transforms from a reactive Q&A tool into a proactive AI that anticipates needs and executes scheduled tasks automatically.

Passive AI vs. Proactive AI: What's the Difference?

Most AI assistants are passive — if you don't ask, they don't say anything. Proactive Agent solves this fundamental problem.

This is ClawHub's #4 most downloaded Skill (126k downloads). Core value: transforms AI from a tool into "an assistant that proactively contacts you."

What Is the WAL Protocol?

WAL (Working Action List) protocol is the core mechanism of Proactive Agent, and it's the most commonly asked-about concept.

WAL is a persistent "pending action list," structured like this:

## WAL — Current Action Queue
 
- [ ] Push daily tech briefing to Telegram every day at 08:00
- [ ] Monitor GitHub repo star count, push notification when it exceeds 1000
- [ ] Generate this week's work summary every Friday at 17:00

The Skill periodically checks this list in the background. When a trigger condition is met, it automatically executes the corresponding task, marks the entry as [x] when done, and records the timestamp. The WAL file is saved locally and persists across restarts.

Step 1: Install Proactive Agent Skill

In WebChat or Telegram, send:

/install @halthelobster/proactive-agent

Verify installation:

pnpm openclaw skills list
# proactive-agent should appear in the list

After installation, the Skill creates these files in your workspace:

~/.openclaw/workspace/skills/proactive-agent/
├── SKILL.md          ← Skill main logic
├── wal.md            ← Working Action List (core config file)
└── buffer.md         ← Working Buffer (temporary scratch pad)

Step 2: Configure Time-Triggered Tasks

Open wal.md or just tell AI to add a time-triggered task:

Add a task to WAL: every morning at 8 AM, search today's tech news, compile 3 key points, and push to me

AI will automatically write to wal.md:

## WAL
 
- [ ] cron: 0 8 * * *
  task: Search today's tech news, compile 3 key points, push to Telegram
  last_run: —

Standard cron expressions are supported, for example:

Every day at 08:00       → 0 8 * * *
Every hour               → 0 * * * *
Every Monday at 09:00    → 0 9 * * 1
Every month on the 1st   → 0 9 1 * *

Step 3: Configure Event-Triggered Tasks

Proactive Agent supports not just time triggers but also event triggers.

Add an event-triggered task: when I send a message containing the word "urgent", automatically forward that message to the #urgent Telegram channel and flag it as high priority

Example event triggers:

## WAL — Event Triggers
 
- [ ] event: keyword="urgent"
  task: Forward message to #urgent channel, add 🔴 flag
 
- [ ] event: file_created="~/Downloads/*.pdf"
  task: Auto-summarize the new PDF and push summary to Telegram
 
- [ ] event: time_idle=30min
  task: Push today's to-do reminder

Step 4: Set Up Proactive Push to Telegram

First confirm you've completed the Telegram setup from Lesson 02. Then specify a push target in the task:

Configure Proactive Agent push: every day at 09:00, check my GitHub issues, and if any are unanswered for more than 24 hours, send a reminder to my Telegram private chat

You can also target a specific group or channel:

- [ ] cron: 0 9 * * 1-5
  task: Summarize yesterday's unread email subjects, push to Telegram group @myteam
  target: telegram://group/@myteam

Test whether the push is working:

Trigger a Proactive Agent push now, send "push test" to Telegram

Step 5: Using the Working Buffer

The Working Buffer (buffer.md) is Proactive Agent's temporary workspace, used for:

• Passing data between tasks (output of task A becomes input for task B)
• Saving intermediate results to avoid recomputing
• Recording task execution logs

## Buffer — 2026-03-30
 
### Tech Briefing Data
- Source: HackerNews Top 10
- Scraped at: 08:00:12
- Item count: 10
 
### Pending Push Content
1. OpenAI releases o4 model, reasoning improved by 60%
2. Rust 2026 edition released, compile speed up 30%
3. Next.js 16 stable, Server Actions fully production-ready

Tell AI how to use the Buffer:

Store each briefing scrape's raw data in the Buffer, keep the last 7 days, so I can review historical data

Proactive Agent vs. Lesson 06 Scheduled Tasks

Lesson 06's scheduled tasks trigger single script executions via the system cron, suitable for simple recurring tasks (like sending briefings).

Proactive Agent is a continuously running intelligent agent — key differences:

Simply put: Lesson 06 is an "alarm clock," while Proactive Agent is a "thinking secretary."

FAQ

What's the difference between Proactive Agent and regular OpenClaw?

Regular OpenClaw requires the user to send a command before it responds, and no state is preserved after the conversation ends. Proactive Agent runs in the background after installation, maintaining an action queue via the WAL Protocol, and automatically triggers tasks based on time or events — no manual initiation needed. The most obvious result: you can have it send you a message every morning without you having to ask.

What is the WAL Protocol and how do I configure it?

WAL (Working Action List) is a locally saved Markdown-format to-do list, where each entry contains a trigger condition (cron expression or event keyword) and a description of the task to execute. You can tell AI in plain language "add a task that runs every day at 8 AM" and AI will automatically write to ~/.openclaw/workspace/skills/proactive-agent/wal.md — no need to manually edit config files.

Will Proactive Agent continuously consume API tokens?

No. Proactive Agent uses a "polling check" mechanism — it checks the WAL list every 5 minutes by default and only calls the AI model when a trigger condition is met. In between it's in a lightweight check state that consumes almost no tokens. You can further control consumption by adjusting the polling interval — for example, changing it to every 15 minutes.

Can proactive push messages go somewhere other than Telegram?

Yes. Proactive Agent supports multiple push targets: Telegram (personal/group/channel), email (with Gmail Skill), Slack (with Slack Skill), and local file writes. Just specify the target field in the WAL task to push to any integrated channel. Telegram is the default recommendation because it's the easiest to configure and has the lowest latency.

Next Steps

• Lesson 16 — Install Multi Search Engine to let OpenClaw simultaneously query 17 search engines including Baidu