BMAD-METHOD/docs/native-skills-lean-shard-do.../phase-1-analysis.md

1.6 KiB

Phase 1 Analysis: Native Skills Lean PoC

Date: 2026-03-07
Branch: feature/native-skills-lean-shard-doc-prototype
North-star reference: docs/native-skills-transition-north-star-thread-2026-03-07.md

Problem Statement

The prior native-skills transition effort overshot scope. This recovery PoC must prove a single end-to-end duplicate native-skill path while preserving all current legacy task/help behavior.

Scope

In scope:

  1. One duplicated native-skill prototype only: bmad-shard-doc-skill-prototype
  2. Source capability remains src/core/tasks/shard-doc.xml
  3. Installer behavior only for supported native-skill tools:
    • discover prototype metadata
    • register/copy to skill output surface
  4. Minimal tests proving prototype duplication for skill tools and no regression for non-skill tools

Out of scope:

  1. Multi-capability conversion
  2. Broad authority/metadata redesign
  3. Command/help surface changes
  4. Repository-wide migration framework

Constraints

  1. Keep module-help.csv behavior unchanged
  2. Keep legacy bmad-shard-doc capability intact
  3. Keep PR lean and reviewable
  4. Avoid touching unrelated installer paths

Risks and Mitigations

  1. Risk: duplicate visible command surfaces
    Mitigation: apply prototype duplication only on skill_format installers
  2. Risk: behavior drift for legacy task/help paths
    Mitigation: no edits to legacy task file, task/help catalogs, or command generation rules
  3. Risk: hidden regressions across tool outputs
    Mitigation: add targeted install-component tests for one skill-format and one non-skill tool