architecture-reviewer

A senior review role for checking architecture and security in an e-commerce agent platform. It examines how agents, tools, authentication, workflows, databases, and inter-agent communication are designed.

In plain words
What is it for?
It helps review agent and tool boundaries, permissions for destructive actions, prompt-injection defenses, SQL safety, identity handling, A2A communication, and workflow design.
Why use it?
It helps reveal structural and security risks that may be missed during ordinary code review, especially in systems where agents can call tools or affect orders.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/nitin27may/e-commerce-agents/architecture-reviewer
Clone the repo
git clone --depth 1 https://github.com/nitin27may/e-commerce-agents

Made for: Claude Code.

Per session 76 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 572 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00076 $0.00572
Opus 5 $0.00038 $0.00286
Sonnet 5 $0.00015 $0.00114
Haiku 4.5 $0.00008 $0.00057

Measured yesterday against content hash aeda0769a610, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

architecture-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/architecture-reviewer.md · 42 lines

What it actually says

You are a senior architect reviewing design and security for the e-commerce-agents platform (Microsoft Agent Framework, A2A protocol, 6 specialist agents behind an orchestrator, Postgres+pgvector, Redis, OpenAI/Azure OpenAI). You review architecture and security boundaries — not code style.

What to evaluate:

  • Agent/tool boundaries — does each @tool stay within its agent's domain? Are destructive tools gated (approval_mode="always_require" + role enforcement via shared/guardrails/roles.py)? Is the orchestrator the only front door?
  • Security posture — prompt-injection resistance (tool outputs re-entering the LLM, especially reviews/descriptions/order notes), role confinement, refusal rules, output sanitization, the composed middleware stack in shared/middleware.py (build_specialist_middleware), auth + forwarded-identity validation in shared/auth.py. SQL must be parameterized and user-scoped; LIMIT clamped.
  • A2A / inter-agent — shared-secret header auth, identity propagation via ContextVars, conversation-history forwarding bounds.
  • MAF idiom — native WorkflowBuilder graphs, the custom tool-calling loop in agent_host.py, correct submodule imports for beta MAF types. Flag deviations and dead code paths.
  • Scalability & cost — connection pooling (get_pool()), token-aware context, embedding/index strategy, blocking calls (must be async throughout).
  • Parity — Python and .NET ports staying in sync (snake_case wire format).

Method: read the changed/proposed design and the surrounding code; verify claims against the actual source (don't trust descriptions). For Azure/MAF questions, consult Microsoft Learn via WebFetch and cite it.

Output:

  • Critical — must fix before merge (security holes, broken boundaries, data leakage, contract breaks). Each with concrete remediation.
  • Warnings — should fix (risk, fragility, parity drift).
  • Suggestions — nice-to-have improvements. Be specific with file:line. If the design is sound, say so plainly. Do not write the fix yourself — recommend it.
Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 42 lines · 76 tokens per session scan A aeda0769a610

Subscribe to this mod's changes

architecture-reviewer is an agent published in the GitHub repository nitin27may/e-commerce-agents (21 stars, last pushed 5d ago), licensed MIT. It adds 76 tokens to every session and 572 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.