A manual context management tool that allows developers to compact AI memory at logical workflow boundaries rather than relying on arbitrary auto-compaction.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install strategic-compact
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 strategic-compact 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 Strategic Compact?
The Strategic Compact skill is designed to give developers using Openclaw Skills precise control over the context window of their AI agents. While most agents use auto-compaction that triggers at arbitrary token counts, this skill allows for manual intervention at natural transition points, such as after completing a research phase or before starting a major implementation.
By utilizing this skill, you can prevent the AI from losing critical mid-task reasoning. It encourages a structured workflow where important findings are saved to persistent files before the conversation history is cleared. This ensures that the agent remains performant during long-running sessions without sacrificing the continuity of the project state.
Strategic Compact Use Cases
- Moving from a research and planning phase to active code implementation.
- Switching between two unrelated tasks or bug fixes within the same session.
- Managing high token usage (200K+) to maintain AI responsiveness and accuracy.
- Clearing out failed reasoning or dead-end debugging paths to start fresh with a new strategy.
- Documenting a major milestone before transitioning to a new developmental sprint.
How Strategic Compact Works
- The user identifies a logical transition point in their workflow, such as the completion of a plan.
- Important reasoning and data are saved into persistent files (like CLAUDE.md) or committed to Git.
- The user triggers the compaction manually using a specific command.
- The agent clears the ephemeral conversation history while retaining files on disk and project instructions.
- The session continues with a fresh context window, referencing the saved files for ongoing project continuity.
Strategic Compact Setup
Strategic Compact is integrated directly into the agent workflow. No additional binary installations are required for users already utilizing Openclaw Skills compatible environments.
# To trigger a strategic compaction with a context message
/compact "Research complete. Moving to implementation of API endpoints."
Strategic Compact Data Schema & Taxonomy
The skill differentiates between ephemeral conversation data and persistent project data to ensure safety during compaction:
| Data Type | Persistence | Description |
|---|---|---|
| Conversation History | Cleared | Previous messages and intermediate AI reasoning are removed |
| Files on Disk | Persistent | All source code, documentation, and saved logs remain untouched |
| Git State | Persistent | Current branch, commits, and staged changes survive the reset |
| Project Rules | Persistent | Core instructions and .md files used for project context remain |
| Memory Files | Persistent | Explicitly created 'memory' or 'context' files survive to bridge phases |
Strategic Compact Advanced Features
- Token Awareness Thresholds: Guidelines for compacting at 150K (consider), 180K (plan), and 200K (execute) tokens.
- Contextual Resumption: The ability to pass a message during compaction to immediately re-prime the agent's focus.
- Multi-phase Persistence: Works in tandem with memory-to-disk strategies to ensure long-term reasoning survives across multiple compactions.
- Checkpoint Discipline: Encourages a Git-centric workflow where commits act as the permanent record of progress before memory clearing.
SKILL.md
Loading
Related Openclaw Skills
LoRA Toolkit
A specialized toolkit for configuring, estimating, and generating LoRA fine-tuning scripts for large language models.
Skill Vetter
Skill Vetter is a dedicated security analysis tool that scans AI agent skills for dangerous commands, network exfiltration, and unauthorized file access before installation.
Fantasy Map Generator
An AI-powered cartography tool that transforms text prompts into professional fantasy maps for RPG campaigns and game development.
Yang Yuanqing Strategic Thinking Perspective v2.0
A high-level strategic mental model designed to simulate the decision-making framework of Lenovo's CEO, focusing on global integration and operational efficiency.
Deep Research Agent
A professional-grade research skill that searches dozens of sources, fetches deep content, and synthesizes evidence-based reports with full citations.
Apple Music Play (Clawtunes)
A macOS skill for controlling Apple Music playback of both library and streaming catalog tracks using command-line tools.
Featured*
