What it does #

When your connected editor runs a Livecheck, Norma reads the current contents of one file — including unsaved edits — applies the rules that match that file’s language, and returns:

• a clear outcome (clean, a list of issues, or “nothing to check”),
• for each issue: the rule, its severity and impact area, the line, a short description, and concrete fix guidance with good/bad examples.

It’s fast and file-scoped, so your AI assistant can run it constantly as it writes, not just at the end.

When you’d use it — Livecheck vs a full scan #

	Livecheck	Full scan
Scope	One file, right now	Your whole repository
When	While you write, before you commit	On demand, after code lands
Sees unsaved code	Yes	No — reads the branch
Uses your scan quota	No	Yes

Use Livecheck to keep a file clean as you write it. Use a full scan to get your repository’s Production Ready Score and the complete issue backlog.

How to use it #

Livecheck runs through your local AI editor once it’s connected over MCP. You don’t call it directly — your assistant does, in two situations:

• Automatically, right after it creates or changes a file, so it can correct itself before moving on.
• On demand, when you ask it to “check”, “review”, or “run quality on” the file you’re in.

If you use the VS Code extension, you can also trigger a check yourself from the editor and see the results as native VS Code problems. See that page for the editor walkthrough.

Reading the result #

Every Livecheck comes back as one of three clear outcomes — never a silent empty response:

• Clean — “No issues found in this file for the applicable rules.”
• Issues — a list, most severe first, each with the rule, severity, impact area, line and column, a description, fix guidance, and good/bad code examples.
• Not checkable — “No rules available for this file type.” This just means no rule in your plan applies to that language — it isn’t an error.

If a file can only be partly parsed (for example, it’s half-written), Livecheck still returns what it can and tells you coverage was reduced, so a “clean” result is never misleading.

Fixing an issue #

Each issue comes with a remediation action and good/bad examples, so the fix is actionable immediately. Your AI assistant can take that guidance, propose a change, and — once you accept it — record the result back to Norma as a prevented issue: something caught and fixed before it ever landed in your codebase. That’s different from fixing an issue already sitting in your scan backlog, and it’s how Livecheck work shows up in your governance record.

Behaviour & rules #

• Rules are chosen by file type. The file’s extension decides which rules run — .py gets the Python rules, .tsx the TypeScript/React rules, and so on. A file type with no matching rules returns “not checkable”.
• Your plan sets the boundary. Livecheck only ever evaluates rules from tools your tier includes. Rules outside your entitlement are never run — or mentioned.
• Deterministic rules only. Livecheck runs the fast, pattern-based rules (the same engine behind your regular findings). Analysis that needs the whole repository — architecture/SOLID-style checks — isn’t part of a single-file Livecheck; those surface in a full scan and on the AI Findings tab.
• Works with or without a linked repository. Rule selection never needs a repo, so Livecheck works even on web-based editors with no local files. Linking a repository just lets the results be attributed to it.

Limits & notes #

• One file per check. No batching.
• File size cap: 1 MB. Larger files are refused with a clear message.
• Up to 50 issues per check, most severe first, with the total count shown when there are more.
• Rate limits: 30 checks per minute per user and 5,000 per day per organisation, shared across your editors and separate from your scan quota. If you hit a limit you get a calm message — “Keep working and retry in a moment.” — and nothing auto-retries.
• Your code isn’t stored. File contents and file names are never saved or logged. Only metadata — language, which rules ran, the outcome — is recorded.

Frequently asked questions #

Does a Livecheck count against my weekly scans? #

No. Livecheck has its own limits and never touches your scan quota.

Why did a check come back “not checkable”? #

No rule in your plan applies to that file’s language. It’s not a failure — there was simply nothing to run.

Can Livecheck find architecture or design problems? #

Not on a single file — those need the whole repository. Run a full scan and check the AI Findings tab for design-level analysis.

Will Livecheck edit my code? #

No. It returns issues and fix guidance; your AI assistant proposes the change and you decide whether to apply it.

Related #

• MCP and local AI editors — connect your editor so it can run Livecheck.
• Norma MCP server on GitHub — the live_check tool and the full setup reference.
• Set up the VS Code extension — trigger checks and see results inside VS Code.
• Running a scan — the whole-repository counterpart to Livecheck.
• AI Findings — design-level analysis that a single-file check can’t do.
• Tiers and quotas — what your plan includes.