Compare targets
Pick two tools to see what Claude Code and Codex: 2 of 10 spec kinds differ in support, 3 in file format. This is the default comparison. Turn on JavaScript to pick other targets, or read the capability matrix. These targets have the same state and file format for every spec kind. Ahead means a target writes a spec kind by default where another column does not. Behind is the reverse.agnostic-ai sync writes for each spec kind, and where. Target C is optional: pick one to compare three. The page URL keeps the choice, so you can share a comparison.Spec kind Claude Code claudeCodex codexAgentsFormat differs .claude/.codex/Skills .claude/.agents/Rules .claude/AGENTS.mdHooks .claude/.codex/MCPFormat differs .mcp.json.codex/CommandsDiffers .claude/SettingsFormat differs .claude/.codex/Code reviewTool-specificDiffers AGENTS.mdDev setupTool-specific Ignored files Where they differ
Where the data comes from
Support states come from the capability matrix. Paths come from a real sync of one sample spec per kind, and CI fails when either drifts from the adapters. File formats come from each path's extension. A file that holds more than one kind, such as .claude/settings.json, appears under each kind that writes to it.
This page compares output files, not how each tool behaves once it reads them. Open a target page for field-level caveats and the config keys an Opt-in kind needs.