Your first architecture session

You will finish with a corrected current-state map, evidence coverage, an attention heat map, and one selected next investigation. Start with this exact request:

Assess architecture and provide an action plan in survey mode, stopping after the attention heat map without creating files.

The survey reads the repository but does not run its code or change it. You make two decisions: whether the conceptual map is right, and which hotspot deserves deeper evidence.

Prerequisite: the architect pack is installed.

1. Read the assessment charter

The first response should bound the repository or system, name baseline/understanding as the primary intent, select survey mode, and list what the agent can read. It should say whether it found an in-repo or private enterprise knowledge surface.

You should not see a score, defect list, or rewrite plan yet.

2. Correct the conceptual map

The agent inspects documentation, source, tests, manifests, CI/CD, deployment definitions, schemas, configuration, operations files, and current local history where available. It then describes context, runtime/deployables, modules/capabilities, data, interactions, delivery/operations, and trust or identity boundaries.

At Map checkpoint, compare the model with what you know. Correct a boundary in plain language:

The worker and API are separate deployables, but they share the same database.

If the map is accurate, reply continue.

The assessment must record your correction or acceptance before it focuses the investigation.

3. Choose where to look deeper

Next you see an attention heat map by system area. It keeps consequence, pressure, coupling or concentration, verification weakness, operational/data/security exposure, and confidence separate. The legend should say that heat selects drill-down priority; it is not proof of a defect or severity.

Each proposed hotspot should explain its architectural role, raw signals, counter-evidence, affected journey or quality scenario, unknowns, and the next check. At Focus checkpoint, redirect the set or reply continue.

Because this tutorial asked for survey mode, the agent now stops. It should call the hotspots hypotheses, not completed findings, and report:

Result: chat only; no file was created.

4. Verify the result

Your completed survey contains:

• a bounded assessment charter;
• a conceptual model that reflects your correction;
• a status for every required evidence surface, including missing ones;
• an attention heat map with raw dimensions;
• bounded hotspot cards and recommended drill-downs;
• coverage limits and the next human decision.

It fails this tutorial if it returns only a folder map, dependency list, compliance checklist, code-smell inventory, or confident action plan without investigation.

5. Continue with one hotspot

Pick a hotspot and ask for standard depth:

Continue in standard mode. Investigate H-2, trace the normal, side-effect, and failure/recovery paths, then propose only actions supported by findings.

The agent asks before any executable check, private knowledge query, runtime access, experiment, or write. When the current-state evidence points to a future-state choice, move to Shape an architecture concept. If you want a durable normative foundation rather than an assessment, follow Create and use your reference.md as a separate journey.

What you have now

You have a chat-only survey with a corrected current-state map, evidence coverage, and selected hotspots. Choose one hotspot for standard-depth investigation, or begin a separate reference-architecture journey.