Read what your agent actually ran.
Linger opens Claude Code and Codex sessions in your terminal. Step through the tool calls, inspect their inputs and outputs, and look up the parts of a command you don’t recognise.
When something is worth coming back to, keep the example and write yourself a note.
A place to stop and read.
An agent can finish a task using commands you’ve never written yourself. You get the result, but some of the work can remain unfamiliar.
Linger keeps the call, its output and the surrounding session close together. You can follow a run as it happens or move back through a recording. Earlier moments show only the evidence that was available then.
The field guide grows from calls you inspect. Keep examples of a command, add a note, and mark what you want to understand or practise. It stays on your machine.
Reading a transcript never executes its commands. Command documentation works locally with an optional pack. You can also request a model interpretation of a selected call; that sends its evidence to your configured provider.
Install Linger
macOS / Apple SiliconmacOS / Intel
Or build the 0.1.0 release with Rust 1.88 or newer:
cargo install --git https://github.com/rickhallett/linger --tag runtime-v0.1.0 --lockedRun linger from a project directory, or pass it a session file. The guide covers keyboard controls and the optional documentation pack.
Early software. macOS binaries are not Apple Developer ID signed or notarized. Linux binary distribution is still to be decided. Checksums and release notes.