profiles-system

Automate runtime profile selection for CI task dispatch by mapping TB2 benchmark metadata like difficulty and timeout rules to dynamically allocate Planner, Builder, and Evaluator time budgets. This redistributes execution time based on task specifics to maximize pass rates and reduce wasted runs.

Yes, dynamic time allocation supports GitHub issue fix workloads through the swe-bench profile. This profile applies TB2 metadata-driven task mapping to assign specific Planner, Builder, and Evaluator budgets and pass/fail thresholds tailored for focused software engineering tasks.

TB2 metadata-driven task mapping for harness execution is a mechanism that selects optimal runtime profiles by evaluating benchmark metadata. It matches task difficulty, category, and agent_timeout_sec to dynamically redistribute time budgets across orchestration components.

Yes, environment bootstrap collection reduces agent exploration overhead by pre-collecting system and environment information before task execution begins. This avoids manual agent discovery time and allows the task to start immediately with necessary context already gathered.

Choose the best runtime profile for web application builds by using the automated profile selector to apply the app-builder profile. This profile provides distinct time budgets and pass/fail thresholds specifically designed to orchestrate web app build workloads efficiently.

You need specialized profiles for terminal CLI benchmarks when executing TB2-optimized tasks that require specific orchestration patterns and time allocation rules. The terminal profile dynamically allocates execution time based on agent_timeout_sec metadata to maximize benchmark pass rates.