Bug Fix Agent

Triage-to-fix is slow and context-heavy. Agents shrink mean-time-to-resolution by systematically exploring the codebase and running the test loop that engineers repeat manually.

A detective with a lab: gathers evidence (logs, tests), forms a hypothesis, runs experiments (repro, patch), and presents a case file (PR) — not a guess on the first clue.

Scenario

Sentry reports `NullPointerException` in checkout. The agent traces to `CartService.applyCoupon`, adds a null guard test, fixes the bug, runs `npm test`, and opens PR #842 with linked Sentry issue.

Solution

In Coding Agents, apply Bug Fix Agent to this scenario: Sentry reports `NullPointerException` in checkout. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Bug Fix Agent (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Bug Fix Agent happens in the code.

1async function bugFixLoop(errorReport) {2  const context = await gatherContext(errorReport); // stack, files, recent commits3  for (let i = 0; i < MAX_ITERATIONS; i++) {4    const hypothesis = await planner.diagnose(context);5    const patch = await coder.proposePatch(hypothesis);6    const result = await sandbox.runTests(patch);7    if (result.passed) return { patch, testsAdded: result.newTests };  # return the result8    context.failures.push(result.output);9  }10  return { status: "needs_human", context };  # return the result11}

• Treating Bug Fix Agent as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for bug fix agent

Quick recap — the most important points from this module.