Sema is an AI-first semantic language and tooling stack for explicit software intent, drift analysis, and safer intervention in live backend systems.

This repository is not optimized for human-first reading. It is optimized for AI agents that need compressed, canonical semantic context.

Canonical entrypoints in the repository root:
1. `llms.txt`
2. `AGENT_CONTEXT_PACK.json`
3. `SEMA_BRIEF.md`
4. `SEMA_INDEX.json`
5. `AGENTS.md`
6. `README.md`
7. `docs/AGENT_STARTER.md`

Capacity routing:
- small/free AI: `llms.txt` -> `AGENT_CONTEXT_PACK.json` -> `SEMA_BRIEF.micro.txt` -> `SEMA_INDEX.json` -> `AGENTS.md`
- medium AI: `llms.txt` -> `AGENT_CONTEXT_PACK.json` -> `SEMA_BRIEF.curto.txt` -> `SEMA_INDEX.json` -> `AGENTS.md` -> `README.md`
- large/tool-using AI: `llms-full.txt` -> `AGENT_CONTEXT_PACK.json` -> `SEMA_BRIEF.md` -> `SEMA_INDEX.json` -> `AGENTS.md` -> `README.md`

Source-of-truth order:
1. root Sema artifacts
2. CLI JSON outputs
3. official docs
4. official examples
5. raw implementation code

Useful commands:
- `sema resumo <arquivo-ou-pasta> --micro --para mudanca`
- `sema prompt-curto <arquivo-ou-pasta> --curto --para review`
- `sema drift <arquivo-ou-pasta> --json`
- `sema contexto-ia <arquivo.sema> --saida ./.tmp/contexto --json`
- `sema validar <arquivo-ou-pasta> --json`
- `sema verificar <arquivo-ou-pasta> --json`

Do not ask the model to infer semantics that are already available in Sema artifacts.
Do not start from raw source code if `AGENT_CONTEXT_PACK`, `SEMA_BRIEF`, `SEMA_INDEX`, `drift`, or `briefing` are available.
