# Showcase

This page is the shortest path for understanding why Agent Spine is more than a repository scaffold.

## The Claim

Agent Spine turns a repository into an agent operating layer:

- instructions are visible in `AGENTS.md`
- native Codex behavior is committed under `.codex/`
- repeatable workflows live under `.agents/skills/`
- Claude Code compatibility is available through the optional `.claude/` adapter
- downstream adapters stay optional
- verification checks the contract before people depend on it

## Five-Minute Tour

1. Read `docs/PRODUCT_THESIS.md`.
2. Read `docs/COGNITIVE_OPERATING_SYSTEM.md`.
3. Inspect `.codex/hooks/pre_tool_use_policy.py`.
4. Inspect `.agents/skills/production-change/SKILL.md`.
5. Run `./scripts/verify-template.sh`.

## What To Demo

Use this sequence:

```bash
./scripts/bootstrap-template.sh sample-fullstack fullstack-product drop-claude
./scripts/smoke-test-installer.sh
./scripts/verify-template.sh
```

Then show:

- the generated README frames template checks as temporary scaffolding
- `fullstack-product` carries real ownership and verification boundaries
- the installer preserves an existing README and AGENTS contract
- the hooks block direct secret reads while allowing safe exclusion globs

## Strongest Evidence

- `docs/PILOT_INFERENCE_GUARDIAN.md`: real downstream adoption findings
- `docs/QUALITY_MODEL.md`: measurable quality rubric
- `scripts/smoke-test-bootstrap.sh`: positive and negative regression tests
- `scripts/smoke-test-installer.sh`: merge-first adoption test