A L M A S ALMASHelp Open in the app

28. Reporting bugs (and requesting features)

ALMAS has two ways to file a report, and both feed into an AI-assisted triage pipeline. Use whichever fits the moment.

Bug or feature? The report modal has a type selector. Leave it on Bug to report something that's broken, or switch it to Feature to request something new. Both create a task and both get an AI-written plan attached as a note shortly after — for a bug it's a fix plan, for a feature it's an impact-and-approach plan with a suggested size and priority. When you submit, you'll see a quick ✓ Task created confirmation; the plan note arrives a moment later on the task itself.

1. The keyboard shortcut: N then B

Press N, then B (two separate key presses, not chord). A bug-creation modal opens.

You'll typically use this when:

  • You notice something looks wrong but it didn't trigger an error popup.
  • You want to file something proactively (a confusing label, a slow page, a layout glitch).
  • You're already on the page where the issue lives and want the URL captured automatically.

Fill in:

  • Title — a one-line description of what's broken. The clearer the title, the better the AI triage's first guess at the cause.
  • Priority — how urgent it is for you. (Triage runs regardless of priority.)
  • Description — anything you'd want to tell a developer. Include "expected vs actual" if you can.

Click Save. The bug is filed instantly.

2. The error popup's "Report" button

Whenever the system fails an action — saving an invoice, loading a list, calling an external service — a popup appears with the error details and two buttons:

  • Dismiss — close the popup.
  • Report — file a bug from this exact error.

Clicking Report auto-creates a bug pre-filled with the error details, the page URL, and a snapshot of what just happened (no need to type a description). Use this when you want a developer to look at exactly the failure you just hit.

What gets captured automatically

In both flows, the system silently attaches a context bundle to the bug:

  • The page URL and the route you were on.
  • Your last few page navigations (so we can see what you were doing right before).
  • Your last few API calls (with passwords/tokens stripped — see "Privacy" below).
  • Any browser console errors from the same session.
  • Browser and viewport info (so we can reproduce on the same device class).
  • The error itself, when filed from the error popup.

You don't need to do anything to enable this — it just works.

What happens after you file

Within roughly a minute (depending on queue load), an AI triage agent runs. It:

  • Reads the context bundle you attached.
  • Pulls matching system logs from around the time you filed.
  • Reads the relevant code paths.
  • Writes a fix plan as a note on the bug. The plan covers a summary, reproduction steps, the suspected root cause, a proposed change, and a test plan — formatted so the development team can act on it immediately.

You'll see the plan attached as a note titled "AI Triage Plan" when you open the bug task.

Duplicates and rate limits

If you happen to report the same issue twice (same URL, same error), the second report is automatically linked to the first instead of running the AI again. You'll see a note titled "AI Triage: Linked to bug #N" on the duplicate.

There's a per-user rate limit (default: 5 AI triages per hour). If you hit it, your bug still gets created — you just won't get an automatic plan for that one. A note titled "AI Triage: Rate limited" is attached so the developer knows why the plan is missing.

Privacy

The context capture is designed to never leak credentials:

  • HTTP headers like Authorization, Cookie, and any header containing token, secret, key, session, or auth are replaced with [REDACTED] before being sent.
  • Body fields whose name matches password, secret, token, key, cvv, ssn, or pin are redacted the same way.
  • Response bodies larger than 8 KB are truncated, with a marker indicating the original size.

If you're concerned a particular bug might contain sensitive data, file it via the keyboard shortcut and skip the description rather than relying on the auto-captured response bodies.

What you should NOT do

  • Don't file the same bug repeatedly hoping for a faster response — the dedup logic links them, but it doesn't accelerate the queue.
  • Don't include passwords, secrets, or customer PII in the title or description — those fields are not auto-redacted.