preflight-checks_skill

Overview

This skill provides test-driven behavioral verification for AI agents to detect silent degradation when memory loads but learned behaviors are not applied. It adds scenario-based checks and canonical answers so agents can self-test after memory updates, restarts, or configuration changes. The result is objective pass/fail reporting and actionable feedback to restore expected behavior.

How this skill works

Define a set of scenario checks and expected answers in human-readable CHECKS and ANSWERS files. On startup or on demand, the agent reads the checks, answers each scenario, and compares its responses to the expected answers to produce a score and failure report. Scripts and templates automate initialization, adding new checks, and running the full suite; optional automation integrates checks into CI or session workflows.

When to use it

• After session restart or when bootstrapping a persistent-memory agent
• Immediately after memory updates or rule changes
• When testing agent behavior after software updates or refactors
• During onboarding of new agent instances or profiles
• On demand whenever behavior consistency is uncertain

Best practices

• Write scenario-based checks that test behavior, not trivia or raw recall
• Include explicit wrong answers to make drift detection precise
• Group checks into categories (Identity, Core Behavior, Communication, Anti-Patterns)
• Keep a living ANSWERS file as the canonical behavioral reference
• Run checks automatically on critical lifecycle events and in CI for regressions

Example use cases

• Detect an agent that loads a tool list into memory but fails to auto-save new tools
• Verify behavioral rules after merging a code change that touches decision logic
• Run nightly or pre-release behavioral checks in CI to catch regressions
• Onboard a new agent profile with role-specific PRE-FLIGHT-CHECKS files
• Automatically trigger reloading and retesting when score drops below threshold