context-window-management
Context Window Management
State file: ~/.openclaw/skill-state/context-window-management/state.yaml
Warning Signs
- Session running for many hours
- Responses getting slower or less coherent
- Carrying context about completed tasks no longer relevant
- Large code blocks in context not being actively used
- Check
last_compacted_atin state — if compacted recently this session, prefer a lighter strategy
Reduction Strategies
Summarize Completed Work
Replace detailed context about finished tasks with short summaries.
After: write state strategy_used: summarize, current_task, last_compacted_at.
Externalize Reference Material
Move large docs/schemas to files. Reference them by filename rather than including in context.
After: write state strategy_used: externalize, current_task, last_compacted_at.
Clean Session Restart
Write comprehensive handoff document, start fresh session, load only what's needed.
After: write state strategy_used: clean_restart, handoff_path, current_task, last_compacted_at.
What to Keep in Context
- Current task objective
- Current step being executed
- Relevant code being modified (just the parts in scope)
- Recent error messages if debugging
What to Offload
- Completed task details (summarize to memory)
- Full file contents of files not currently being edited
- Conversation history from different topics hours ago
- Reference docs that can be re-read on demand
More from archieindian/openclaw-superpowers
heartbeat-governor
Enforces per-skill execution budgets for scheduled cron skills — pauses runaway skills that exceed their token or wall-clock budget before they drain your monthly API allowance.
25fact-check-before-trust
Triggers a secondary verification pass for any agent output containing factual claims, numbers, dates, or named entities before the output is acted on
19long-running-task-management
Breaks multi-hour tasks into checkpointed stages with resume capability. Use when a task is expected to take more than 30 minutes or multiple sessions.
19agent-self-recovery
Detects when the agent is stuck in a loop and escapes systematically. Use when you notice repeated failures or loss of direction.
18skill-vetting
Reviews a ClawHub skill's source code for security risks before installation. Use before installing any new skill.
18persistent-memory-hygiene
Keeps OpenClaw's memory store clean, structured, and useful. Use at session end and during periodic maintenance.
18