fix(subagent): retry announce on timeout

[Limitless2023]

Fixes #17000

Problem

Sub-agent task results are silently lost when announcement delivery exceeds the 60-second gateway timeout. No retry, no error shown to user.

Solution

• Add retry logic with exponential backoff (up to 3 attempts)
• Delay between retries: 5s, 10s, 20s (capped at 30s)
• Log retries and final failures for debugging

Impact

• Subagent results are more reliably delivered to parent session
• Timeout failures are now logged instead of silently dropped
• Backoff prevents overwhelming busy gateway

Greptile Summary

Adds retry logic with exponential backoff (up to 3 attempts, 5s/10s/20s delays) to sendAnnounce in src/agents/subagent-announce.ts to handle gateway timeout failures when delivering subagent task results.

• The retry mechanism correctly catches timeout errors and applies capped exponential backoff before retrying
• Issue: A new idempotencyKey is generated on each retry attempt. Since the gateway deduplicates agent method calls by idempotency key, retries bypass dedup and can cause duplicate message delivery if the original request succeeded but its response timed out
• The previously-flagged log variable issue (undefined/not imported) still needs to be resolved before merging

Confidence Score: 2/5

• This PR has a runtime crash bug (undefined log) and a duplicate delivery risk from regenerating idempotency keys on retries.
• Score of 2 reflects two issues: the log ReferenceError (already flagged, not yet fixed) will crash the retry/error paths at runtime, and the new idempotencyKey per retry bypasses gateway deduplication, risking duplicate subagent announcements to the parent session.
• src/agents/subagent-announce.ts — both the log import and idempotencyKey reuse need to be addressed

_{Last reviewed commit: 59e87d9}

[greptile-apps]

_{1 file reviewed, 2 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

log is not defined or imported, will throw ReferenceError. Use defaultRuntime.error instead (already imported on line 15).

Suggested change

	log.warn(`Subagent announce timeout, retrying (attempt ${attempt + 1}/${maxAttempts})`, {
	defaultRuntime.error?.(`Subagent announce timeout, retrying (attempt ${attempt + 1}/${maxAttempts})`, {

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/agents/subagent-announce.ts
Line: 142:142

Comment:
`log` is not defined or imported, will throw ReferenceError. Use `defaultRuntime.error` instead (already imported on line 15).

```suggestion
      defaultRuntime.error?.(`Subagent announce timeout, retrying (attempt ${attempt + 1}/${maxAttempts})`, {
```

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

log is not defined or imported, will throw ReferenceError. Use defaultRuntime.error instead (already imported on line 15).

Suggested change

	log.error(`Subagent announce failed after ${maxAttempts} attempts: ${errorMsg}`, {
	defaultRuntime.error?.(`Subagent announce failed after ${maxAttempts} attempts: ${errorMsg}`, {

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/agents/subagent-announce.ts
Line: 151:151

Comment:
`log` is not defined or imported, will throw ReferenceError. Use `defaultRuntime.error` instead (already imported on line 15).

```suggestion
      defaultRuntime.error?.(`Subagent announce failed after ${maxAttempts} attempts: ${errorMsg}`, {
```

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

New idempotencyKey per retry risks duplicate delivery

Each retry generates a fresh UUID here. The gateway's agent method deduplicates requests using idempotencyKey (see src/gateway/server-methods/agent.ts:209: context.dedupe.get(\agent:${idem}`)`). If the first attempt was processed by the gateway but the response timed out on the client side, the retry with a new key will bypass dedup and could trigger a duplicate message delivery to the parent agent.

Generate the key once before the try block and reuse it across retries. For example, add an idemKey parameter (defaulting to a new UUID on the first call) and pass it through on recursive calls:

// At the top of sendAnnounce, before the try block:
const idempotencyKey = idemKey ?? crypto.randomUUID();

// Then on recursive retry:
return sendAnnounce(item, attempt + 1, maxAttempts, idempotencyKey);

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/agents/subagent-announce.ts
Line: 131:131

Comment:
**New idempotencyKey per retry risks duplicate delivery**

Each retry generates a fresh UUID here. The gateway's `agent` method deduplicates requests using `idempotencyKey` (see `src/gateway/server-methods/agent.ts:209`: `context.dedupe.get(\`agent:${idem}\`)`). If the first attempt was processed by the gateway but the *response* timed out on the client side, the retry with a new key will bypass dedup and could trigger a duplicate message delivery to the parent agent.

Generate the key once before the try block and reuse it across retries. For example, add an `idemKey` parameter (defaulting to a new UUID on the first call) and pass it through on recursive calls:

```
// At the top of sendAnnounce, before the try block:
const idempotencyKey = idemKey ?? crypto.randomUUID();

// Then on recursive retry:
return sendAnnounce(item, attempt + 1, maxAttempts, idempotencyKey);
```

How can I resolve this? If you propose a fix, please make it concise.