#11104: feat(agent): add Ralph persistence loop for autonomous continuation

## Summary - Add configurable Ralph persistence loop that allows the agent to continue working autonomously after each reply - When `ralph.enabled` is true in `openclaw.json`, the agent re-dispatches with a continuation prompt after each reply, up to `maxIterations` times - Trigger internal hook `agent:reply-complete` for hook-based loop control - Assign unique `MessageSid` per continuation to bypass inbound dedup ## Config Example ```json { "ralph": { "enabled": true, "maxIterations": 5, "continuationPrompt": "Continue working. Complete the remaining tasks." } } ``` ## Files Changed - `src/config/types.openclaw.ts` - Add `ralph` config type - `src/config/zod-schema.ts` - Add `ralph` to validation schema - `src/auto-reply/dispatch.ts` - Add continuation loop in `dispatchInboundMessage` - `src/auto-reply/reply/dispatch-from-config.ts` - Add `ralphContinue` flag and hook trigger ## Test plan - [x] TypeScript compilation passes - [x] Config validation accepts `ralph` key - [x] Gateway API test: ralph loop runs for maxIterations - [x] Discord channel test: bot sends continuation replies - [x] Inbound dedup bypass verified with unique MessageSid 🤖 Generated with [Claude Code](https://claude.com/claude-code) <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR adds a new optional `ralph` config section (types + Zod schema) and wires a persistence loop into the inbound dispatch path. When enabled, `dispatchReplyFromConfig` can signal `ralphContinue` after a reply completes; `dispatchInboundMessage` then re-dispatches with an injected continuation prompt and a fresh `MessageSid` to bypass inbound de-duplication. It also introduces an internal `agent:reply-complete` hook event intended to let hooks control whether the loop continues. Key behavior change is that a single inbound message may now cause multiple reply cycles (up to `maxIterations`) when Ralph continuation is active. <h3>Confidence Score: 3/5</h3> - This PR is close to safe, but has a functional gap in the new hook-driven continuation flow. - Core changes are localized (config schema/types + dispatch loop) and the loop trigger is guarded by `cfg.ralph?.enabled` in `dispatchReplyFromConfig`. However, the new `agent:reply-complete` hook event exposes a `messages` channel that is currently never dispatched when the loop triggers, which will break expected hook behaviors. There’s also a maintainability hazard where the loop in `dispatchInboundMessage` is not itself gated on `cfg.ralph.enabled`, relying on downstream behavior. - src/auto-reply/reply/dispatch-from-config.ts; src/auto-reply/dispatch.ts <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->