Skip to content

Implementation Workflow

This repository is set up for a docs-first workflow that works well with AI tools without depending on the Speckit package itself.

Artifact Flow

For each meaningful feature slice:

  1. Start from Requirements.
  2. Add or update an ADR if the slice introduces a cross-cutting design decision.
  3. Write a spec for the slice.
  4. Write a task that points to exact requirements and acceptance criteria.
  5. Implement in small changes and verify against the task.

Established Baseline

The first four meaningful slices are now defined:

This gives the repo a stable base path from requirements to ADR to spec to implementation task.

For external AI tools, the task pack is the preferred handoff because it includes:

  • smaller scoped tasks
  • recommended model class and reasoning effort
  • copy/paste prompts
  • dependency order

Core Completion State

The final merged core slice is now closed out:

Roadmap

The remaining slices are now summarized in Prototype Roadmap.

Short version:

  • the merged core roadmap is complete through SLICE-004
  • about 2-3 optional expansion slices after that if you want a richer demo/teaching system

Definition of Ready

A slice is ready for implementation when:

  • scope and non-scope are explicit
  • command guards and terminal states are named
  • failure behavior is described
  • touched layers and boundaries are known
  • verification is concrete enough for code review

Definition of Done

A slice is done when:

  • the task scope is fully implemented
  • acceptance criteria are satisfied
  • automated tests cover the critical behavior promised by the slice
  • docs are updated if the slice changes stable contracts or decisions

Next Suggested Follow-Ups

After SLICE-004 close-out, the next likely docs work is:

  • a slice for trend charts and operator metrics if the richer run data proves useful
  • a slice for performance instrumentation and memory profiling hooks
  • a conditional ADR if workflow complexity ever outgrows the current approach

Docs-first project memory for AI-assisted implementation.