LLM MODULE FEATURE
LLM Module Overview
The LLM module is not "chat pasted onto a framework". It gives your project a governed way to expose its own commands as AI-usable skills.
Feature Guide
A quick orientation block that answers the essential questions: what this feature does, how it works, why it matters, and the key concepts behind it.
What this does
What `semitexa/llm` adds to the framework and how your project can expose its own CLI skills to the assistant.
Project Extension
Turn your own commands into skills
The point of `semitexa/llm` is not only to ship one assistant command. The point is to let each project expose its own safe command surface to the assistant through explicit skill metadata.
Your commands stay primary. A skill starts as an ordinary Semitexa console command owned by your project or package.
Metadata makes it usable. `#[AsAiSkill]` adds summary, use/avoid guidance, confirmation mode, and argument policy.
The manifest stays reviewable. Only the commands you expose become part of the skill manifest the assistant may use.
| Command | Purpose | Why it matters |
|---|---|---|
#[AsCommand(...)] |
Define the real console command first. | The assistant extends your CLI; it does not replace it with a second hidden automation layer. |
#[AsAiSkill(...)] |
Attach AI metadata to the command you want to expose. | This is where you define risk, confirmation, exposed arguments, and usage guidance. |
bin/semitexa ai:skills --json |
Verify what the assistant can actually see. | Treat the manifest as the review surface for your project-specific skills. |
Minimal custom skill
#[AsCommand(name: 'reports:rebuild')]
#[AsAiSkill(summary: 'Rebuild reporting snapshots', argumentPolicy: 'allowlisted')]
final class RebuildReportsCommand extends Command {}Review the exposed manifest
bin/semitexa ai:skills --jsonVerified against Semitexa Ultimate 2026.09.19.1020
LLM Module Overview
The semitexa/llm package gives a Semitexa project a governed AI surface: your own console commands can become discoverable skills instead of living behind ad-hoc prompt instructions.
How it works
You keep writing normal console commands. When one should be usable by the assistant, you add #[AsAiSkill] next to #[AsCommand], choose the risk and argument policy, and let the manifest expose it to the LLM layer.
Why this matters
This keeps AI integration inside the framework contract. Teams extend the system by adding real commands and explicit metadata, not by teaching a model private tribal knowledge about the project.