How martbln/benchmark-loop works
benchmark-loop is a Claude Skill that guides ambitious builds through reference-based scoping, staged implementation, and blind evaluation. The graph reflects the README’s documented workflow and reference-file loading; no source excerpts were available, so implementation-level call relationships are not inferred.
Main entry points: SKILL.md Core Loop · judging.md Blind Judge Loop · domains.md Domain Adaptations · prompt-template.md Prompt Template
Components 10
Benchmark Skill
- Core Loop
SKILL.md
Build And Review
- Set The Bar
SKILL.md - Freeze Foundation
SKILL.md - Parallel Work
SKILL.md - Blind Judge Loop
references/judging.md - Integration Pass
SKILL.md - Stop Conditions
SKILL.md
Reference Guides
- Domain Adaptations
references/domains.md - Prompt Template
references/prompt-template.md
Other components
- Builder
Connections 10
- Builder → Core Loop: requests ambitious build
README.md - Core Loop → Set The Bar: starts with
README.md - Set The Bar → Freeze Foundation: precedes
README.md - Freeze Foundation → Parallel Work: enables
README.md - Parallel Work → Blind Judge Loop: produces artifact for
README.md - Blind Judge Loop → Integration Pass: precedes
README.md - Integration Pass → Stop Conditions: checks
README.md - Core Loop → Blind Judge Loop: loads before comparison
README.md - Core Loop → Domain Adaptations: loads when scoping
README.md - Core Loop → Prompt Template: offers example prompt
README.md