One click. A read-only AI agent audits your project's code quality and streams findings live — then each finding becomes a pipeline task that fixes itself.

Optimize View: AI Deep Analysis

What it is

Deep Analysis is a dedicated view where a single click — AI Deep Analysis — spawns a read-only AI agent that audits your project's code quality and streams findings live into the dashboard. Read-only means it can never change your code — it only reports.

The 5 audit areas

Each run scans your codebase across five categories:

🏛
Architecture & SOLID violations

God objects, tight coupling, and single-responsibility breaks — flagged with the file and the specific principle violated.

Error handling gaps

Unguarded async calls, swallowed exceptions, and missing try/catch around risky I/O that could fail silently in production.

🔒
Security issues

Injection risks, unsafe input handling, and secrets accidentally left in code.

Performance problems

N+1 queries, unnecessary re-renders, and blocking calls sitting on hot paths.

🧹
Dead code

Unused functions, unreachable branches, and orphaned files with zero callers.

How it works

Results are cached per project, so revisiting the Optimize view doesn't re-run the analysis. A 5-minute per-project cooldown prevents accidental repeated — and costly — runs. Analysis runs against the project's knowledge base (the same contracts, variables, and DOM map system described in How AITM Understands Your Project), so findings reference concrete files instead of vague descriptions.

From finding to fix

Each finding can be turned into a pipeline task with one click: Optimize finds the problem, and the 9-step pipeline (architect → code → review → fix → test → e2e → docs → merge, see The Pipeline) fixes it, tests it, and merges it. That makes Optimize continuous AI-driven code health, not a one-off report.

Tips

  • Run Deep Analysis after a batch of merged tasks — fresh findings reflect what actually changed, not stale code.
  • Create tasks from the top 3-5 findings and chain them so they run back-to-back without you babysitting the queue.
  • Respect the 5-minute cooldown — it exists because a full analysis run has a real AI cost, not as an arbitrary throttle.