Supported tools
cnfgr reads configuration from, and writes it to, the AI coding tools below. It stores everything in Claude Code’s data shape (the richest model) and projects each tool’s native format at write time, so some tools receive a faithful subset of what others support.
Support matrix
| Tool | Read | Write | Sessions | Notes |
|---|---|---|---|---|
| Claude Code | ✅ Full | ✅ Full | ✅ Backup only | Reference implementation; richest data model |
| Codex CLI | ✅ Full | ✅ Full | ✅ Backup only | TOML config; AGENTS.md context |
| Gemini CLI | ✅ Full | ✅ Full | ❌ Skip | TOML commands; GEMINI.md context |
| GitHub Copilot | ✅ Full | ✅ Full | ❌ Skip | copilot-instructions.md + AGENTS.md + *.agent.md |
| Cursor | ⚠️ Partial | ⚠️ Partial | ❌ Skip | .cursor/rules/*.mdc only |
Legend — ✅ Full: complete support. ⚠️ Partial: a reduced subset. ❌ Skip: not handled. Sessions refers to archiving a tool’s session transcripts.
What “fidelity loss” means
When a tool can’t represent something cnfgr stores (for example, a tool with no native skill concept, or fewer agent fields), cnfgr writes the closest faithful equivalent and preserves the full canonical record so nothing is lost on the tools that do support it. Each tool page below calls out where that happens.