v0.20.0

Added

• blueprint skill — Product Experience Direction (lightweight UX-intent
  support; skill manifest 0.4.0 → 0.5.0). The blueprint now captures enough
  product-experience direction to keep downstream architecture and
  implementation from inventing UX assumptions, without turning blueprint into a
  full UX-design stage. The boundary is explicit: blueprint defines UX intent
  · architecture defines the UX-enabling technical structure · a later UX-design
  skill defines the detailed experience · implementation-plan turns it into
  build tasks. Folded into the existing prompts/templates/example (no new
  manifest tasks), tech/domain-neutral:
  • New §9 Product Experience Direction (the output template grows from 18
    to 19 sections; §9 sits after Workflow Model and before Logical
    Architecture, so UX intent exists before the architecture handoff). It
    captures, compactly (1–2 pages, tables): Primary Experience Thesis · Primary
    User / Operator · Primary Job-to-Be-Done · Primary Interaction Mode
    (CLI/Web/GUI/TUI/API/AI-Skill/MCP/Hybrid, with MVP stage + rationale) ·
    Secondary / Future Interaction Modes · Critical Trust, Control, and
    Transparency Requirements · Human-in-the-Loop Experience · Failure and
    Recovery Expectations · UX Assumptions for Architecture · Product Experience
    Handoff to Architecture.
  • Product Experience Gate added to the quality-gate prompt and the
    Appendix A self-check (eight rows: primary user, job-to-be-done, experience
    thesis, interaction mode, trust/control/transparency, human-in-the-loop,
    failure/recovery, UX handoff), with explicit FAIL and WARNING conditions.
  • UX-intent boundary enforced — §9 must not contain screen layout,
    wireframes, CSS/visual design, full user journeys, exact CLI syntax/flags,
    exact MCP/API schemas, copywriting, mobile navigation, full accessibility
    checklists, or implementation tasks (UX over-reach is a gate FAIL). SKILL.md
    gains UX trigger phrases and negative triggers (detailed UI/UX → a later
    UX-design skill); the forbidden-content checklist gains a detailed-UX
    section.
  • New reference references/product-experience-direction.md — the section
    template, the boundary rule, the grill-me-style clarification-question
    format (good vs. bad questions), the Product Experience Gate, and the output
    budget.
  • Chain handoff — the architecture skill's blueprint-parse prompt now
    extracts product_experience (interaction mode, trust/control/transparency,
    human-in-the-loop, failure/recovery, UX assumptions) so it preserves UX
    intent instead of inventing it.
  • Updates the manifest, SKILL.md, prompt 04 (generate), prompt 05
    (quality-gate), the output template, both test checklists, and the worked
    example (now modelling §9 + a PASS-with-warning PE-gate row), with new guard
    tests including a Copilot description-length check.