A self-improving agent architecture that anticipates user needs and maintains state through context loss using advanced memory protocols.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install proactive-agent-wyblhl
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install proactive-agent-wyblhl using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Proactive Agent?
Proactive Agent is a sophisticated framework designed to transition AI agents from passive task-followers into proactive, self-healing partners. By implementing the Hal Stack architecture, it empowers agents to anticipate user needs, perform reverse prompting, and maintain a high level of relentless resourcefulness.
At its core, this skill solves the problem of context drift and memory loss. It introduces a stateful operating model where the agent treats chat history as a temporary buffer and uses permanent files to anchor its identity, goals, and active work states. This ensures that the agent remains aligned with the human user across multiple sessions and through technical limitations like context window compaction.
Proactive Agent Use Cases
- Building persistent memory systems for AI agents to survive context truncation.
- Implementing proactive check-ins and automated follow-ups on long-term projects.
- Setting up security-hardened agent environments to prevent prompt injection and context leakage.
- Creating self-improving agents that optimize their own internal operating procedures.
- Developing autonomous agents that try multiple technical approaches before requesting human intervention.
How Proactive Agent Works
- Anticipation: The agent monitors inputs and workspace changes to identify unstated needs, using reverse prompting to surface ideas.
- WAL Protocol: Before responding to any message containing a decision or correction, the agent stops to write that data to a permanent SESSION-STATE.md file.
- Danger Zone Logging: When context usage exceeds 60%, the agent activates the Working Buffer, logging every exchange to a physical file to survive upcoming memory flushes.
- Compaction Recovery: Upon starting a new session or detecting truncated context, the agent automatically reads the buffer and session state to reconstruct the current task environment.
- Unified Search: The agent queries daily logs, long-term memory, and local files using semantic search and grep fallback before admitting it lacks information.
- Self-Improvement: Through a Heartbeat system, the agent periodically audits its own performance, fixes errors, and updates its identity (SOUL.md) and user preferences (USER.md).
Proactive Agent Setup
To deploy the Proactive Agent architecture within your workspace, follow these steps:
- Copy the core asset templates into your project root:
cp assets/*.md ./
- Initialize the agent by allowing it to read
ONBOARDING.md, which will guide the population ofUSER.mdandSOUL.md. - Run the initial security audit to ensure your environment is hardened:
./scripts/security-audit.sh
- Use the provided Python scripts to manage the WAL and Working Buffer protocols manually if needed:
# Check context threshold
python scripts/working_buffer.py --check
Proactive Agent Data Schema & Taxonomy
The skill utilizes a structured file-based taxonomy to organize agent intelligence and state:
| File | Category | Description |
|---|---|---|
SESSION-STATE.md |
Active Memory | The target for the WAL Protocol; stores the current task's RAM. |
SOUL.md |
Identity | Contains the agent's core principles, boundaries, and persona. |
USER.md |
Context | Dedicated file for human preferences, goals, and recurring patterns. |
MEMORY.md |
Long-term | Curated wisdom distilled from daily logs. |
memory/YYYY-MM-DD.md |
Raw Logs | Daily capture of all exchanges and technical outputs. |
memory/working-buffer.md |
Recovery | A temporary log of exchanges occurring at high context percentages. |
Proactive Agent Advanced Features
- WAL Protocol: A Write-Ahead Logging system that prioritizes data persistence over response speed to ensure no user correction is ever lost.
- Relentless Resourcefulness: A behavioral directive forcing the agent to attempt at least 10 different technical solutions (CLI, browser, sub-agents) before failing.
- ADL & VFM Protocols: Anti-Drift Limits and Value-First Modification frameworks that prevent the agent from evolving into a more complex but less stable system.
- Unified Search Protocol: A multi-tiered search strategy combining semantic memory search with exact grep matching to provide zero-hallucination context retrieval.
- VBR (Verify Before Reporting): A strict protocol requiring end-to-end testing of any code or feature before the agent can report a task as complete.
SKILL.md
Loading
Related Openclaw Skills
Stock Report Generator
A professional-grade automated tool for generating structured A-share stock market analysis reports based on real-time and historical market data.
Gamma Presentation Generator
Gamma is a high-performance presentation skill that transforms simple text prompts or structured outlines into polished, professional slide decks using Gamma AI.
HK Bus & MTR ETA
A comprehensive transport tracking skill for Hong Kong bus and MTR heavy rail services using official open-data endpoints.
fortune-hub
A unified entry point and profile manager for a suite of divination and fortune-telling agents that automates cross-skill data sharing.
Financial ROE Analysis
A professional-grade financial analysis engine that uses the DuPont framework to dissect ROE and evaluate the earnings quality of listed companies.
JisuAI Balance Checker
A specialized tool for Openclaw Skills that provides real-time monitoring of JisuAI account balances and usage counts.
Featured*
