#23048: feat(session): add `announceDeliver` option to suppress auto-delivery of subagent results

## Summary Add `session.announceDeliver` (boolean, default `true`) to let users opt out of automatic delivery of subagent announce replies to the chat channel. ## Motivation When agents use structured message formats — Feishu interactive cards, Slack Block Kit, Telegram InlineKeyboard, etc. — instead of plain text, the current hardcoded `deliver: true` in the announce flow causes raw text to leak into the chat channel *before* the agent can send a properly formatted message via its own tools. Today the only workaround is patching `deliver` in the compiled dist files, which breaks on every update. A first-class config option lets users cleanly control this behavior. Related: #13267 ## Changes **`src/config/types.base.ts`** - Add `announceDeliver?: boolean` to `SessionConfig` with JSDoc **`src/config/zod-schema.session.ts`** - Add `announceDeliver: z.boolean().optional()` to `SessionSchema` **`src/agents/subagent-announce.ts`** - `sendAnnounce()` (queue drain path): read `cfg.session.announceDeliver` and pass it to the `deliver` param - `sendSubagentAnnounceDirectly()` (direct path): same change - `buildAnnounceReplyInstruction()`: when `announceDeliver === false`, instruct the agent to use its own tools for delivery and reply `NO_REPLY` - `runSubagentAnnounceFlow()`: thread the config value into `buildAnnounceReplyInstruction()` ## Behavior | `announceDeliver` | Trigger injected? | LLM reply auto-delivered? | Reply instruction | |---|---|---|---| | `true` (default) | ✅ | ✅ | "Convert and send user-facing update now" | | `false` | ✅ | ❌ | "Use your own tools to deliver; reply NO_REPLY" | ## Usage ```json { "session": { "announceDeliver": false } } ``` ## Backward compatibility Fully backward-compatible. The option defaults to `true` when omitted — existing behavior is unchanged. ## Test plan - [ ] Verify default behavior (no config / `true`): subagent results auto-delivered as before - [ ] Verify `announceDeliver: false`: trigger injected into session, agent processes result, but LLM reply is NOT auto-sent to channel - [ ] Verify agent can still deliver via its own tools (e.g. card API) when `announceDeliver: false` - [ ] Verify nested subagent announces (requesterIsSubagent) are unaffected by this option 🤖 Generated with [Claude Code](https://claude.com/claude-code) <!-- greptile_comment --> <h3>Greptile Summary</h3> Added `session.announceDeliver` config option (boolean, defaults to `true`) to control whether subagent completion results are automatically delivered to the chat channel. When disabled, the announce trigger is still injected but the LLM's reply is not auto-sent, allowing agents to deliver via structured message tools (Feishu cards, Slack blocks, etc.) instead of leaking raw text. - Config plumbing: added field to `SessionConfig` type with JSDoc, added to Zod schema validation - Modified three announce delivery paths (`sendAnnounce`, `sendSubagentAnnounceDirectly`, `runSubagentAnnounceFlow`) to read config and conditionally set `deliver` parameter - Updated `buildAnnounceReplyInstruction` to provide different instructions when auto-delivery is disabled (instructs agent to use own tools and reply with `NO_REPLY`) The implementation correctly defaults to `true` (maintaining backward compatibility) and handles the three code paths that control delivery. <h3>Confidence Score: 4/5</h3> - Safe to merge with minor style inconsistency - The implementation is logically correct and backward-compatible. Config schema, type definitions, and all three delivery paths are properly updated. The inconsistency in default value handling (line 1122 passes raw value vs normalized `!== false` in other locations) works correctly but creates maintenance risk if the check in `buildAnnounceReplyInstruction` changes. - No files require special attention _{Last reviewed commit: 94598c5} <!-- 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 -->