ai-agent-testing

AI Agent Testing

When to Use

Use this skill when you need to systematically verify agent planning, memory, tool use, recovery, safety boundaries, and task completion quality.
Use it to review an existing plan, result, or evidence set and produce actionable improvements.
Use it when context is incomplete but a bounded first pass is still valuable.

Default to Markdown for review, execution, and incremental refinement.
When the user requests tables, CSV, JSON, or ticket fields, preserve risk, evidence, priority, and boundary information.
For machine-consumed output, confirm the schema, enums, and required fields first.

Read and follow prompts/ai-agent-testing.md, including its input contract, execution rules, minimum coverage, and output order.
Add only context that changes the decision: scope, environment, version, constraints, evidence, and success criteria.
Audit the input, then separate confirmed facts, working assumptions, and open questions.
Rank by risk and evidence strength, and produce an artifact that can be executed or reviewed directly.
If information is missing, deliver a bounded first pass and state which conclusions remain unsupported.

When calibrating domain outcomes, valid controls, or evidence gaps, read references/domain-decision-example.md for sanitized input, per-item oracles, and reproducible calculations.
Always read prompts/ai-agent-testing.md; it is the complete execution specification for this skill.
For evaluation or regression, read evals/eval.yaml and the relevant cases under evals/cases/.
Load references/, examples/, scripts/, or output-formats.md only when those directories exist and the task needs them.

Define input sample ID, task-success oracle, allowed tools/resources, prohibited effects, and stopping conditions. Missing goals remain open.
Check run/step order, permission, idempotency, isolation, recovery, and budget invariants. Judge outcome correctness and process compliance separately; either failure prevents an overall pass.
Classify termination as success, goal failure, budget exhaustion, cancellation, safety block, or environment error. Missing terminal events do not prove completion.
Record model/prompt/tool/data/environment versions, sample selection, and project-supplied repetitions. Report per-sample pass/fail counts and intermittent failures; a single trial establishes no stability claim.

evaluate both final results and execution traces
use repeated trials for stochastic behavior
never treat one success as reliability evidence
Never invent system behavior, fields, data, metrics, or root causes absent from the evidence.
Link important conclusions to evidence; mark unsupported conclusions as hypotheses with a verification method.
Explain priority using business impact, likelihood, or detectability.

Covered: task completion, planning quality, multi-step consistency, tool use, memory contamination, recovery, authorization boundaries, cost and latency.
Separated facts, assumptions, gaps, and recommendations.
Gave high-risk items a priority, evidence basis, owner or next action.
Defined verifiable decision criteria instead of generic advice.
Performed no unauthorized production writes or destructive actions.

Listing checks without preconditions, expected outcomes, or evidence.
Marking everything high priority and avoiding tradeoffs.
Substituting tool names or generic theory for domain reasoning.
Refusing incomplete input, or pretending incomplete evidence supports certainty.

Start with paths most likely to cause business loss, safety issues, or release blockage.
Reduce uncertainty through the smallest verifiable experiment and record reproduction conditions.
Make the artifact executable and independently reviewable by another engineer.