| name | proactive-agent |
| version | 1.0.0 |
| description | Converts passive agents into autonomous, self-improving partners. Use when building long-lived agent systems that need initiative, self-correction, and autonomous operation. Covers write-ahead logging, working buffer management, compaction recovery, reverse prompting, proactive check-ins, autonomous cron scheduling, and ADL/VFM guardrails. Enables agents to anticipate needs rather than waiting for instructions. |
| source | https://lobehub.com/skills/openclaw-skills-proactive-agent (4.92/5 rating) |
| changelog | Initial release — autonomous agent architecture patterns |
Proactive Agent
Convert passive agents into autonomous, self-improving partners that anticipate needs and take initiative.
Core Architecture
Write-Ahead Logging (WAL)
All agent actions are logged before execution to prevent data loss:
- Every decision, action, and outcome recorded in sequence
- Enables crash recovery and state reconstruction
- Provides audit trail for debugging autonomous behavior
Working Buffer
Short-term memory management for active tasks:
- Current context and goals held in buffer
- Automatic compaction when buffer exceeds threshold
- Priority-based eviction for least-relevant items
Compaction Recovery
When context is compressed or lost:
- WAL entries reconstruct recent state
- Working buffer repopulated from log
- No task continuity is lost during compaction
Proactive Behaviors
Reverse Prompting
Instead of waiting for user input, the agent asks itself:
- What would the user likely ask next based on current work?
- What blockers exist that the user hasn't noticed?
- What can be prepared in advance to save time?
Proactive Check-Ins
Scheduled self-assessment triggers:
- Review task progress against goals
- Identify stalled or blocked items
- Surface issues before the user asks
Autonomous Cron Scheduling
Recurring tasks managed without user intervention:
- Health checks, dependency updates, status reports
- Scheduled at appropriate intervals
- Results logged and surfaced on next interaction
Guardrails
ADL (Action Definition Language)
Every autonomous action must:
- Have a clear purpose linked to user goals
- Be reversible or low-risk
- Include a rollback plan
- Stay within defined permission boundaries
VFM (Value Verification Metric)
Before executing autonomously, verify:
- Necessity: Does this need to happen now?
- Value: Does this create measurable value?
- Frequency: Is this the right cadence?
- Scope: Is this within the agent's authority?
Activation Triggers
- Building long-running agent systems
- Implementing autonomous workflows
- Creating self-monitoring agent processes
- Designing agent architectures that anticipate needs