A test-driven framework to verify that AI agents correctly apply learned behaviors and memory patterns without silent degradation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install preflight-checks
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 preflight-checks 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 Pre-Flight Checks?
The Pre-Flight Checks skill provides a robust behavioral verification framework for AI agents, inspired by aviation safety protocols and automated software testing. It solves the critical problem of silent degradation, where an agent may successfully load memory files but fail to apply the specific rules or behavioral patterns defined within them. By implementing this within your suite of Openclaw Skills, you can move beyond subjective assessments and establish an objective, quantifiable measure of agent reliability.
This skill enables agents to perform self-diagnostics by running through a series of behavioral scenarios and comparing their internal logic against a set of canonical answers. This ensures that every session starts with an agent that is truly aligned with its documented identity, communication standards, and operational constraints.
Pre-Flight Checks Use Cases
- Validating agent behavior immediately after a session restart or memory clear command.
- Ensuring consistency across different agent instances or onboarding new instances.
- Detecting behavioral drift after updating core memory files or rule sets.
- Automating quality assurance for agent logic within a CI/CD pipeline.
- Performing on-demand diagnostics when an agent shows signs of inconsistent decision-making.
How Pre-Flight Checks Works
- The system utilizes a CHECKS file containing specific scenarios that require behavioral responses from the agent.
- Upon initialization, the agent reads these scenarios and generates responses based on its current loaded state.
- The agent then compares its answers against a PRE-FLIGHT-ANSWERS file which contains the expected correct behavior and known wrong answers.
- A scoring mechanism evaluates the results (e.g., 23/23), providing a clear pass/fail status.
- Based on the score, the agent either confirms it is ready for operation or identifies specific rules that require re-loading or review.
Pre-Flight Checks Setup
To integrate this framework into your environment, install the skill via the hub:
# Install the skill
clawhub install preflight-checks
# Initialize the system in your workspace
cd ~/.openclaw/workspace
./skills/preflight-checks/scripts/init.sh
This will create the necessary Markdown templates and update your AGENTS.md file to ensure that Openclaw Skills are verified at the start of every session.
Pre-Flight Checks Data Schema & Taxonomy
The skill organizes its verification data through a structured taxonomy in Markdown files:
| File Name | Content Type | Role |
|---|---|---|
PRE-FLIGHT-CHECKS.md |
Scenarios | Defines behavioral unit tests and categories (Identity, Anti-Patterns, etc.). |
PRE-FLIGHT-ANSWERS.md |
Expectations | Contains canonical answers, rationales, and common mistakes to avoid. |
AGENTS.md |
Integration | Orchestrates the automatic execution of checks during the session lifecycle. |
scripts/ |
Automation | Shell scripts for initializing, adding, and running checks. |
Pre-Flight Checks Advanced Features
- CI/CD Integration: Run behavioral unit tests automatically via GitHub Actions whenever memory files are updated.
- Multi-Profile Support: Maintain distinct behavioral check-lists for different agent roles, such as development, production, or research profiles.
- Interactive Check Builder: Use the provided CLI scripts to interactively add new behavioral scenarios based on real-world mistakes discovered during operation.
- Self-Correction Logic: Configurable thresholds that trigger automatic memory re-reads if the agent fails more than a specific number of checks.
- Living Documentation: The answers file serves as a canonical reference for intended behavior, evolving alongside your Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Video Games
A powerful multi-source gaming intelligence skill for checking deals, compatibility, and playtime via Python.
Zillow Real Estate Intelligence
A professional-grade real estate analysis skill for interpreting Zillow data, calculating investment metrics, and formulating property pricing strategies.
Zigbee
A comprehensive guide for deploying, managing, and troubleshooting high-performance Zigbee mesh networks for smart home automation.
Zendesk
A professional-grade AI agent skill for managing Zendesk tickets, user records, and automated support workflows via API integration.
WorthClip
WorthClip is an AI-driven tool that evaluates YouTube videos against your personal learning goals to provide scores, summaries, and curated feeds.
Liuren Divination
A Node.js-powered tool for performing traditional Chinese Xiao Liu Ren divination based on the current lunar calendar.
Featured*
