#18893: fix(message): use currentMessageId for react fallback and suppress recoverable react warnings

## Summary Improve agent-initiated `message(action=react)` behavior by wiring `currentMessageId` through the embedded tool pipeline and using it as the fallback source when `messageId` is missing. Also suppresses recoverable react input warnings (for example `messageId required`) from user-visible payloads to avoid noisy error text leaks. ## Changes - Add `currentMessageId` to `ChannelThreadingToolContext` and threading context builder. - In `runMessageAction`, for `react`, fill `params.messageId` from `toolContext.currentMessageId` when missing. - Wire `currentMessageId` through embedded runner/tool creation path: - `run/params.ts`, `run.ts`, `run/attempt.ts` - `pi-tools.ts`, `openclaw-tools.ts`, `message-tool.ts` - Add/adjust tests: - `message-action-runner.test.ts` - `reply-plumbing.test.ts` - `message-tool.e2e.test.ts` - `payloads.e2e.test.ts` ## Validation - `pnpm vitest run src/infra/outbound/message-action-runner.test.ts src/auto-reply/reply/reply-plumbing.test.ts` - `pnpm vitest run --config vitest.e2e.config.ts src/agents/pi-embedded-runner/run/payloads.e2e.test.ts src/agents/tools/message-tool.e2e.test.ts` - `pnpm build` Refs #18154 Refs #17782 Refs #17651 <!-- greptile_comment --> <h3>Greptile Summary</h3> This PR wires `currentMessageId` through the embedded tool pipeline so that `message(action=react)` can use it as a fallback when the LLM omits `messageId`. It also adds a suppression mechanism for recoverable react input errors in the payload builder to avoid leaking noisy warnings into user-facing chats. - Adds `currentMessageId` to `ChannelThreadingToolContext` and propagates it through `buildThreadingToolContext`, `RunEmbeddedPiAgentParams`, `createOpenClawCodingTools`, `createOpenClawTools`, `createMessageTool`, and the embedded runner/attempt pipeline. - In `runMessageAction`, fills `params.messageId` from `toolContext.currentMessageId` when the action is `react` and `messageId` is missing — this is the core behavior fix. - Adds `isRecoverableMessageReactInputError` in `payloads.ts` to suppress user-visible warning payloads for react input errors. **However, the suppression is overly broad**: it matches any `RECOVERABLE_TOOL_ERROR_KEYWORDS` (e.g., "required", "missing", "invalid") for react actions, not just `messageId`-related ones. This could silently hide legitimate errors like Discord's `"emoji required"` when the emoji string is empty. - Tests cover the new `messageId` inference, context forwarding, and suppression behavior, though the suppression test only exercises the `"messageId required"` case. <h3>Confidence Score: 3/5</h3> - The core `messageId` fallback logic is correct and well-tested, but the error suppression in `payloads.ts` is too broad and could hide legitimate react errors. - The plumbing of `currentMessageId` through the pipeline (12 of 14 files) is clean, consistent with existing patterns (`currentChannelId`, `currentThreadTs`), and well-tested. The `runMessageAction` fallback logic is correctly placed and guarded. However, `isRecoverableMessageReactInputError` suppresses all recoverable-keyword errors for react actions, not just `messageId`-specific ones — this could silently hide errors like "emoji required" from Discord. Score reflects the one meaningful logic concern in an otherwise well-executed PR. - `src/agents/pi-embedded-runner/run/payloads.ts` — the `isRecoverableMessageReactInputError` function's error matching is too broad and should be narrowed to `messageId`-related errors specifically. _{Last reviewed commit: 42a1f41} <!-- greptile_other_comments_section --> _{(3/5) Reply to the agent's comments like "Can you suggest a fix for this @greptileai?" or ask follow-up questions!} <!-- /greptile_comment -->