agent-eval-harness

Generate eval.yaml for the agent eval harness. Two modes - (1) Skill-based - examines SKILL.md, sub-skills, scripts, test cases to verify implementation quality, OR (2) Prompt-based - tests agent capabilities using custom analysis prompts (documentation effectiveness, pattern understanding, API usage, constraint compliance). Produces complete config with execution mode, dataset schema, outputs, judges, models, thresholds. Use when setting up evaluation, testing skills/documentation, adding quality checks, or benchmarking. Auto-triggered by /eval-run when eval.yaml missing. Triggered by "how do I know if my skill/docs work?"

Generate evaluation test cases for an eval.yaml. Sources cases per generation.strategy - skill analysis (default), synthetic LLM generation from generation prompts (documentation and agent-capability evals), or MLflow production traces. Bootstraps a starter dataset or augments an existing one to improve coverage. Use when setting up evaluation, when the user needs test cases, when coverage is too thin, or after /eval-analyze when no dataset exists yet. Triggers on "create test cases", "generate test data", "need test inputs", "make a dataset", "add more cases", "improve coverage", "generate documentation eval cases". Also useful when /eval-run reports "no test cases found."

MLflow integration for evaluation — sync datasets, log run results, push/pull feedback between the harness and MLflow traces. Use when the user wants to log eval results to MLflow, sync test cases to MLflow datasets, connect judge scores to traces, pull MLflow annotations for eval-optimize, or view results in the MLflow UI. Triggers on "log to mlflow", "sync dataset", "push results", "mlflow integration", "view in mlflow".

Automated skill improvement loop. Runs eval, identifies judge failures, reads traces and rationale, edits the SKILL.md to fix issues, re-runs to verify, and checks for regressions. Use when the user wants to automatically improve a skill based on eval results, fix failing judges, make the skill better, auto-fix quality issues, improve scores, or iterate until all judges pass. Triggers on "optimize the skill", "make it pass", "auto-fix", "improve the scores", "why is it failing". Works best after /eval-run has produced results to learn from.

Interactive review of evaluation results. Presents judge scores and skill outputs for human feedback, then proposes SKILL.md improvements based on what the user identifies. Use when the user wants to review eval results, look at results, check scores, see what went wrong, give qualitative feedback on skill outputs, or iterate on a skill based on human judgment rather than automated fixes. Triggers on "review the run", "how did my skill do", "what failed", "look at the eval results", "check the scores". Complements /eval-optimize (automated) with human-in-the-loop review.

Execute an evaluation against test cases (skill or prompt mode), score with judges, and report results. Requires eval.yaml (generated by /eval-analyze). Use when the user wants to test a skill, run eval, benchmark, compare models, detect regressions, check skill quality, or verify changes didn't break anything. Triggers on "run eval", "test the skill", "evaluate", "benchmark", "check for regressions", "how does my skill perform", "score the skill", "run the tests", "run my evals", "compare against baseline", "did I break anything", "test my changes". Also called by /eval-optimize for automated iterations.

Optional environment configurator for the agent-eval-harness. Configures MLflow tracking, verifies API keys, and troubleshoots dependency issues. Detects available skills and agentic documentation (CLAUDE.md, AGENTS.md, ai-docs/) to suggest appropriate evaluation modes. Not required for basic usage — dependencies auto-install via SessionStart hook and agent_eval is available via symlinks. Use when the user wants to configure MLflow tracking, troubleshoot import errors, verify the environment, or set up a remote MLflow server. Also triggers on "configure mlflow", "set up tracking", "ModuleNotFoundError", "mlflow not installed", "missing dependencies", or "check my eval environment".

Evaluate the full harness configuration as a system. Scans all skills, commands, CLAUDE.md, and hooks for redundancy, overlap, type misclassification, and structural issues. Produces an informational report with restructuring suggestions. Use when the user wants to check their overall setup health, find redundant skills, detect overlapping triggers, or get restructuring recommendations before diving into individual skill evaluation. Triggers on "check my setup", "harness health", "are my skills redundant", "what should I merge", "setup overview", "configuration check".

Query Jira for bugs in a project and summarize coverage gaps. Test fixture for e2e external-state field detection.