Automate Terminal-Bench integration for Lattice agent benchmarking and failure analysis.
npx skills add https://github.com/latticeHQ/latticeWorkbench --skill tbench-latticehq
Please help me install this Agent Skill. Skill: tbench Source: https://github.com/latticeHQ/latticeWorkbench/tree/main/.lattice/skills/tbench Command: npx skills add https://github.com/latticeHQ/latticeWorkbench --skill tbench-latticehq
Terminal-Bench integration for Lattice agents to benchmark performance and diagnose failures in automated pipelines.
Run the full Terminal-Bench benchmark suite in CI to evaluate all tasks using the default dataset.
High-intent search queries and answers about installing and using this skill.
Automate Terminal-Bench benchmarking in CI by running the full task suite with default datasets to evaluate Lattice agent performance and diagnose pipeline failures. It generates per-task logs and leaderboard-ready reports.
Terminal-Bench integration benchmarks Lattice agent performance and diagnoses failures in automated pipelines. It uses Harbor-backed tasks to generate centralized results, logs, and failure-rate analysis.
Yes, Harbor is required as the evaluation harness to run Terminal-Bench benchmarks. It supports local execution and Daytona cloud sandboxes for parallel workloads across benchmark tasks.
Yes, Terminal-Bench tasks can run in parallel using Daytona cloud sandboxes. Configure TB_CONCURRENCY and TB_DATASET environment variables to manage parallel workloads and dataset selection.
Analyze Terminal-Bench failure rates using centralized results and logs from benchmark runs. The integration provides scripts for analyzing failure data and exporting leaderboard-ready reports to identify optimization opportunities.
Configure Terminal-Bench execution using TB_DATASET, TB_CONCURRENCY, TB_TIMEOUT, TB_ENV, TB_TASK_NAMES, and TB_ARGS environment variables. LATTICE_RUN_ARGS controls Lattice agent-specific parameters.
lwyBZss8924dAutomate Harbor TB2 environment evaluations for ReCodeAgent workflows.
47networkRuns standardized benchmark suites and generates consolidated leaderboard reports for AI models.
blaxel-aiTranslate Daytona sandbox code to Blaxel equivalents for TypeScript and Python.
simplysmartaiEvaluate LLM applications with automated metrics, human feedback, and LLM-as-judge patterns.
