truth-first
by Roy Yuen
Eliminate hallucinations with an evidence-first verification framework for system state, configs, and file contents.
About This Skill
What it does
The Truth-First Verification skill enforces a rigorous, evidence-based workflow for AI agents. It prevents "hallucinated certainty" by requiring real-time verification of system states, configurations, and file contents before the agent is allowed to provide an answer. Instead of assuming a service is running or a config is correct, the agent must prove it using available diagnostic tools.
Why use this skill
Standard LLM prompts often lead to agents making assumptions based on outdated context or training data. This skill eliminates that risk by implementing a strict classification framework (Verified, Inferred, Unknown). It is essential for developers working in environments where accuracy is non-negotiable, such as infrastructure management, debugging complex deployments, or auditing configurations. It ensures every response is backed by specific command outputs or file citations.
Supported tools
- Verification via
ripgrep(rg) for targeted file searches. - System status checks (ls, stat, logs) for environment validation.
- Connectivity and mount verification.
- Configuration auditing through direct file reads and environment inspections.
The Output
The agent delivers a structured report listing every claim analyzed, its verification status, and the direct evidence (paths, log snippets, or command results) supporting the conclusion. If a state remains unknown, the agent provides the exact commands needed to find the answer.
How to Install
unzip truth-first.zip -d ~/.claude/skills/Free
One-time purchase • Own forever
Security Scanned
Passed automated security review
Permissions
File Scopes
Tags
Frequently Asked Questions
Learn More About AI Agent Skills
Similar Skills
env-doctor
Diagnoses why your project will not start. Checks runtime versions, dependencies, environment variables, databases, ports, and build artifacts systematically.
git-commit-writer
Writes conventional commit messages by analyzing your staged git changes. Detects commit type, scope, and breaking changes automatically.
diagnosing-rag-failure-modes
RAG fails quietly. It retrieves documents, returns confident-looking answers, and misses the question entirely — because the question required connecting facts across documents, reasoning about sequence, or tracing causation. This skill gives you a five-question diagnostic checklist that classifies any failing query as either RAG-safe or structurally RAG-incompatible, then maps it to the specific failure pattern and the architectural fix that resolves it.
codeberg
Fully automate Codeberg repository management, file operations, and releases via high-level Gitea API integration.