fix: handle session.status cooldown retries in runtime fallback

[code-yeongyu]

Summary

• route session.status retry events through runtime fallback so cooldown retry loops can switch to the next fallback model
• stop model-fallback from handling session.status when runtime_fallback is enabled and dedupe retry updates by attempt, model, and retry class
• add regression coverage for runtime fallback status handling, plugin gating, and retry-status normalization helpers

Testing

• bun test
• bun run typecheck
• bun run build

---

Summary by cubic

Handle provider cooldown auto-retry signals from session.status via runtime fallback to switch to the next model immediately and avoid countdown spam. Also gate model-fallback when runtime_fallback is enabled.

• New Features

  • Runtime fallback handles session.status "retry" events, aborts in-flight requests, and prepares the next fallback model.
  • Shared utils to extract retry attempt/model and normalize messages for stable dedupe.

• Bug Fixes

  • Deduplicate session.status countdown updates by attempt, model, and retry class.
  • Match more cooldown/quota/capacity messages as retryable.
  • Prevent model-fallback from reacting to session.status when runtime fallback is enabled.
  • Clear session.status retry state on session deletion.

^{Written for commit b1946a6. Summary will update on new commits.}

[cubic-dev-ai]

2 issues found across 8 files

Confidence score: 3/5

• There is some merge risk because both findings are medium severity (6/10) with high confidence (9/10), indicating concrete behavior gaps rather than speculative nits.
• In src/hooks/runtime-fallback/event-handler.ts, not clearing the session-status retry key after session.idle/session.stop/session.error can cause retry deduplication to leak across requests in the same session, leading to incorrect retry handling.
• In src/shared/retry-status-utils.ts, the regex currently misses : in the character class, so model IDs containing colons (such as Ollama-style IDs) may not be matched correctly and can break retry-status detection for those models.
• Pay close attention to src/hooks/runtime-fallback/event-handler.ts and src/shared/retry-status-utils.ts - retry-key lifecycle and model-ID parsing need fixes to avoid request-level retry regressions.

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="src/hooks/runtime-fallback/event-handler.ts">

<violation number="1" location="src/hooks/runtime-fallback/event-handler.ts:38">
P2: Clear the session status retry key when a request completes (`session.idle`, `session.stop`, `session.error`) to avoid incorrectly deduplicating retry events across different requests in the same session.</violation>
</file>

<file name="src/shared/retry-status-utils.ts">

<violation number="1" location="src/shared/retry-status-utils.ts:17">
P2: Add `:` to the regex character class to correctly match model IDs that contain colons (e.g., Ollama models).</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[cubic-dev-ai]

P2: Clear the session status retry key when a request completes (session.idle, session.stop, session.error) to avoid incorrectly deduplicating retry events across different requests in the same session.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At src/hooks/runtime-fallback/event-handler.ts, line 38:

<comment>Clear the session status retry key when a request completes (`session.idle`, `session.stop`, `session.error`) to avoid incorrectly deduplicating retry events across different requests in the same session.</comment>

<file context>
@@ -33,6 +35,7 @@ export function createEventHandler(deps: HookDeps, helpers: AutoRetryHelpers) {
       sessionRetryInFlight.delete(sessionID)
       sessionAwaitingFallbackResult.delete(sessionID)
       helpers.clearSessionFallbackTimeout(sessionID)
+      sessionStatusHandler.clearRetryKey(sessionID)
       SessionCategoryRegistry.remove(sessionID)
     }
</file context>

[cubic-dev-ai]

P2: Add : to the regex character class to correctly match model IDs that contain colons (e.g., Ollama models).

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At src/shared/retry-status-utils.ts, line 17:

<comment>Add `:` to the regex character class to correctly match model IDs that contain colons (e.g., Ollama models).</comment>

<file context>
@@ -0,0 +1,51 @@
+}
+
+export function extractRetryStatusModel(message: string): string | undefined {
+  return message.match(/model\s+([a-z0-9._/-]+)(?=\s+(?:are|is)\b)/i)?.[1]?.toLowerCase()
+}
+
</file context>