Pre-Flight Checks Skill

Test-driven behavioral verification for AI agents

Catches "silent degradation" - when agent loads memory but doesn't apply learned behaviors.

Quick Start

# Install
cd ~/.openclaw/workspace/skills
git clone https://github.com/IvanMMM/preflight-checks.git

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

# Edit your checks and answers
vim PRE-FLIGHT-CHECKS.md
vim PRE-FLIGHT-ANSWERS.md

# Update AGENTS.md to run checks after loading memory

What It Does

Problem: Agent knows rules but doesn't follow them.

Memory loaded ✅ → Rules understood ✅ → Behavior wrong ❌

Solution: Behavioral unit tests.

1. Agent loads memory
2. Agent reads PRE-FLIGHT-CHECKS.md
3. Agent answers scenarios
4. Agent compares with PRE-FLIGHT-ANSWERS.md
5. Agent reports score: N/N ✅ or failures ❌

Files

skills/preflight-checks/
├── SKILL.md              # Full documentation
├── README.md             # This file
├── templates/
│   ├── CHECKS-template.md    # Blank template
│   └── ANSWERS-template.md   # Blank template
├── scripts/
│   ├── init.sh               # Setup in workspace
│   └── add-check.sh          # Add new check
└── examples/
    ├── CHECKS-prometheus.md  # Real example (23 checks)
    └── ANSWERS-prometheus.md # Real answers

Usage

Add Check

./skills/preflight-checks/scripts/add-check.sh
# Interactive prompts guide you through

Run Checks

Manual (recommended for agents):

1. Read PRE-FLIGHT-CHECKS.md
2. Answer each CHECK-N
3. Compare with PRE-FLIGHT-ANSWERS.md
4. Report score

Automated (optional, TBD):

./skills/preflight-checks/scripts/run-checks.sh

Examples

See examples/ directory for real working pre-flight system:

• 23 behavioral checks
• Categories: Identity, Saving, Communication, Telegram, Anti-Patterns
• Covers common degradation patterns

When to Use

• ✅ Building agent with persistent memory
• ✅ Want behavioral consistency across sessions
• ✅ Testing agent after updates
• ✅ Catching drift/degradation early

Integration

Add to AGENTS.md:

## Every Session

1. Read SOUL.md
2. Read USER.md
3. Read memory/YYYY-MM-DD.md
4. If main session: Read MEMORY.md
5. **Run Pre-Flight Checks**

### Pre-Flight Checks

1. Read PRE-FLIGHT-CHECKS.md
2. Answer each scenario
3. Compare with PRE-FLIGHT-ANSWERS.md
4. Report discrepancies

Learn More

See SKILL.md for:

• Detailed problem description
• Check writing guidelines
• Scoring rubrics
• Advanced usage (CI/CD, automation)
• Contributing guide

Credits

Created by Prometheus (OpenClaw agent)
Inspired by aviation pre-flight checklists

License

MIT