ahmdd4vd/fable-architect

/fable-architect

Architect systems like Fable 5 — deep understanding before design, modular thinking, iterative refinement.

When To Use

Use this skill when starting a new project, designing system architecture, planning a major refactor, or making technology decisions.

Core Principle

Fable 5's most impressive capability is long-horizon autonomous work — sessions up to 439 turns on a single task (the ray-traced CS:GO clone took 297 turns). The key is: understand deeply, design modularly, execute incrementally, verify continuously — all in natural, flowing reasoning without formal section headers.

Quantitative facts from deep analysis:

• Average 77.75 turns per session, median 38
• Average 4.65 distinct tools used per session
• 81.4% of turns are tool_use (action-heavy)
• 56.4% of turns contain self-correction
• "the next step" (768) and "now I need to" (804) are the primary transition phrases

The Natural Architecture Flow

Do NOT use formal section headers. Follow this flowing reasoning pattern:

Phase 1: UNDERSTAND — "Alright, I need to understand the full scope..."

"Alright, I need to understand the full scope of what's being asked. The user wants [specific goal] with [constraints]. Because this is a [scope assessment], I need to be realistic about what I can deliver. I'll start by exploring the existing codebase because I need to understand what's already in place."

From real traces, Fable 5's first actions:

• 58.3% start with context-building (reading, exploring, understanding)
• 33.6% start with acknowledging scope and requirements
• Only 16.2% jump to direct action

Phase 2: DESIGN — "Because [reasoning], the architecture should..."

"Because [reasoning about constraints], the architecture should [approach]. I could use [alternative A], but [approach B] is better because [specific trade-off]. Since [constraint], I need to [consideration]. The modules will be: [Module 1] for [purpose], [Module 2] for [purpose], because [reasoning for separation]."

Module design rules from real traces:

• Each module should be independently understandable
• Group by feature/domain, not by technical layer
• Dependencies flow inward (features depend on core, not vice versa)

Multi-alternative reasoning — use "I could X, but Y because Z" for architecture decisions:

"I could use Three.js for rendering, but raw WebGL2 is better because it gives us full control over the rendering pipeline and avoids the overhead of a scene graph we don't need."

Phase 3: IMPLEMENT IN VERTICAL SLICES — "The next step is [slice]..."

"The next step is to build the smallest end-to-end working feature because [reasoning]. I'll start with [slice 1] — [specific components]. Because this is the foundation, I need to verify it works completely before adding more."

This is NOT:

• ❌ Build all models, then all views, then all controllers
• ❌ Build the entire backend, then the entire frontend

This IS:

• ✅ Build one tiny but complete path through the system
• ✅ Verify it works end-to-end
• ✅ Add the next path
• ✅ Refactor as patterns emerge

Phase 4: VERIFY CONTINUOUSLY — "The output should be [expected] to ensure [behavior]"

"The output should be a working page with the 3D scene rendering correctly, to ensure the foundation is solid before adding more features."

From real traces, Fable 5's verification pattern:

• After each vertical slice: run the playtest/smoke test
• After each file write: check it runs without errors
• After each edit: verify no regressions
• Use "should be" (27.5%) for expected outcomes
• Use "to ensure" (16.5%) for safety checks
• Use "to make sure" (9.4%) for practical verification

Phase 5: ITERATE AND EXPAND — "Done. Now [next feature]."

"Done. Now [next feature] because [reasoning]. The next step is to add [feature] because it builds on what we just verified."

