Works with the AI Agents you already use
Agent Architecture: Should This Be an Agent?
Decide if your problem needs an LLM agent at all, pick the simplest pattern that works, and record it as an Architecture Decision Record.
Free
Agent Architecture: Should This Be an Agent?
Example session with this skill installed
We get ~3,000 supplier invoices a month as PDFs. A clerk extracts supplier, invoice number, date, VAT and total, checks them against the purchase order in our ERP and flags mismatches. My CTO wants a multi-agent system with a planner, an extraction agent, a validation agent and a supervisor. Is that the right architecture? Write the decision in adr.md.
- Read your context and instructions
- Compiled the agent architecture
Recommendation: not multi-agent. Rung 1 + rung 0 hybrid: one structured-output extraction call per invoice, deterministic PO matching in code against the ERP (read-only tool), and human review only for mismatches. The steps are fixed and enumerable, so an agent adds cost and variance without benefit. adr.md: requirements table (≥98% field accuracy on a 200-invoice eval set, ≤$0.02/invoice), tool table (ERP lookup: read, no approval; ERP write-back: reversible, human approval above tolerance), budgets, rejected alternatives (4-agent system: coordination overhead, untestable), launch preconditions.
Connects securely to your tools. The creator never sees your data.
About this skill
Most failed agent projects are over-built, not under-built: a free-roaming multi-agent system where a three-step workflow would have been cheaper, faster and testable.
This skill runs a short, evidence-based design conversation and ends in a written decision:
- Frame the job: inputs/outputs, volume, latency and cost budgets, cost of a wrong action, systems touched, success metric.
- Four tests (complexity, value, viability, recoverability) that must all pass before autonomy is justified.
- A 9-rung ladder from deterministic code → single call → prompt chain → router → fan-out → orchestrator-workers → evaluator-optimizer → autonomous agent → multi-agent, with the typical failure of each when mis-chosen.
- Runtime choice: manual loop vs SDK tool runner vs agent SDK (Claude Agent SDK, Google ADK, LangGraph) vs managed agent service.
- A filled ADR (template included) with tools, read/write classification, approval points, budgets, success metric and launch preconditions.
Includes a pattern reference with pseudo-code and the checkpoint each pattern needs, plus a bundled eval case.
Part of the Production AI Agents series. Once you know what to build, go deeper with Agent Guardrails (security), Agent Evals (measurement), MCP Server in Production, and Text-to-SQL Data Agent.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Does not design tool schemas, prompts, the run loop or security controls in depth; those are covered by the companion skills. Does not write implementation code.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 1 day ago
- Free with an Agensi account
Requires
Nothing to install.
Permissions
No special permissions declared or detected