configlisted

# adversarial-review-loop configuration ```! ${CLAUDE_PLUGIN_ROOT}/scripts/arl.sh config --args-stdin <<'ARL-ARGUMENTS-EOF' $ARGUMENTS ARL-ARGUMENTS-EOF ``` This command is unrelated to any armed activation — it never registers the review gate, and it does not require one to be running. It only ever touches the user's own config file (`~/.config/adversarial-review-loop/config.json` by default) or, with `--repo`, the repository's own `.adversarial-review-loop.json`. - **No arguments**: prints every supported key, its currently resolved value, and which layer set it — `default`, `user`, `repo`, `activation` (an override an armed session was given via `--harness`/`--model`/`--variant`/`--guide`), or `env`. `model` shows `default: <harness>`, because its default is the selected harness's own rather than a constant. - **`<key> <value>`**: writes `key` into the user config. Add `--repo` to write the repository config instead. If the value itself contains a token that looks like a flag (a `verify_cmd` such as `pytest --maxfail=1`), put `--` before it: `config verify_cmd --repo -- pytest --maxfail=1`. - **`<key> --unset`**: removes `key` from that layer (add `--repo` for the repository config). - Setting `model` additionally checks the name against the selected harness's model list (`opencode models`). An unreachable reviewer only warns — arming will refuse on its own if it is still unreachable then — but a name the reviewer actively rejects is refused; pass `--force` to set it an