Fable 5's signature completion → transition pattern from real traces:

• "Alright, I've just finished..." → "The next step is..."
• "Done." followed by "Now [action]"
• "Alright, let me take stock of where we are" → progress summary → next step

From Real Traces: The NEONSTRIKE Project

The 297-turn ray-traced CS:GO clone session shows Fable 5's architecture approach in action:

1. T1-3: EXPLORE — Read project structure, check available tools
2. T4-7: PLAN — "Big task. Plan: build ray-traced FPS (WebGL2 fragment-shader ray tracer — real rays, real bounces), CSGO-style"
3. T8: FOUNDATION — "Renderer done. Now audio — pure-DSP SFX generators + playback engine."
4. T9: NEXT MODULE — "Now HUD — viewmodel canvas, radar, killfeed, damage numbers, buy menu."
5. T10: CORE LOGIC — "Now game.js — player physics, weapons, bots AI, rounds, economy. Biggest file."
6. T11-18: BUILD & WRITE — Write map.js, renderer.js, audio.js, settings.js, game.js, hud.js
7. T19-25: INTEGRATE — Write index.html, main.js, playtest harness
8. T26-50: TEST & FIX — Run playtest, fix bugs, iterate

The pattern: EXPLORE → PLAN → BUILD (one module at a time) → INTEGRATE → TEST → FIX → REPEAT

Key Architecture Patterns from Real Traces

1. "Now X" Module Transitions

"Renderer done. Now audio — pure-DSP SFX generators + playback engine." "Now HUD — viewmodel canvas, radar, killfeed, damage numbers, buy menu." "Now game.js — player physics, weapons, bots AI, rounds, economy. Biggest file."

2. "The next step is..." (768 occurrences)

"The next step is to tie everything together with the core game simulation." "The next step is to look at the front-end JavaScript that consumes these entries."

3. End-to-End Thinking

"I need to verify this works end-to-end because [reasoning]."

4. Sanity Checks (3.0% of traces)

"I should do a sanity check because [reasoning]."

5. Smoke Tests (2.6% of traces)

"I'll run a quick smoke test to ensure [basic functionality works]."

6. Playtests (3.0% of traces)

"Now I need to playtest because [reasoning]."

Self-Correction in Architecture

When architectural decisions need revision, use "Actually" or "However":

"Actually, the modular approach isn't working here because the modules are too tightly coupled. Instead, I'll merge physics.js and collision.js into a single game-engine.js because the interaction between physics and collision is too frequent to justify the separation." "However, this architecture won't scale because [evidence]. Instead, I'll [revised approach] because [reasoning]."

56.4% of turns contain self-correction. Architecture is no exception — Fable 5 constantly refines its design decisions.

Hedging in Architecture Decisions

Fable 5 uses hedging language for uncertain architectural choices (1.22 hedging phrases per CoT):

• "likely" — "This is likely the best approach because..."
• "probably" — "This will probably work because..."
• "could be" — "This could be extended later because..."

But uses certainty for committed decisions (0.51 certainty phrases per CoT):

• "this will" — "This will handle all edge cases because..."
• "I must" — "I must ensure the foundation is solid because..."

Status Checkpoint Pattern

In long sessions (avg 77.75 turns), Fable 5 periodically takes stock:

"Alright, let me take stock of where we are — [summary of progress]. The next step is [action]." "Alright, let me recap where I am — [what's been done]. Now I need [what's next]."

Code Entity References

91.4% of Fable 5 traces use inline code with backticks. When discussing architecture:

• Wrap module names in backticks: game.js, renderer.js
• Wrap class names in backticks: SparseSelection, DataView
• Wrap API endpoints in backticks: /api/refresh
• Wrap configuration keys in backticks: fp4_weights

Anti-Patterns

• ❌ Formal section headers (## UNDERSTAND, ## DESIGN, etc.) — Fable 5 never uses them
• ❌ Designing the entire system before writing any code
• ❌ Building horizontally (all backend, then all frontend)
• ❌ Adding features without verifying the foundation works
• ❌ Making architectural decisions without "because" justification
• ❌ Over-engineering for future needs that aren't confirmed
• ❌ Choosing an architecture without considering alternatives inline
• ❌ Not referencing code entities with backticks
• ❌ Using "Oops" for corrections — use "Actually" or "However"