SafePaste is a local security and compatibility layer that analyzes third-party prompts and configurations before they are applied to your OpenClaw environment.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install safepaste
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 safepaste 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 SafePaste?
SafePaste acts as a defensive barrier for your AI agent workspace, preventing the creation of conflicting or incoherent configurations known as Frankenclaws. It automatically intercepts pasted content—such as AGENTS.md rules, SOUL.md persona shifts, or skill installation commands—and compares them against your actual local setup to identify contradictions, redundancies, or security risks. By utilizing these Openclaw Skills, developers can confidently experiment with community-shared tips while ensuring their agent core logic remains stable and secure.
The skill is designed for 100% local execution, meaning no data leaves your machine during the analysis. It provides a sophisticated reasoning layer that doesn't just check for syntax, but evaluates the semantic impact of changes on your agent's behavior, cost efficiency, and security posture. This makes it an indispensable tool for anyone building complex agentic workflows.
SafePaste Use Cases
- Testing viral prompts from social media or community forums without breaking your existing agent behavior.
- Merging new persona traits into SOUL.md while preserving your agent's established voice and values.
- Auditing third-party tool configurations and skill installs for hidden API costs or security vulnerabilities.
- Preventing redundant rule additions that bloat the context window and reduce performance.
- Safely rolling back workspace changes if a new configuration produces undesirable results.
How SafePaste Works
- The skill monitors incoming messages for signals suggesting a prompt or configuration tip is being shared.
- When detected, or when manually triggered, SafePaste escalates the analysis to a high-reasoning model tier to ensure deep logic checking.
- The agent reads your local workspace files, including AGENTS.md, SOUL.md, and openclaw.json, to understand your current state.
- SafePaste performs a multi-point analysis checking for conflicts, behavior drift, permission expansions, and security injection patterns.
- A detailed report is generated with a quick verdict, specific line-by-line conflict data, and suggested modifications.
- Upon confirmation, the skill creates a timestamped backup of your workspace before applying the changes.
- Users can invoke a rollback command at any time to restore the workspace to its exact state prior to the application.
SafePaste Setup
To install SafePaste, run the following command in your terminal:
clawhub install safepaste
Once installed, the skill will automatically offer to check content it recognizes as OpenClaw configuration. You can also manually trigger an analysis by prefixing your paste with "SafePaste this:" or "Check this before I add it:".
SafePaste Data Schema & Taxonomy
SafePaste organizes its local state and backups within the standard OpenClaw directory structure to ensure easy management and privacy.
| Path | Purpose |
|---|---|
~/.openclaw/safepaste-state.json |
Stores local usage statistics and internal state. |
~/.openclaw/safepaste-history.json |
(Optional) Maintains a log of analyzed items and verdicts for power users. |
~/.openclaw/safepaste-backups/ |
Directory containing timestamped full-workspace snapshots for rollback support. |
~/.openclaw/workspace/ |
The target directory where safe modifications are applied. |
SafePaste Advanced Features
- Intelligent Model Escalation: Automatically switches to higher-tier models for complex analysis to ensure high-fidelity conflict detection.
- Smart Merging logic: Specifically tuned to handle SOUL.md files, preserving the agent's unique voice while incorporating new behavioral traits.
- Batch Processing: Summarizes large sets of configuration items (10+) into categorized reports for efficient cherry-picking.
- Security Injection Detection: Flags patterns like "Ignore previous instructions" or unauthorized data exfiltration attempts.
- Automated Rollback System: Provides a non-destructive path to experiment with new Openclaw Skills and configurations.
SKILL.md
Loading
Related Openclaw Skills
Clawtar
A demonstration skill designed for AI agents to practice detecting, settling, and retrying Cashu-gated HTTP 402 payment flows.
Cashu Emoji
A specialized tool for embedding and extracting Cashu tokens or hidden messages within standard emojis using Unicode variation selectors.
agent-self-care
A specialized maintenance skill that automates health monitoring, process cleanup, and performance optimization for autonomous agents.
Zoho Mail CLI
A robust command-line interface for Zoho Mail that outputs machine-readable JSON for seamless automation and agent workflows.
WeChat Pack
WeChat Pack is a specialized utility for converting Markdown or Word documents into perfectly formatted HTML and organized assets for WeChat Official Accounts.
Studio Service Catalog
An automated tool to fetch and display studio service details, pricing, and durations via a local script execution.
Featured*
