Results for “eval”

Use when the user has run multiple evaluation skills and wants a comprehensive analysis — maturity assessment, cross-skill signals, trends, prioritized actions, and an executive summary. Also use when the user mentions eval health check, evaluation audit, ship readiness, evaluation maturity, or "how good is my evaluation system itself." This is a read-only analysis skill.

The English summary is being prepared. Showing the original description.

Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.

The English summary is being prepared. Showing the original description.

Run evaluations for Hugging Face Hub models using inspect-ai and lighteval on local hardware. Use for backend selection, local GPU evals, and choosing between vLLM / Transformers / accelerate. Not for HF Jobs orchestration, model-card PRs, .eval_results publication, or community-evals automation.

The English summary is being prepared. Showing the original description.

Guide for diagnosing and improving MSBuild project evaluation performance. USE FOR: builds slow before any compilation starts, high evaluation time in binlog analysis, expensive glob patterns walking large directories (node_modules, .git, bin/obj), deep import chains (>20 levels), preprocessed output >10K lines indicating heavy evaluation, property functions with file I/O ($([System.IO.File]::ReadAllText(...))), multiple evaluations per project. Covers the 5 MSBuild evaluation phases, glob optimization via DefaultItemExcludes, import chain analysis with /pp preprocessing. DO NOT USE FOR: compilation-time slowness (use build-perf-diagnostics), incremental build issues (use incremental-build), non-MSBuild build systems.

The English summary is being prepared. Showing the original description.

Scaffolds eval.yaml evaluation specs for skills, custom agents, and redistributable gh-aw workflow packages in the dotnet/skills repository. Use when creating skill or workflow-package tests, writing evaluation stimuli, defining graders and rubrics, sizing an eval for statistical power, or setting up test fixture files. Handles the Vally eval.yaml schema, fixture organization, and overfitting avoidance. Do not use for running or debugging existing evals (use improve-skill-quality) nor for skills authoring (use create-skill).

The English summary is being prepared. Showing the original description.

Braintrust is a hosted platform for evaluating and tracing LLM applications: it runs a task over a dataset, scores the outputs, stores each run as an experiment and logs production traces. Use when a user asks to write or run a Braintrust eval, score LLM outputs with autoevals, compare prompt or model versions as experiments, run evals in CI and fail on regressions, trace OpenAI or Anthropic calls with Braintrust, manage evaluation datasets, or use the bt CLI.

The English summary is being prepared. Showing the original description.

Run AI agent and LLM evaluations in CI/CD pipelines — automated quality gates that fail the build when AI output quality drops. Use when someone asks to "test my AI agent", "add evals to CI", "catch prompt regressions", "compare models", "evaluate LLM output quality", "set up AI quality gates", or "benchmark my agent before deploying". Covers eval frameworks (Cobalt, Promptfoo, Braintrust), LLM-as-judge scoring, threshold-based assertions, and GitHub Actions integration.

The English summary is being prepared. Showing the original description.

Audit whether a paper's EVALUATION DESIGN actually measures what it claims and whether its reporting is complete — the validity layer family D (experiment-forensics) cannot reach. Three patterns: train/test leakage means the reported score may not measure generalization (HP-EVAL-LEAKAGE — adopts the Kapoor & Narayanan 8-type / 3-category leakage taxonomy; the illegitimate-proxy / sampling-bias / pretraining-contamination subtypes hand off as needs_external_check, naming but NEVER running Oren-2023 exchangeability / Shi-2023 Min-K% / Golchin-2023 Time-Travel / BIG-bench canary); a load-bearing LLM judge is conflicted (same model/family as a compared system) or unvalidated (no human-agreement, no bias control) (HP-JUDGE-VALIDITY); a declared condition/metric is dropped or switched to favor the method, or 'best' is chosen with no held-out set (HP-SELECTIVE-REPORTING). Verdict-bearing at L0/L1 from the DESCRIBED protocol — NOT repo-gated like experiment-forensics; L2 only CONFIRMS against split/preprocessing/result files. A fresh cross-model reviewer (gpt-5.6-sol xhigh, read-only, fresh thread per pass) PROPOSES findings, each span-anchored to a ledger claim_id; tools/adjudicate_findings.py DECIDES the verdict. Leakage and under-reporting are usually HONEST methodological errors — every finding describes a discrepancy to CHECK, never an accusation. An LLM generating GROUND-TRUTH labels is HP-FAKE-GT (experiment-forensics) — routed there, not here. Emits eval-design-forensics.findings.json; computes NO verdict. Detect-only. Triggers: "eval design audit", "evaluation validity", "train/test leakage", "data leakage", "is the score measuring generalization", "LLM judge bias", "is the judge validated", "selective reporting", "cherry-picked results", "评估设计审计", "评测有效性", "数据泄漏", "训练测试集泄漏", "裁判模型有没有验证", "选择性报告".

The English summary is being prepared. Showing the original description.

Advanced Evaluation workflow skill. Use this skill when the user needs This skill should be used when the user asks to "implement LLM-as-judge", "compare model outputs", "create evaluation rubrics", "mitigate evaluation bias", or mentions direct scoring, pairwise comparison, position bias, evaluation pipelines, or automated quality assessment and the operator should preserve the upstream workflow, copied support files, and provenance before merging or handing off.

