§ decision · 2026-04-21-spec-v0-scope
dated
2026-04-21
status
accepted
deciders
[email protected], agent:claude-opus-4-7

# Spec v0 scope

# Context

Petitgen is entering Wave 1 of its roadmap: prove the .remedy substrate by shipping a minimum spec, tooling, sovereign runtime, and a first dogfood. Earlier iterations in Complaintery/code/remedy built a multi-file workspace layout (features/<slug>/{feature.yml, complaint.yml, code.yml, …}) and a VS-Code-style workbench; a parallel FEATURE_MODEL.md in the strategy hub proposed a single-YAML-per-remedy model with embeddings baked in from day one.

Reconciling those two shapes required a product decision.

# Decision

  1. Remedy is a specification for directories, not a single-file format. The unit is the folder; the .remedy file at its root is the entry-point contract.
  2. File syntax is YAML with canonical key order enforced by remedy fmt.
  3. Embeddings ship in v0, not v1. Local-first via Ollama nomic-embed-text; OpenAI as opt-in.
  4. Kinds are fixed in v0 to five values (feature, tool, product, workspace, doc-set). Extensible in v1.
  5. Marketplace is deferred to v1. v0 proves the spec + tooling + sovereign runtime; marketplace is additive.
  6. First dogfood is TimeBill (not Complaintery), because it is smaller, Sebastien runs it weekly, and failure modes will surface within a week of shipping.
  7. Surfaces in v0: CLI, MCP server for Claude Code, Tauri Mac Desktop app, Astro-based web surface at remedy.build. All four consume remedy-core.

# Consequences

# Alternatives considered

# Follow-ups