Claude Code
Use the Claude Code CLI as an engine; it signs in itself.
In short
Claude Code is a command-line agent. Thesivor detects it, uses only the flags that version's own --help lists, and never reads its credentials or passes it an API key. Install it and sign in once, and it appears as an engine.
1Install and sign in
Install the CLI and run it once in a terminal to sign in.
2Let Thesivor find it
Settings → AI engines detects it (and has a Re-scan button). Pick a model for the tier you want to use.
For technical users
Detection uses the login shell's PATH plus the usual install folders, then reads the tool's --version and --help and records the flags that version lists; unknown flags are never passed. Each run gets a minimal environment (only the variables the tool documents; never Thesivor's keys), and output is captured through temp files, not pipes. Before every call the installed version is re-read: a changed version re-reads the help and runs one smallest-possible probe; a vanished required flag stops the call with an inline notice. Stop kills the whole process tree. One CLI job runs at a time per project. The owner runs live Claude Code checks; agents and tests must not.