// Raw Task Input
Outputs one structured system prompt for Claude Code. Copy / save .md.
Pre-loads task-type-specific guardrails and DONE-WHEN criteria.
Type / coverage of programmatic testing the agent must write specs and tests for (Red→Green). Not a library choice — the runner is still picked from the detected stack.
all
Downloading model...
Drop a .gguf or .onnx model here, or click to choose
Loaded in-browser via transformers.js (WebGPU, CPU fallback). Cached locally. Not uploaded anywhere.
Route inference to the Hugging Face Inference API. Optional — public model lists work with no key. An API key (stored only in this browser's localStorage) enables authenticated / rate-limited models. Never logged.
Anti-loop: repeat_penalty / presence / frequency auto-scale up as context fills (>70% ctx). Detected repetition loops auto-bump penalties.
// Agentic Claude Code Prompt
◈
Output appears here