proactive-agent

The Proactive Agent skill turns standard task-following AI agents into autonomous, long-term partners capable of anticipating user needs, maintaining state across restarts, and executing background maintenance. Designed for developers and power users working within the OpenClaw ecosystem, this skill implements the Write-Ahead Logging (WAL) protocol to preserve critical context, ensuring that sessions never lose momentum due to crashes or timeouts. By leveraging a structured memory architecture including a Working Buffer for active task tracking and a curated MEMORY.md for long-term intelligence, agents can effectively manage their knowledge base over time.

• WAL Protocol: Implements a robust logging system with SESSION-STATE.md, working-buffer.md, and MEMORY.md to prevent context loss.
• Autonomous Crons: Enables context-aware automated tasks, including heartbeats for periodic system checks, daily maintenance, and weekly optimizations.
• Self-Improving Architecture: Facilitates pattern recognition to suggest relevant actions, automate repetitive file management, and compact memory logs.
• Context-Aware Automation: Detects recurring user patterns to proactively suggest follow-ups, calendar syncs, and system health updates.
• Memory Management: Supports structured compaction, daily logging, and long-term curation to keep the agent's active memory clean and efficient.

Users should integrate this skill into their workspace to manage complex workflows that require persistent state. Typical inputs involve defining cron schedules, heartbeat intervals, and memory retention policies in the provided JSON configuration. Outputs consist of automated status reports, compacted state files, and proactive task triggers. Constraints include strict file structure adherence and the need for periodic manual curation to ensure long-term memory quality. This skill is ideal for building agents that function like virtual administrative assistants, ensuring that tasks are not only executed but also tracked, refined, and prioritized without constant user intervention.