Fable Architect
Prosa de arquitectura que justifica cada llamada y define una verificación por segmento
Probado · Funciona
Qué hace
Una single-file architecture skill que forces the agent to justify each design decision con an explicit reason, order the build as vertical end-to-end slices instead of layer by layer, y state a verification check for every slice. Se activa cuando se start a new project, plan a major refactor, o weigh a technology choice. No scripts, no dependencies — one SKILL.md that changes how the design answer is written.
Informe de la prueba
Se ejecutó el same architecture brief twice — once cold, once following the skill body strictly. La skill run carried 8 explicit per-slice acceptance checks against 0 in the baseline, 18 because/since justifications against 0, y caught a real coupling bug que el cold run missed: el WebSocket URL declared independently in two modules. El cost es bulk y scannability — 786 words against 231, y el body bans section headers, so you get a wall of prose instead of a design doc you can skim. Su headline claim, Grade A at 100% emulation, comes from the author's own keyword-checkpoint harness tuned across nine rounds, y la advertised 31/31 verification only checks the skill's own statistics against its own shipped data dump, not whether the skill improves anything.
Probado el: 2026-08-10 · Claude Code 2.x (agent harness)
Instalación
git clone https://github.com/ahmdd4vd/Fable5res.git mkdir -p ~/.claude/skills cd Fable5res && cp -r skills/fable-architect ~/.claude/skills/fable-architect
Comandos y prompts de ejemplo
/fable-architectProsa de arquitectura que justifica cada llamada y define una verificación por segmento
Los skills se activan con peticiones en lenguaje natural, sin comandos que memorizar. Tras instalarlo, prompts como estos lo activan (en inglés):
Design the architecture for a Markdown live-reload preview CLIPlan this refactor as vertical slices with a verification checkBreak down the build order for this new project's architecture