Agent Claude Code
The definitive read-only authority on the .NET backend — architecture, project structure, every standard/rule, and all canonical patterns. Reviews backend changes (apps/api) for rule adherence and hunts for real bugs, returning severity-bucketed findings. Never edits code.
Agent Claude Code
Read-only frontend reviewer for the Next.js/React app. Validates adherence to the project's own frontend standards and hunts for bugs. Use to review frontend changes under apps/web. Never edits code. (Fully exercised once apps/web is scaffolded.).
Agent Claude Code
Read-only reviewer of the API↔BFF seam and cross-stack concerns — request/response contract parity, ProblemDetails handling, SSE relay, auth/tenancy across the boundary. Use when a change spans apps/api and apps/web. Never edits code.
Agent Claude Code
Adversarially verifies a single architect finding by reading the actual code line by line, returning a real/noise verdict with evidence. Used by the architect-review workflow to pre-filter findings before the main agent's final triage.
Agent Claude Code
Implements an approved implementation plan in the .NET/Next.js codebase. Spawned by the impl-build workflow to build a plan section — edits C# via Serena, builds, and reports files changed. Not for ad-hoc use.
Agent Claude Code
Read-only root-cause investigator for the .NET backend — diagnoses runtime/production issues (errors, anomalies, failures, performance) by combining code analysis, SQL, and telemetry. Produces an evidence-backed root cause + a minimal-fix proposal (a fix-list impl-build can consume). Never edits code.
Agent Claude Code
The backend security authority for the .NET API — audits against OWASP Top 10 (mapped to .NET 10 / Minimal API / EF Core on Npgsql), tenant isolation, secret handling, and EU data residency. Read-only; returns severity-bucketed, verifiable findings. Backend-specific (the frontend has its own auditor).
Agent Claude Code
Writes and runs the tests specified by an implementation plan's exact test list, then reports exact pass/fail/skip counts. Used by the impl-build workflow. Owns test authorship; never weakens a test to make it pass. (Fully exercised once apps/api/tests + an approved test stack exist.).
Agent Claude Code
Read-only validator that checks an implementation against its plan and the project rules and returns a structured pass/fail verdict. Used by the impl-build workflow after the implementer; gates the fix loop. Never edits code.