0.10.0-alpha · core 2026-06-01
release 0.10.0-alpha on GitHub →
Changed
- BREAKING: Removed
complexity_estimatefield from proposal-structure.md — replaced withappetite_fit(fits/cuts_needed/reshape). Appetite is now treated as a constraint, not a target for estimation. The LLM checks whether the shaped proposal fits the declared appetite, rather than estimating effort on an ordinal scale. - BREAKING: plan-critique/SKILL.md appetite violation check now uses
appetite_fitcase-based logic instead of ordinal comparison (Light vs XS/S/M/L/XL).reshapehalts critique with exit 1. - shape-up/SKILL.md: Validation guard updated to check
appetite_fitfield instead ofcomplexity_estimate. Conceptual callout rewritten to emphasize appetite as constraint. - scope-executor/SKILL.md: Template reference updated from
Complexity EstimatetoAppetite Fit. - execution-critique/SKILL.md: Fixed stale XS/S/L/XL appetite labels → Light/Balanced/Deep.
- setup.md: Comment updated to reference
appetite_fitconstraint model. - README.md: Complete restructure — new section ordering, Key Differentiators as proper heading, "Measure thrice, cut once" as blockquote, Appetite & Mode promoted, Evidence-Based Design + Radical Transparency merged into unified Evidence & Limitations section, Mode system added to differentiators, links added for all Known Limitations papers.
- appetite-consistency.test.ts: Tests updated for new schema; execution-critique added to stale-label scan.
Added
- State coverage baseline: Standardized coverage formula
(✅ + ⬆️) / (✅ + ❌ + ⬆️)across all 4 skills (interface-alternatives, interface-rules, checklists, ui-audit-dimensions) - Component Typing section in
ui-audit-dimensions.md: Int/Disp classification with baseline applicability rules - N/A ⬆️ semantics: ⬆️ (inherited) counts toward coverage, with named system reference requirement
- Guardrails: Component grouping (×N for repeated components), platform-aware states (web vs mobile), self-audit checklist before finalizing coverage table, N/A justification rule (>1 N/A per Int = required note), Display misclassification self-check
- Quantitative consistency checks: Coverage plausibility, N/A inflation detection, Display cell cap (≤3)
- Escape hatch expansion beyond Archetype D: all Archetype A tables may declare escape rows with ⬆️^DS
- Baseline relaxation: read-only/kiosk, voice-only, simple Int components (single-state: toggle, badge)
Fixed
- Coverage formula contradiction:
interface-rules.mdnow uses(✅ + ⬆️) / (✅ + ❌ + ⬆️)(was excluding ⬆️), matchingoutput-format.md - Bugged coverage example: Row showing
4/4corrected to6/6(5 ✅ + 1 ⬆️ = 6 applicable cells) - ⬆️ missing from N/A Semantics table in
interface-rules.md: Added inheritance row - Missing coverage formula in
checklists.md: Added formula + numeric example - Missing scoring threshold alignment: Added
Maps to (present tells)column + cross-file reference for inverted scales - Missing Int/Disp typing in
ui-audit-dimensions.md: Added Component Typing section