feat: mission autopilot and continuation loops - #1150

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

Good instinct - splitting the mission logic out of #1148 and constraining it to a per-scope file (getMissionScopeId) instead of execSync('git rev-parse...') is the right call, and the read of use-send-message.ts's retry loop is a legitimate concern. But the execution needs another pass:

1. The stated translation isn't actually done. The PR description claims Portuguese strings were translated, but mission-store.ts's buildMissionPrompt still emits **PROTOCOLO DE DOUBLE-CHECK OBRIGATÓRIO**... não há itens \- [ ]`... CONTINUE TRABALHANDO, commands/mission.ts's fallback message is 'Uso: /mission [status|start |...]', and mission-todos-tracker.tsxhasMETA:, ETAPAS, 'Sem meta definida'. Even the new tests assert on Portuguese strings ('MISSÃO ATIVA', 'não encerre', 'evidência'`). This is a half-finished pass, not a fix.

2. The retry-cap fix has no direct test. use-send-message.ts now recurses via runSendMessage(nextArgs) and stashes retryCount as a hidden property on arguments[0], mutating the params object rather than threading it as a real parameter. It works, but it's a fragile pattern in a file this central, and none of the added tests (mission-autopilot.test.ts, mission-store.test.ts) actually exercise it.

3. Scope/design concerns for a maintainer to weigh: the autopilot effect in chat.tsx automatically resubmits prompts and auto-renews ended sessions (autoRestart in session-ended-banner.tsx) without user confirmation - a meaningful behavior/product change bundled into what's framed as a bugfix PR. refreshMissionCompletion hardcodes a single task_plan.md path convention that isn't documented or validated anywhere else in the diff.

4. Size (+561/-12 across 11 files, three new modules) is large for a single PR building an entirely new autonomous subsystem; splitting the mission-store/autopilot core from the UI/tracker component from the retry fix would make this reviewable.

Recommend: finish the translation pass, remove the arguments hack in favor of an explicit param, and split into smaller PRs.