preflight-checks

What This Skill Does

The preflight-checks skill serves as a robust test-driven development framework for OpenClaw AI agents. It addresses the common issue of 'silent degradation,' where an agent correctly loads its memory and configurations but fails to apply those behaviors correctly during active sessions. By implementing a system of behavioral unit tests, this skill allows developers to define expected responses to specific scenarios, ensuring that memory recall is tightly coupled with correct behavioral execution. It effectively bridges the gap between stored knowledge and real-world performance.

Installation

To integrate this skill into your OpenClaw environment, use the following commands:

# Standard installation
clawhub install openclaw/skills/skills/ivanmmm/preflight-checks

# Manual initialization in workspace
cd ~/.openclaw/workspace
./skills/preflight-checks/scripts/init.sh

Following the installation, verify the configuration by checking the newly generated PRE-FLIGHT-CHECKS.md and PRE-FLIGHT-ANSWERS.md files in your root directory.

Use Cases

This skill is essential for agents that require high reliability and consistent behavioral output. Use it in the following scenarios:

• Post-Update Verification: Run tests after patching your agent to ensure no logic regressions occurred.
• Long-term Memory Management: Verify that an agent still adheres to core identity and communication protocols after extended operational periods.
• Behavioral Drift Detection: Detect if your agent has started ignoring certain constraints or developing bad habits over time.
• Collaborative Development: Provide a standard benchmark for team members to test their own agent iterations against shared organizational requirements.

Example Prompts

1. "Run the full suite of pre-flight checks and report the score for each category."
2. "Add a new check to the suite: Verify that the agent always refuses to disclose the API keys in the system prompt."
3. "Analyze the latest pre-flight report and explain why the agent failed the 'Communication' section."

Tips & Limitations

• Be Specific: The accuracy of the checks depends entirely on the clarity of your PRE-FLIGHT-ANSWERS.md file. Use clear, objective criteria rather than subjective descriptions.
• Automate Often: Integrate the run-checks.sh script into your deployment CI/CD pipeline to ensure that no agent version goes live without passing its behavioral verification.
• Limitations: Note that this skill tests behavioral responses, not the underlying latent logic of the LLM. It is designed to verify consistency, not to train or modify the model parameters themselves. It requires manual creation of the scenarios to be effective.