Mobile Engineering Agents

Appearance

Workflow: Investigate a Bug ​

Systematic debugging — find the root cause with evidence before fixing. Pairs with the systematic-debugging mindset.

Objective ​

Reproduce, diagnose the root cause (not the symptom), fix it, and prevent regression with a test.

Inputs ​

  • Bug report: steps, expected vs actual, environment, frequency, logs/crash report.

Outputs ​

  • A reliable reproduction, root-cause explanation, minimal fix, and a regression test.

Step-by-Step Process ​

  1. Reproduce reliably — establish exact steps and a failing case. If you can't reproduce, gather more data (logs, device, OS, account state).
  2. Write a failing test that captures the bug at the lowest possible level (unit > UI).
  3. Form a hypothesis about the root cause; locate the responsible component.
  4. Confirm with evidence — logs, breakpoints, Instruments, the memory graph — don't guess.
  5. Fix the root cause (Refactoring/relevant specialist), keeping the change minimal.
  6. Verify the failing test now passes and no others regress.
  7. Review (Code Reviewer) the fix + test.

Validation Steps ​

  • The regression test fails before the fix and passes after.
  • The original reproduction no longer occurs.
  • No new failures introduced; the fix addresses cause, not symptom.

Failure Scenarios ​

  • Can't reproduce → add logging/telemetry; narrow environment; consider a flaky/timing issue.
  • Symptom masks cause → keep tracing inward; don't patch the surface.
  • Heisenbug (timing/concurrency) → suspect data races; use Thread Sanitizer.
  • Fix breaks other tests → reassess the hypothesis.

AI Agent Instructions ​

  • Do not propose a fix before reproducing and identifying the root cause with evidence.
  • Always add a failing-first regression test at the lowest level.
  • Prefer the smallest change that fixes the cause; avoid opportunistic refactors in the same PR.
  • For concurrency bugs, enable Thread Sanitizer and review actor/isolation boundaries.

Acceptance Criteria ​

  • [ ] Reliable reproduction documented.
  • [ ] Root cause identified with evidence.
  • [ ] Regression test fails before, passes after.
  • [ ] Minimal fix; no new regressions.
  • [ ] Reviewed and approved.