Skip to main content

AI Diagnosis

A page for generating and reading comprehensive diagnostic reports that analyze your account's overall operational state from AWS-native data.

AI Diagnosis report

Features​

Run a diagnosis​

  • Diagnosis depth (Tier): choose the scope of analysis.
    • Light: a fast check of core items only (currently identical catalog/token budget to Mid)
    • Mid: a balanced standard diagnosis (default)
    • Deep: a broad analysis covering 15+1 sections (16 rendered). Only Deep lets you additionally pick a model (Sonnet / Opus), and choosing Opus shows an accompanying cost note.
  • Report language: choose the report body language (한국어 / English / 中文 / 日本語). It defaults to the current UI language, and switching only the language within the same hour starts a new report.
  • Press the 진단 실행 (Run diagnosis) button and the report is produced asynchronously by a background worker.

Progress indicator​

  • While generating, an mm:ss elapsed timer and a per-section completed/pending checklist grid accompany the progress bar; a completed report shows a section count · duration · report id stats bar (duration omitted on older reports without a recorded finish time).
  • While the report is being produced, a live progress bar shows the data collection → section analysis → assembly stages.
  • The report opens automatically once it is done.

Report list​

The left sidebar lists your recent reports — completed rows carry inline MD/DOCX download links so you can download without opening the report:

  • An auto-generated title, the id, depth (Tier), model, status, and creation date (KST)

Report body​

  • The body renders as collapsible section cards with severity icons and a sticky table of contents (TOC) sidebar on the right.

  • Use the top buttons to export as MD / DOCX / PDF, or open Print view — a new tab with a white A4 layout (cover, numbered TOC, per-section page breaks) for direct browser Print-to-PDF.

  • Manual report handoff is available for succeeded or partial reports with readable Markdown. Preview, copy or download selected excerpts with omission/truncation notices. Template headings, notices and checklists are English; excerpts preserve the source report's language. Transfer is manual; no publishing or external agent invocation.

Insight badges​

  • Invariant assessment coverage separates total, assessed, passed, violated and unassessed counts. Unassessed reasons appear in the UI, report body and exports; unassessed results or an empty violation list do not establish health or improvement. Historical reports without recorded coverage show assessment unavailable.
  • The current collector path still needs relationship-resolution and encryption-aggregate integration, so its six invariant kinds remain unassessed. Read these separately from observations in other diagnosis sections.
  • An intended-vs-actual / change insights badge row summarizes invariant violations and changes versus the previous report.
  • The Intent (invariant candidates) panel lets you propose, accept, and reject candidates. (Admin-only; read-only for everyone else.)

Title, tags, and delete​

  • The report owner or an admin can edit the title inline, add and remove tags, and soft-delete the report.

Scheduled diagnosis & notifications​

  • Scheduled diagnosis: besides the cadence (weekly/biweekly/monthly) you can pick a weekday (weekly/biweekly), a day of month 1–28 (monthly), a run hour (KST), and the report language; the next run and last run times are both shown. Unset fields keep the interval-only behavior.
  • Diagnosis mailing list: admins can, in addition to adding/removing subscribers, press Send test to deliver one test email to every confirmed subscriber and verify delivery. The email notification switch at the top of the panel pauses report/digest emails without a deploy (admin-only) — reports completed while paused are dropped from email (not re-sent on resume; a pause shorter than the ~15-minute digest cadence may drop nothing — the flag is checked per run), and the test-send button still works while paused (delivery-path verification). The same switch and subscriber list also govern the compliance benchmark completion mail on the shared topic.

How to use​

  1. In the sidebar, click AI Operations > AI Diagnosis.
  2. Select a diagnosis depth (Light / Mid / Deep).
  3. If you chose Deep, optionally specify the model (Sonnet / Opus).
  4. Choose the report language (default: the current UI language).
  5. Press the 진단 실행 (Run diagnosis) button to start generation.
  6. Watch the stages on the progress bar; when it finishes, review the report that opens automatically.
  7. Navigate sections with the table of contents sidebar on the right (section cards collapse/expand), and export as MD / DOCX / PDF from the top.
  8. Edit the title and tags or delete the report if needed (when you have permission).

Tips​

Duplicate-run protection

If you run the same conditions (depth, model, and language) again within the hour, the existing report is returned instead of generating a new one, along with a duplicate-run notice.

Permissions

Editing the title, tags, and deletion are available only to the report owner or an admin. Proposing, accepting, and rejecting invariant candidates is admin-only and is shown read-only to other users.

Read-only analysis

AI Diagnosis only observes and analyzes AWS resources — it never modifies them. Reports are for analysis purposes only.

AI analysis tips​

After reviewing a report, you can ask follow-up questions from the floating button or the AI Assistant page, for example:

  • "What is the riskiest item in this diagnosis?"
  • "What changed compared to the previous report?"
  • "Explain the cause of the invariant violations shown."
  • AI Assistant - follow-up questions about report results
  • Async Jobs - check the progress of diagnosis generation