Fixed Agents Debugging Interview

Frequency: Reported

Browse the sanitized 2026 starter repository.

The starter repository contains the complete client, server, tests, bug tickets, and pull-request patch. Its embedded Git metadata, editor settings, and packaging-only cleanup/build hooks were removed; the mock pull-request author was anonymized. The client, server, tests, bug tickets, and patch were otherwise preserved.

Project context

Fixed Agents are AI assistants that workspace members can trigger to summarize pages, translate content, or generate new material. Triggering an agent creates a run that enters a processing queue.

All interview code is under src/. The client/ is stated to be bug-free and exists to help inspect the queue. The server/ contains the agent queue system being debugged. It is a mock server and sends no actual network requests.

middleware/actions.ts       api.ts                  queue.ts
  │                          │                        │
  │  runAgent(user, ...)     │                        │
  │─────────────────────────>│  authorizeAndEnqueue() │
  │  (permission pre-check)  │─────────────────────>  │    enqueue(...)
  │                          │  (auth check)          │──> AgentRunner
  │                          │                        │        │
  │                          │                        │     onProgress
  │                          │                        │     onComplete

• middleware/actions.ts pre-checks permissions.
• api.ts performs the authoritative permission check.
• queue.ts manages the pending → running → complete lifecycle.
• processor.ts dispatches runs to summarize, translate, or generate handlers.

Setup

Prerequisites are Node.js v22, npm, and Git.

nvm use
npm install
npm run dev
npm test

The supplied baseline has failing tests.

AGENT-401: Creative Agent runs never finish

Status: Open
Priority: P1
Component: Agent Queue

Creative Agent runs stay stuck in running and never transition to complete or failed. Summary Agent and Translate Agent runs finish normally.

Steps to reproduce:

1. Enqueue a Creative Agent run (generate).
2. Wait for the run to progress.
3. Observe whether it ever leaves running.

The console reports:

Uncaught (in promise) Error: Cannot process agent run with no messages

The ticket suggests running npm run test:watch processor. The run should eventually reach complete, or failed if something goes wrong, and the tests should pass.

AGENT-512: Queue API returning more errors than expected

Status: Open
Priority: P1
Component: Agent Queue / Permissions

The queue API is returning more error responses than expected, and failed runs are clogging the queue. The system appears to be making unauthorized requests. npm test src/server/tests/permissions.test.ts passes. A new user role named external was recently added and may be related.

AGENT-738: Cancelling a run leaves the next run stuck

Status: Open
Priority: P0
Component: Agent Queue

When a run is cancelled, the run enqueued immediately after it never finishes and remains running.

Steps to reproduce:

1. Enqueue run A, then enqueue run B immediately afterward.
2. Cancel run A while it is active.
3. Observe that run B does not complete.

The ticket asks the candidate to get npm run test:watch queue passing. Cancelling A should affect only A; B should still run and complete normally.

PR #52 review: Retry configuration for agent runs

A teammate submitted a patch adding configurable retry behavior for failed agent runs. The proposed implementation:

• adds retryCount to AgentRun;
• adds retry logic, defaults, exponential backoff, and a timeout guard to agent-runner.ts; and
• adds an agent-runner.test.ts suite.

The stated defaults are three retries, a one-second base delay, and a 30-second timeout. The candidate is asked to apply and review the patch in an IDE diff view:

npm run pr:apply

The sanitized repository supports reversing the patch with:

npm run pr:reset

No supplied fixes for the three bug tickets or supplied review answer for PR #52 were included.