The English summary is being prepared. Showing the original description.

Measures and improves the quality of AI models and agents on Google Cloud using the Eval Quality Flywheel methodology. Use when evaluating an agent or model, building an eval dataset, picking or writing evaluation metrics, analyzing failures, comparing results before and after a fix, or when guidance is needed on Agent Platform eval methodology — including dataset schema, LLM-as-judge scoring, and common failure causes. For fine-tuning, use agent-platform-tuning. For general production deployment, use agent-platform-deploy.

The English summary is being prepared. Showing the original description.

Evaluate Jira ticket complexity + inconsistencies via the jira-evaluator agent. Read-only. Triggers: 'evaluate KEY', 'analyze ticket complexity for KEY', 'check ticket quality for KEY'. NOT for estimation (mk:jira-estimator); NOT for full RCA (mk:jira-analyst).

The English summary is being prepared. Showing the original description.

Improve AI application with evaluation-driven development. Define eval criteria, instrument the application, build golden datasets, observe and evaluate application runs, analyze results, and produce a concrete action plan for improvements. ALWAYS USE THIS SKILL when the user asks to set up QA, add tests, add evals, evaluate, benchmark, fix wrong behaviors, improve quality, or do quality assurance for any Python project that calls an LLM model.

The English summary is being prepared. Showing the original description.

Improve AI application with evaluation-driven development. Define eval criteria, instrument the application, build golden datasets, observe and evaluate application runs, analyze results, and produce a concrete action plan for improvements. ALWAYS USE THIS SKILL when the user asks to set up QA, add tests, add evals, evaluate, benchmark, fix wrong behaviors, improve quality, or do quality assurance for any Python project that calls an LLM model.

The English summary is being prepared. Showing the original description.

Investigate AI observability evaluations of both types — `hog` (deterministic code-based) and `llm_judge` (LLM-prompt-based). Find existing evaluations, inspect their configuration, run them against specific generations, query individual pass/fail results, and generate AI-powered summaries of patterns across many runs. Use when the user asks to debug why an evaluation is failing, surface common failure modes, compare results across filters, dry-run a Hog evaluator, prototype a new LLM-judge prompt, or manage the evaluation lifecycle (create, update, enable/disable, delete).

The English summary is being prepared. Showing the original description.

Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.

The English summary is being prepared. Showing the original description.

Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.

The English summary is being prepared. Showing the original description.

Add or extend Paperclip end-to-end evaluations that exercise real browser workflows across the server, database, and runner. Define expected outcomes, check independent evidence, and connect results to the existing reporting pipeline for local or Daytona execution.

• Adding an evaluation for a complete user workflow
• Updating test data and outcome checks
• Including screenshots and artifacts in evaluation reports

Train and evaluate models that estimate human poses from WiFi signals. Compare results against a model that always predicts the average pose, check for data leakage, and report measured improvements with clear evaluation conditions.

• Train a pose model using camera labels
• Evaluate pose estimation without a camera at inference
• Compare accuracy against an average-pose baseline

Create and run automated tests that score LLM answers against expected outputs or judging criteria. Monitor progress, inspect individual results, compare models on the same tests, and use score thresholds in CI, an automated checking workflow.

• Creating tests with expected answers
• Monitoring evaluation progress
• Inspecting failed samples and scores

Create a fixed JSON evaluation suite from a repository's existing tests and check its structure for errors. Use the suite with harness-evolve to compare variants against consistent tasks, independently of the repository's normal test command.

• Create a fixed evaluation suite from existing tests
• Check evaluation suite structure in CI
• Compare variants across commits using consistent tasks

Benchmark how CodeGraph affects an agent investigating a real codebase. Choose a version, language, repository, and execution mode, then compare cost, tool calls, file reads, and answer correctness with and without CodeGraph.

• Evaluating a local development build
• Auditing a published CodeGraph version
• Comparing investigation cost and file reads

Build and evaluate time-to-event models with scikit-survival, including records whose events remain unobserved when follow-up ends. Keep preprocessing separate from evaluation data, compare models, predict survival probabilities, and summarize censoring-aware metrics.

• Validate event indicators and follow-up times in local CSV data
• Train survival models without leaking evaluation data
• Evaluate survival probabilities with censoring-aware metrics

Create consistent records for evaluating AI agents, including task success, invalid tool calls, token use, and actual costs. Produce CSV tables, database definitions, and JSON validation schemas to support reproducible comparisons and checks for performance declines.

• Recording agent evaluation results in a shared format
• Comparing a candidate agent with a baseline
• Tracking actual cost per solved task

Turns supplied AI agent evaluation results into reports that distinguish autonomous success, human-assisted success, failure, timeout, and invalid runs. Makes metric denominators, retries, timing, costs, and comparison conditions explicit to support evidence-based decisions.

• Separating autonomous and human-assisted success rates
• Reporting evaluations with retries and invalid runs
• Comparing agents or releases under matched conditions