id: SMH-PHASE-A0RPUJ_Q
type: phase
title: "Prototype"
label: proto
Prototype
Operating posture
Prove the plugin seam with real consumers.
Grow the smith:plugin world organically from working verticals; revisit names, shapes, and structure freely.
Gates and session safety laws never relax.
Priorities
API discovery over API stability: signatures earn their shape from real consumers.
Dogfood parity: built-ins load through the same sandbox and effects.
Working verticals over breadth: one vendor end to end beats three half-done.
Fast plugin loop: scaffolding and dev-links keep edit-run cycles short.
Decision defaults
Compatibility: none owed before 1.0; breaking changes are free, deprecation ceremony unnecessary — but the additive-only and world-versioning design stays in the spec so the plan for compatibility exists.
Migrations: plugin authors get no upgrade path pre-1.0.
Stability: host internals and world shape may churn without notice.
Quality: zero-warning, clippy-pedantic, test gates always on.
Experimentation: lives in plugins/ and prototypes, never in core crates.
Debt: naming and glossary audits run at every world shape change.
Non-goals
Public plugin ecosystem or third-party support.
Semver promises, stable WIT, or SDK crates.
Backward or forward compatibility guarantees.
Exit criteria
A plugin-registered provider completes an eval turn with no native vendor adapter.
A subscription login flows end to end through the sandbox and secret proxy.
Every world interface has at least two real consumers and passes the naming and glossary audit.
The world is small enough to review in one sitting, preparing 1.0 stabilization.
---
id: SMH-PHASE-A0RPUJ_Q
type: phase
title: "Prototype"
label: proto
---
# Prototype
## Operating posture
Prove the plugin seam with real consumers.
Grow the `smith:plugin` world organically from working verticals; revisit names, shapes, and structure freely.
Gates and session safety laws never relax.
## Priorities
1. API discovery over API stability: signatures earn their shape from real consumers.
2. Dogfood parity: built-ins load through the same sandbox and effects.
3. Working verticals over breadth: one vendor end to end beats three half-done.
4. Fast plugin loop: scaffolding and dev-links keep edit-run cycles short.
## Decision defaults
- Compatibility: none owed before 1.0; breaking changes are free, deprecation ceremony unnecessary — but the additive-only and world-versioning design stays in the spec so the plan for compatibility exists.
- Migrations: plugin authors get no upgrade path pre-1.0.
- Stability: host internals and world shape may churn without notice.
- Quality: zero-warning, clippy-pedantic, test gates always on.
- Experimentation: lives in `plugins/` and prototypes, never in core crates.
- Debt: naming and glossary audits run at every world shape change.
## Non-goals
- Public plugin ecosystem or third-party support.
- Semver promises, stable WIT, or SDK crates.
- Backward or forward compatibility guarantees.
## Exit criteria
- A plugin-registered provider completes an eval turn with no native vendor adapter.
- A subscription login flows end to end through the sandbox and secret proxy.
- Every world interface has at least two real consumers and passes the naming and glossary audit.
- The world is small enough to review in one sitting, preparing 1.0 stabilization.