#20840: fix(message-tool): include other configured channels in tool description

## Summary - **Problem:** `buildMessageToolDescription()` and `resolveMessageToolSchemaActions()` both return early when `currentChannel` is set, only listing that channel's actions. Cron and isolated agents inherit a default channel (e.g. Signal), so they can neither discover nor invoke actions from other configured channels. - **Why it matters:** The description hides other channels from the LLM, and the schema enum rejects cross-channel actions at validation time. Agents that need to send messages cross-channel require hardcoded workarounds in every prompt. - **What changed:** Both functions now include actions from all configured channel plugins when `currentChannel` is set. Updated existing schema scoping tests and added new description tests covering multi-channel and single-channel output. - **What did NOT change (scope boundary):** No changes to tool execution, routing, permissions, or sandbox logic. `runMessageAction()` is untouched — it already routes to the correct channel at runtime via the `to`/`target` parameter. ## Change Type (select all) - [x] Bug fix - [ ] Feature - [ ] Refactor - [ ] Docs - [ ] Security hardening - [ ] Chore/infra ## Scope (select all touched areas) - [ ] Gateway / orchestration - [x] Skills / tool execution - [ ] Auth / tokens - [ ] Memory / storage - [ ] Integrations - [ ] API / contracts - [ ] UI / DX - [ ] CI/CD / infra ## Linked Issue/PR - Closes #20754 - Closes #20995 ## User-visible / Behavior Changes Tool description seen by the LLM changes from: ``` Send, delete, and manage messages via channel plugins. Current channel (signal) supports: react, send. ``` To: ``` Send, delete, and manage messages via channel plugins. Current channel (signal) supports: react, send. Other configured channels: telegram (delete, edit, react, send, topic-create), whatsapp (react, send). ``` Tool schema `action` enum now includes the union of all configured channels' actions instead of only the current channel's. This means cross-channel actions (e.g. `poll` from Discord when scoped to Telegram) pass validation and are routed correctly. No changes to config, defaults, or CLI behavior. ## Security Impact (required) - New permissions/capabilities? `No` - Secrets/tokens handling changed? `No` - New/changed network calls? `No` - Command/tool execution surface changed? `No` - Data access scope changed? `No` — the schema and description now accurately reflect channels the agent could already reach via `runMessageAction()`. No new access is granted; `currentChannel` was never a security boundary (the message tool is either fully enabled or disabled via `disableMessageTool`). ## Repro + Verification ### Environment - OS: macOS (Darwin 25.3.0) - Runtime/container: Node 22.16.0, pnpm 10.23.0 - Model/provider: N/A (description/schema change) - Integration/channel (if any): N/A - Relevant config (redacted): N/A ### Steps 1. Register two channel plugins (e.g. signal + telegram) 2. Create message tool with `currentChannelProvider: "signal"` 3. Inspect `tool.description` and `tool.parameters` action enum ### Expected - Description includes both signal actions AND other configured channel actions. - Schema action enum includes the union of all channels' actions. ### Actual - Description only includes signal actions; other channels are invisible to the agent. - Schema action enum only includes signal actions; cross-channel actions are rejected at validation. ## Evidence - [x] Failing test/log before + passing after New tests in `message-tool.e2e.test.ts`: - `"includes other configured channels when currentChannel is set"` — verifies multi-channel description and schema - `"does not include 'Other configured channels' when only one channel is configured"` — verifies single-channel deployments produce clean descriptions Updated existing schema scoping tests to assert cross-channel actions are included. Full e2e suite: 13/13 passed. Unit suite: 748/748 passed, 6177/6177 tests. ## Human Verification (required) - Verified scenarios: Built from source, ran `pnpm build && pnpm check && pnpm test:fast`, ran e2e tests for message-tool specifically. - Edge cases checked: Single channel (no "Other configured channels" suffix), channels with no actions (skipped), `filterActionsForContext` still applies to current channel, execution routing via `runMessageAction()` unchanged. - What I did **not** verify: Live gateway with real channel connections (no API keys configured). ## Compatibility / Migration - Backward compatible? `Yes` - Config/env changes? `No` - Migration needed? `No` ## Failure Recovery (if this breaks) - How to disable/revert this change quickly: Revert the single commit; both early returns are restored. - Files/config to restore: `src/agents/tools/message-tool.ts` - Known bad symptoms reviewers should watch for: Tool description becoming excessively long if many channels are configured (unlikely in single-user setups), or unexpected action validation passes (mitigated: `runMessageAction` still validates at execution time). ## Risks and Mitigations - Risk: Description string grows longer with many configured channels, consuming more prompt tokens. - Mitigation: OpenClaw is a single-user assistant — typical setups have 2–4 channels. The added text is minimal (one line per channel). - Risk: Wider action enum could let the LLM attempt an action unsupported by a specific channel. - Mitigation: `runMessageAction()` validates at execution time and returns clear errors. This matches the existing fallback behavior (no `currentChannel` path) which already exposes all actions. ## AI-Assisted Disclosure This PR was reviewed and improved with assistance from Claude Code (Claude Opus 4.6). Specifically: - **Code review**: Identified a stale comment on `message-tool.ts:542` that contradicted the new behavior, missing test cleanup (`afterEach`), and a gap in single-channel test coverage. - **Fixes applied**: Updated the misleading comment, added `afterEach` registry cleanup to the description test block, and added a new test for single-channel deployments. - **Manual verification**: Generated a throwaway script to verify multi-channel and single-channel tool description/schema output without requiring real channel connections. All changes were human-reviewed before committing.