New to EvalGlass? Try it before the concepts.
These guides explain why it works. To see it work first: install the plugin, run the quickstart, and learn what to say to your coding agent.
Learn
Learn AI quality control for agentic builders
Coding-agent-operated AI quality control means letting your coding agent help create, run, and explain project-specific checks for your AI application, while human-reviewed records in your repo keep authority over what passes.
Learn how coding agents can help evaluate AI applications without becoming the source of truth. These field guides explain project-specific checks, agentic workflows, tool-use and trajectory evaluation, scorecards, CI gates, and the no-self-approval boundary that keeps authority with your repo.
Three actors, one governed loop
Every guide here preserves the same model: the coding agent operates the evaluation work, and your repo keeps authority over what it means.
-
01 · authority
Human / host ownerOwns product intent, validation, thresholds, calibration, baselines, and gates. The agent never takes this over.
-
02 · operator
Coding agentInspects the repo, scaffolds checks, runs EvalGlass, compares results, and explains scorecards. It operates — it does not authorize.
-
03 · subject
AI application / agentic workflowPerforms behavior across calls, tools, retrieval, steps, trajectories, and sessions — the system under evaluation.
Start here
The category, defined: your coding agent operates the evaluation, your repo keeps authority, and the scorecard stays bounded. Start here.
Controlled delegation for Claude Code and Codex: what the agent can run, and the authority boundary it can never cross.
The deepest trust idea, made explicit: whatever creates the evidence can never grant the authority to gate on it. No false green, sharpened for the agent era.
How to read the artifact every run returns — values, status, the four verdicts — and what a green one does not license.
Exactly how far a passing result reaches and where it stops: the four filters a claim must survive, and why exit 0 alone is not even a pass.
Evaluate agentic behavior
Beyond final-answer scoring: the EvalUnit ladder (call, step, trajectory, session), process vs outcome checks, and traces as evidence.
Turn the behavior your app must perform into repeatable evals — examples, metrics, rubrics, thresholds, baselines, authority — and see why generic benchmarks can't.
The concrete checks — tool choice, arguments, step order, recovery, outcome — scored from observable trace evidence, not the model's self-narrated reasoning.
How exported OTel / OpenInference / local-JSON behavior becomes evidence for your checks — imported locally, never a live feed — and why a malformed span is a diagnostic, not a fabricated zero.
When your agent writes your evals: the line between evidence it proposed and reference data a human stood behind — and why generated cases land proposed and can't gate themselves.
Ship changes with bounded confidence
Put scorecards inside the coding-agent loop so a prompt, model, retriever, or tool change is checked before it becomes unexamined product behavior.
When an opt-in gate is appropriate and why a fresh install gates nothing: from an informational first run to a deliberately activated, host-owned gate — and who actually blocks the merge.
What you'll find here
The hub is organized into four tracks. Guides are publishing now — each one teaches a concept, shows the common mistake, and states plainly what a result does not prove.
- Start here. Coding-agent-operated AI quality control, the no-self-approval boundary, and what a scorecard for agentic behavior can claim. New to the idea? Begin with how it works.
- Evaluate agentic behavior. Evaluating agentic workflows, project-specific checks, and tool-use & trajectory evaluation — going beyond final-answer scoring. See it applied in agentic app evaluation.
- Ship changes with bounded confidence. Eval-driven agentic development and AI evals in CI, so a prompt, model, retriever, or tool change is checked before it becomes unexamined product behavior. See CI for AI applications.
- Deepen the model. Why the actor that creates evidence can't authorize it, and where human authority stays decisive. Start from no self-approval and what a green result does not mean.
No self-approval. No false green.
A coding agent can do a lot of the evaluation work — but the actor or tool that creates the evidence can never grant itself the authority to gate on it. Capability is not authority. That boundary is what keeps an agent-operated evaluation honest, and it runs through every guide here.
Start with the product
How it works
the six-step loop: Ask, Observe, Shape, Run, Read, Improve
Use cases
the trigger moments where you reach for EvalGlass
Scorecards
the bounded artifact every evaluation returns
Docs
the commands, files, and contracts to operate it