U130 Phase 2 — Static Code Reviewer check on editor buffer

[LosAICode]

Parent: #211
Plan: U130-CMS-Editor-Diagnostics-Plan.md § Phase 2
Depends on: Phase 1 (#211 subtask)
Effort: ~3h

Scope

Run `CODE_REVIEWER_SYSTEM_PROMPT` against the current editor buffer and return findings.

• New module: `modules/services/magnus_diagnostics.py`
• Function: `run_code_reviewer_on_buffer(html: str, script: str, project_id: str) -> list[Finding]`
  • Composes buffer with `<script>` wrapper mirroring U129 render-time injection
  • Calls `_call_llm` with `CODE_REVIEWER_SYSTEM_PROMPT` at `code_reviewer` model role
  • Parses returned JSON checklist into `Finding` dataclass: `id`, `severity`, `title`, `detail`, `fix_hint`
• Cache: in-memory LRU keyed on SHA256(html+script), 5-min TTL (env: `MAGNUS_DIAGNOSTICS_CACHE_TTL`)
• Credits: deduct via `_calculate_magnus_credits(module='magnus_code_reviewer')`
• Failure: return a single warning Finding (`"Code reviewer unavailable"`) — never 500

Acceptance

POSTing the pre-U129 TaskMaster `/tasks` HTML returns findings including:

• `code_reviewer:12` (dead dashboard from U129 check Fix paths to hooks and run git in project root. #12)
• `code_reviewer:9` (font-family without matching link from U128F check Dev #9)

Re-posting identical buffer within 5 min = zero LLM cost (cache hit).

[LosAICode]

Created in wrong repo — moving to LosAICode/NeuroGenServer