Agent history is where claims and decisions come from. Mantyl reads each tool’s local session store directly, on your machine, and every transcript passes through the same fail-closed secret redaction before anything is stored or analysed. Whatever the agent, its statements enter the passport as agent reported and are never upgraded without executed evidence.
| Agent | Status | Requirement |
|---|---|---|
| Claude Code | Full support | None beyond the CLI |
| Cursor | Beta | Node 22.13 or newer |
| OpenAI Codex CLI | Beta | None beyond the CLI |
Claude Code writes one transcript file per session under your home directory, in a folder derived from the project’s path. Run mantyl from the same project root you used with Claude Code and the sessions are found automatically. If your transcripts live somewhere unusual, mantyl scan --session-dir points at them directly. This is the deepest integration and the reference for what every other adapter must reach.
Cursor keeps chat history in local databases inside its application storage. Mantyl matches your project by the folder you opened in Cursor, then reads every conversation belonging to it, including history spread across multiple storage entries from re-opens and Cursor upgrades. Reading these databases uses a capability built into Node 22.13 and newer; on older versions Cursor support reports itself unavailable and everything else keeps working. Cursor’s storage format is undocumented, which is why this adapter is beta: it is tested against real installations, and a future Cursor update could change the layout before we catch it.
Codex records each session as a file under ~/.codex/sessions, and every session names the working directory it ran in. Mantyl matches your project against that record, so run it from the same folder you ran Codex in. Injected instructions, reasoning traces and tool traffic are skipped: only the actual conversation between you and the agent feeds the passport. The ChatGPT desktop app is not supported, because its conversations live in OpenAI’s cloud without project scoping, and there is nothing local and reliable for Mantyl to read.
Start with mantyl doctor. It reports each agent separately: whether the tool’s store exists on this machine, and whether it holds sessions for the current folder. The most common cause of an empty result is running mantyl from a different directory than the one the agent worked in, since every adapter matches sessions to the exact project folder. A project with no agent history at all still produces a passport from repository evidence alone, just without claims or decisions.