Combines floor results and capability-ceiling delta to emit a verdict.
npx skills add https://github.com/mifunedev/openharness --skill benchmark-mifunedev
Please help me install this Agent Skill. Skill: benchmark Source: https://github.com/mifunedev/openharness/tree/main/.claude/skills/benchmark Command: npx skills add https://github.com/mifunedev/openharness --skill benchmark-mifunedev
The Benchmark skill provides a structured gate to decide whether a landed change actually improves the harness by combining the regression floor status with the capability-ceiling delta and emitting a single verdict that informs subsequent routing.
Run the benchmark step after a change to obtain a decisive verdict and route the loop accordingly.
High-intent search queries and answers about installing and using this skill.
To decide if a code change improves an automation harness, evaluate regression-floor results alongside capability-ceiling deltas to emit a single BENEFICIAL or NOT-BENEFICIAL verdict that informs subsequent routing.
A capability-ceiling delta in benchmark evaluation measures the improvement in maximum performance capabilities, which is combined with regression-floor status to determine if a landed change is beneficial.
Automate revert decisions in a continuous evaluation loop by running a benchmark step after a change to obtain a decisive verdict, which then guides the control flow to trigger a revert or repeat action.
An automation benchmark returns a NOT-BENEFICIAL verdict through a deterministic decision path where the first failing signal from either the regression floor or capability ceiling immediately ends the evaluation.
You can gate deployment routing based on benchmark evaluation results by integrating the final STATUS line verdict into an executable loop to satisfy gating constraints and direct control flow.
Deterministic decision paths for harness evaluation are limited by strict sequential logic, meaning the first failing signal ends the evaluation as NOT-BENEFICIAL without evaluating remaining signals.
tiankong0101-byteMeasure system performance and capability differences across iterations.
DubDev720Route InputMap actions through a single authority node for deterministic debounce and repeat behavior.
KappTech88Execute standardized multi-benchmark evaluations of code-generation models with the BigCode Evaluation Harness.
jwmossmozPolls the Lango API to determine landing job statuses and outcomes.
