[quickbeard]

While a turn is running, Ctrl+C now aborts it on a single press, and the prompt footer leads with ctrl+c interrupt instead of esc interrupt. escape keeps its existing double-press interrupt, so nobody's muscle memory breaks.

Behavior

ctrl+c	prompt has draft	prompt empty
busy	clears draft	aborts immediately
idle	clears draft	arms exit confirm, second press quits

While busy, Ctrl+C can no longer arm the double-press quit from #51. Ctrl+D, <leader>q, and /exit still quit immediately, including mid-turn.

Approach

Added a session_interrupt_now keybind (default ctrl+c) and its session.interrupt.now command rather than adding ctrl+c to session_interrupt — the two need different press counts, and inspecting the triggering key inside one handler breaks as soon as someone rebinds.

Precedence runs through enabled predicates rather than binding order. Order is not a stable disambiguator here: same-priority layers resolve by registration order, and useBindings re-registers a layer on every reactive change, which bumps it to newest.

• session.interrupt.now requires a busy session and an empty prompt, so ctrl+c falls through to prompt.clear with a draft present and to app.exit.confirm while idle.
• app_exit_confirm moves to its own binding layer that also requires the session to be idle. app_exit keeps the original predicate.

Notes

• The home screen's "stop the AI mid-response" tip now points at session.interrupt.now, so it agrees with the footer.

• Permission and question prompts keep ctrl+c as their scoped reject — unchanged.

• The interactive split-footer mode (codev --mini, and --mini --attach) has its own separate two-press ctrl+c logic; left alone, it's an independent job. Plain non-interactive codev run "…" is unaffected either way — it builds no renderer, keymap, or footer, and ctrl+c there is an ordinary SIGINT.

• Adding a keybind drifts packages/sdk/js/src/gen/types.gen.ts, which is not regenerated here — following the precedent set by feat(tui): require double Ctrl+C to quit #51, which added app_exit_confirm the same way.

  Correction, added after merge: that claim was wrong. Nothing regenerates src/gen — packages/sdk/js/script/build.ts outputs to src/v2/gen and merely runs prettier over src/gen (line 115). The v2 output contains no keybind types at all, and a freshly generated spec (bun dev generate) has no keybind schema and no keybinds property on Config, because keybinds were moved out of the server config: packages/opencode/src/config/config.ts:54-57 deletes keybinds as a legacy key and config/tui-migrate.ts migrates it into the TUI's own config.

  So KeybindsConfig (src/gen/types.gen.ts:772) and Config.keybinds (line 1184) are fossils of a pre-migration API. This PR does not drift them — regenerating correctly would delete them, which is a separate, breaking cleanup unrelated to this change. The schema that actually validates keybinds (packages/tui/src/config/index.tsx:56, built from TuiKeybind.KeybindOverrides) derives from Definitions at runtime and already knows about session_interrupt_now.

Testing

• New packages/tui/test/interrupt.view.test.tsx boots the real app against a session the fixture server reports as busy, then drives actual ctrl+c key events: draft cleared with no abort, then abort on a single press, with the exit confirm verified disarmed.
• bun typecheck clean across all 31 packages.
• Full tui suite: 232 pass / 1 fail. The failure (app.exit prints the session epilogue after scoped cleanup) is pre-existing and fails identically on an unmodified tree.

Docs (en + es), DIVERGENCES.md, and CHANGELOG.md updated.

🤖 Generated with Claude Code