Fix Inngest agent routing

[Jackson57279]

Summary by cubic

Fixes Inngest coding agent routing to use stable tool-calling models and a deterministic retry chain, improving AgentKit reliability and failover behavior.
Centralizes model selection logic and adds better error reporting and metadata for debugging.

• Bug Fixes

  • Normalize tool-calling model (strip :nitro) and fallback from unreliable tool models to a safe default.
  • Replace OpenRouter implicit fallbacks with an explicit retry chain; retry only on provider errors.
  • Persist agent failures to Convex with requested model, tool-calling model, retry chain, and error details.

• Refactors

  • Extract model planning to src/inngest/model-routing.ts with resolveCodingModelPlan and getToolCallingModelForAgent.
  • Add unit tests for routing behavior in tests/inngest-model-routing.test.ts.
  • Include manager model, retry chain, and attempt count in fragment metadata for better observability.

^{Written for commit 1215e9d. Summary will update on new commits.}

[vidoc-ai-security-engineer]

Vidoc Security Report

Great! No high-severity issues detected.

Details

{ "scanId": "019dc876-2f94-7087-acca-ee7e9a54cf92", "codebaseId": "019d74e8-a953-70b7-a19d-914ab632ae39", "installationId": "019d74e8-a4a8-772f-bdb7-b896739e2b23", "internalPullRequestId": "019dc876-2f79-7468-824f-00949d36f395" }

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

📝 Walkthrough

Walkthrough

Model routing logic is extracted from functions.ts into a new centralized model-routing.ts module that defines fallback chains, classifies models as reliable/unreliable, and exports functions to resolve tool-calling models and comprehensive coding model plans including manager models and retry sequences.

Changes

Cohort / File(s)	Summary
Model Routing Module src/inngest/model-routing.ts	New module introducing getToolCallingModelForAgent, resolveCodingModelPlan, and CODING_FALLBACK_CHAIN constant. Classifies models as unreliable for AgentKit tools, determines manager model eligibility, and builds deduplicated retry model sequences.
Functions Integration src/inngest/functions.ts	Refactors coding-agent model selection to use resolveCodingModelPlan instead of getToolCallingModelForAgent. Removes openrouterAgentModel fallback chain construction, updates fallback orchestration to accept explicit retry lists, augments run results with managerModel, attemptedModels, and attemptCount, adds persistAgentFailure function for failed fragments, and removes tool_choice: "required" from agent configuration.
Model Routing Tests tests/inngest-model-routing.test.ts	New Jest test suite validating :nitro suffix stripping, dedicated manager model selection for unstable tools, and deterministic retry model sequencing.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Poem

🐰 Hops through models with purpose clear,
Fallback chains standing ever near,
From functions wide to routing tight,
The agent picks its tools just right.
No more scattered, now aligned—
A cleaner path through code designed! 🌱

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'Fix Inngest agent routing' directly and clearly describes the main change - refactoring the agent model selection and fallback logic into a centralized routing module.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/fix-inngest-agent-routing

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1215e9d5e6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Require tool calls in coding agent configuration

The coding agent is now created with the default tool_choice behavior (auto), so it can legally return a plain <task_summary> without invoking createOrUpdateFiles. In this workflow, the router stops once a summary is present and the run is treated as successful, which can produce “completed” fragments with no code changes. Restoring tool_choice: "required" (or an equivalent guard) is needed to preserve the prior guarantee that code-writing tasks actually execute tool calls.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve fallback for post-processing model calls

Both post-processing agents now call anthropic/claude-haiku-4.5 without any fallback chain, so a transient provider outage/rate-limit in title or response generation causes the whole job to fail after coding has already completed. Previously these steps had fallback routing; removing it regresses reliability for users in the exact failure mode this patch is trying to address.

Useful? React with 👍 / 👎.