Architect

Your agent can assess the architecture you have, design the one you need, draw it, and review it.

Install

You need agentbundle first. Install it once for your user account:

agentbundle install --pack architect

In Claude Code you can install it as a plugin instead:

claude plugin install architect@agent-ready-repo

What your agent can do

Assess a repository: map the system, show where its evidence is thin, drill into hotspots, and propose action waves in dependency order.
Design a system, subsystem, or change from a one-page concept, with the rationale beside each model.
Draw Mermaid diagrams for flows, sequences, deployments, data models, and roadmaps, with an optional editable draw.io file.
Review an assessment, design doc, diagram, RFC, or ADR and return a verdict with findings ranked by severity.
Write a solution reference architecture for a coordinating repo with no app code.
Turn a design into a physical architecture and bill of materials (BOM) covering cloud services, container images, open-source libraries, hardware, network connections, and licences, with a delta BOM for a change.
Check a design against well-architected lenses, including cloud and agentic workloads.

Skills

architect-assess — Maps the implemented system, shows evidence coverage and attention hotspots, and proposes action waves. Read-only by default.
architect-design — Shapes a concept first, then writes a design doc matched to the scope of the change. Also writes a solution reference architecture and, on request, a physical architecture and BOM.
architect-diagram — Draws a Mermaid diagram chosen from what you ask for, and checks it before it shows you.
architect-review — Critiques an architecture artifact and gives a verdict with severity-tagged findings.
architecture-lenses-reference — An internal knowledge router for the others. You never invoke it yourself.

Subagent: design-reviewer is a read-only reviewer that works in a fresh context, so it does not grade its own work.

Works with

Works with Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Kiro IDE, and Kiro CLI. Add Core to save a design where your repository keeps its architecture records, and to reconcile it after the change ships. Your reference architecture guides each design. Product Engineering offers it at some shaping steps when both are installed. Guides that use this pack: Design the architecture and Assess a repository.

Questions

Does an assessment change my repository? No. It reads documents, source, tests, CI, and local Git history. Saving a file is optional, and you approve it.

Does it run my build, tests, or deploys? No, not without asking. Runtime access and experiments need separate approval.

What if I have no reference architecture? The design says so in its concept. It offers Core’s adapt-to-project or init-project to set one up when Core is installed. For a solution repo with no app code, it authors a solution reference architecture itself in its solution reference mode.

Does the BOM include prices? Each line is priced with its source, date, and region, unpriced when no grounded price exists, or no direct cost. Unpriced lines are counted and left out of the total. No price comes from model memory, and prices are in USD unless you name another currency.

What does a BOM line identify? Its kind and the exact item. A cloud service shows its plan code (SKU) and tier. A container image shows its registry address. An open-source library shows its package URL, a standard name that includes the version, and its licence. A node shows its hardware model and spec. A network link shows its endpoints and bandwidth. A commercial licence shows its product and edition. A design with no deployment section gets no BOM until the design names what runs where. If the physical choices contradict the design, the agent raises it as a design finding.

Can it work without Core? Yes. Without Core it returns the result in chat or saves to a personal workspace you choose.

Details

Scope: user (default), repo · Version: 0.15.19 · Source: packs/architect