A diagnostic utility that identifies prompt context bloat and generates a safe, reversible plan to reduce token costs and improve response speed.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install context-clean-up
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 context-clean-up 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 Context Clean Up?
Context Clean Up is a specialized diagnostic utility designed to maintain the performance and cost-efficiency of AI agent sessions. Over time, prompt contexts can become bloated with oversized tool outputs, automation noise, or massive bootstrap documents, leading to slower response times and increased API expenses. This tool provides a safe, audit-only mechanism to identify these offenders within the ecosystem of Openclaw Skills, ensuring your agents remain lean and focused.
Instead of making destructive changes, this skill generates a comprehensive report that ranks risk factors and proposes a reversible optimization plan. It follows a strict safety model that avoids unattended configuration edits or silent pruning, giving developers full control over their agent environment while maintaining high-signal interactions.
Context Clean Up Use Cases
- Prompt context is bloating, causing slow agent replies or rising API token costs.
- Automation transcripts are filled with repetitive noise or unnecessary heartbeat messages.
- Memory files like AGENTS.md or MEMORY.md have grown too large for efficient processing.
- You need a ranked list of context offenders before applying manual optimizations to your Openclaw Skills setup.
How Context Clean Up Works
- Determine the scope by identifying the workspace and state directory (e.g., ~/.openclaw).
- Execute the bundled Python audit script to analyze session history and system state.
- Analyze the generated report to identify high-signal offenders like tool result dumps or bootstrap reinjection bloat.
- Review the multi-phase fix plan which ranks actions from low-risk noise discipline to higher-risk configuration narrowing.
- Implement manual changes based on the proposed plan and verify the results in a fresh session to confirm reduced context growth.
Context Clean Up Setup
To begin auditing your Openclaw Skills environment, ensure python3 is installed and run the following command to generate an audit report:
python3 scripts/context_cleanup_audit.py --workspace . --state-dir ~/.openclaw --out context-cleanup-audit.json
For Windows environments:
py -3 scripts/context_cleanup_audit.py --out context-cleanup-audit.json
Context Clean Up Data Schema & Taxonomy
The audit focuses on authoritative high-signal fields to measure bloat precisely. Data is organized by offender class and impact.
| Field | Description |
|---|---|
| eligible skills | Count of active skills in the prompt surface |
| skills.promptChars | Character count contributed by skill definitions |
| projectContextChars | Size of the project-specific context files |
| promptTokens | Total token usage for the current prompt session |
Offenders are categorized into classes: Tool Result Dumps, Automation Transcript Noise, Bootstrap Reinjection Bloat, Ambient Specialist Surface, and Summary Accretion.
Context Clean Up Advanced Features
- Audit-only execution ensures no automatic deletions or unattended config edits occur without review.
- Reversible fix plans include exact change proposals, expected impact, and detailed rollback notes.
- Multi-phase trim ladder allows for incremental optimization starting with lowest-risk noise discipline.
- Authority-based measurement prioritizes fresh-session JSON receipts over subjective claims of context reduction in Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Skill Hub
A comprehensive toolset for discovering, security vetting, and managing Openclaw Skills from multiple registries and catalogs.
OpenClaw Whisperer
A comprehensive diagnostic, error-fixing, and skill recommendation toolkit designed to streamline Openclaw Skills management and troubleshooting.
EvoLink Cheapest Image Generation
A high-performance, cost-effective tool for generating AI images directly through your AI agent using the EvoLink API.
EvoLink Best Image
A professional-grade image generation and editing skill powered by the EvoLink Nano Banana Pro API for high-resolution visual creation.
Context Scope Tags
A portable tagging protocol that defines explicit context boundaries and memory persistence rules for AI chat sessions.
Cron Worker Guardrails
A reliability-first checklist for hardening POSIX-based OpenClaw background workers and unattended automation against shell-specific failures.
METADATA
Featured*
