# Revalidation > Revalidate Codex project instructions, agents, skills, or migrated AI configuration. Use when the user asks for agent revalidation, skill revalidation, config revalidation, or to check that AGENTS.md and .codex/skills match the codebase. Use when this capability is needed. - Skill: `tomevault-io/revalidation` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/revalidation` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/revalidation/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/revalidation --- # Agent And Skill Revalidation Use this workflow to keep Codex configuration accurate and coherent with the actual codebase. ## Step 1: Structural Validation Re-read the target agent or skill configuration file in full. Validate every documented fact against the current codebase, including current-session changes: - File and directory paths exist and match the documentation. - Referenced tools, reference docs, and linked files exist. - Named types, classes, dialogs, and namespaces are accurate. - Documented conventions still match existing code. Apply objective structural fixes automatically. These corrections do not need approval. ## Step 2: Logical Analysis Analyze the configuration beyond mechanical path checks: - Responsibilities align with the available tools and reference files. - Conventions still make sense for the current codebase. - Sections do not overlap, contradict each other, or duplicate stale guidance. - Commonly edited areas are covered. - Self-maintenance rules catch likely side effects. - Reference files are structured well enough to scale. Present logical findings one at a time, ordered by importance. For each finding: 1. Describe the issue and concrete suggestion. 2. Ask the user to approve, reject, or modify the suggestion. Use a concise plain-text question when no structured question tool is available. 3. Apply approved changes immediately before moving to the next finding. Do not batch logical findings. Do not apply logical-analysis changes without user approval. --- > Source: [BitzArt-VS/UI-Tweaks](https://github.com/BitzArt-VS/UI-Tweaks) — distributed by [TomeVault](https://tomevault.io).