800 AI

Learn how AI fits on top of Stackpress after the normal app workflow is familiar. AI can connect to a project, expose tools, generate artifacts, and guide app-building phases, but every AI result still needs source-backed verification.

Previously: The previous course, Import / Export, covered moving project data in and out of Stackpress. This course shifts to AI-assisted work, where the risk is not only whether the tool can answer, but whether the answer matches the current Stackpress source.

800.1. Why AI Comes Last

AI help is useful only when the developer can judge the output. If you cannot recognize a valid scaffold, schema, plugin boundary, generated file, or runtime check, a confident AI response can lead you away from the project instead of toward it.

Think of AI as a fast teammate who still needs the map. The map is the Stackpress source, the scaffolded app, the generated output, and the evidence from commands or runtime behavior.

800.2. What This Level Covers

The first part covers the implemented Stackpress AI package. 810 MCP explains how Stackpress creates MCP tools from config and generated registries, while 811 stdio Transport shows the local transport entrypoint that connects those tools to an agent process.

The middle part is deliberately cautious. 820 Artifacts gives review rules for AI output, and 830 Hooks through 832 Transform Hooks separate confirmed source behavior from areas that still need owner confirmation before they can become full lessons.

The last part covers the repo skills used by agents. 840 Skills through 847 Verification Skill explain how app discovery, coordination, scaffolding, idea authoring, plugin routing, and verification hand work to each other without turning a whole project into one vague prompt.

800.3. The Evidence Rule Still Applies

AI output should not be trusted because it sounds complete. It should be trusted only when current files, commands, generated output, runtime behavior, or review evidence prove the result matches the Stackpress workflow.

For example, an agent can draft a plugin plan that says a feature belongs in runtime code. That claim is not useful until the plugin lane is checked against schema.idea, generated output needs, route/view ownership, and the verification evidence for the phase being changed.

800.4. How To Read This Level

Read the MCP pages first because they describe the connection between Stackpress and an external agent. Then read the artifact and hook pages as review boundaries, especially where the current source is not complete enough to support confident how-to instructions.

Finish with the skills pages because they describe the safest way to use AI across a Stackpress app build. The goal is not to memorize every skill name, but to learn which phase owns which decision and what evidence should exist before moving on.

Learning checkpoint: Before moving on, make sure you can explain AI as an assistance layer, not a replacement for Stackpress source knowledge. You should also be able to name the evidence you would check before accepting an AI-generated scaffold, schema change, plugin change, or verification claim.

Next course: Continue with MCP. That course starts with the implemented tool connection layer before moving into higher-level AI workflows.