Show HN: Back up Claude Code sessions from Codespaces to a private repo

Logkeeper addresses a concrete durability gap for developers who run Claude Code inside GitHub Codespaces instead of on a local machine. When a remote Codespace is deleted or rebuilt, Claude Code session logs vanish with the environment. The project, shared as a Show HN, backs those sessions up so conversation history is not tied to the lifetime of a disposable remote workspace.

On each Claude Code session stop, Logkeeper writes two artifacts into a designated private repository: a human-readable conversation log and an age-encrypted copy of the full JSONL transcript. The split keeps day-to-day reading simple while the encrypted JSONL preserves the complete session record under age encryption. Setup is limited to creating that private repository, adding the required Codespaces secrets, and committing a small set of hook and configuration files into the project so the backup runs when a session ends.

For engineers and builders, the value is session continuity across Codespace rebuilds and teardown. Agent work often spans multiple days; losing JSONL history means losing prompts, tool traces, and decision context that would otherwise be available only while the Codespace still exists. Backing that state into a private repo turns ephemeral remote compute into something you can reopen after the machine is gone.

The product sits in a practical niche rather than competing with general cloud backup or full IDE sync. It is scoped to Claude Code on Codespaces: private-repo destination, Codespaces secrets, and stop-time hooks. That is narrower than local disk retention or vendor-side chat history, but it matches how many people already isolate agent work in short-lived remote environments and still need the transcript afterward.

If you already prefer Codespaces for Claude Code, the next step is whether Logkeeper’s private-repo plus secrets plus hooks model fits your repo security model. Watch how well the human-readable log and encrypted JSONL pair holds up across rebuilds, and whether stop-time capture is enough for sessions that end abnormally rather than cleanly.