profiles-system

What problem does it solve?

Automatically choose and apply the most appropriate runtime profile for a given task so agents run with optimized time budgets, orchestration patterns, and pre-collected environment data to reduce wasted execution time and increase success rates.

Core Features & Use Cases

• Multiple specialized profiles: terminal (TB2-optimized), app-builder, swe-bench, reasoning, and base profiles with distinct budgets and thresholds.
• Dynamic time allocation: redistribute Planner/Builder/Evaluator time based on TB2 metadata and task timeout rules to maximize pass rates.
• TB2 metadata-driven selection: map tasks using TB2 benchmarks metadata (difficulty, category, agent_timeout_sec) to choose profiles automatically.
• Environment bootstrap: pre-collect system and environment information to avoid agent exploration overhead.
• Use cases: run CLI benchmark tasks with terminal profile, orchestrate web app builds with app-builder, perform focused GitHub issue fixes with swe-bench, and handle short reasoning/Q&A workloads.

Quick Start

Run the profile selector to automatically detect and apply the best profile for your task description using the harness profile-selector tool.