prompt-testing

Prompt Testing (Chinese Version)

When to Use

You need to verify the correctness, stability, and controllability of prompts under representativeness, boundaries, adversarial conditions, and version changes.
You need to use prompt-regression mode to compare a baseline and a candidate version, and record dataset identity, behavior differences, and evidence status.
You need to conduct a risk review of an existing plan, result, or evidence, and form actionable improvement items.
The input is incomplete, but a usable initial version with assumptions and information gaps still needs to be provided first.

The default output is Markdown, suitable for review, execution, and continuous supplementation.
When the user requests a table, CSV, JSON, or ticket format, retain the same risk, evidence, priority, and boundary fields.
If the output will enter an automated process, first confirm the field schema, enum values, and required items.

First read and follow prompts/prompt-testing.md input, execution rules, minimum coverage checklist, and output order.
Choose normal testing or prompt-regression mode; regression mode must read the baseline, candidate version, dataset/test prompt identity, and comparability information.
Supplement the scope, environment, version, constraints, evidence, and success criteria that truly affect judgment.
First conduct an input audit, then distinguish confirmed facts, reasonable assumptions, and open questions.
Sort by risk and evidence strength, and produce results that can be directly executed, reviewed, or verified.
When information is insufficient, do not stop at asking questions: first deliver a constrained initial version, and explain which conclusions cannot yet hold.

Each execution must read prompts/prompt-testing.md; it is the complete execution specification of this Skill.
When this Skill needs to be evaluated or regressed, read evals/eval.yaml and the matching evals/cases/ cases.
Only read references/、examples/、scripts/ or output-formats.mdwhen the directory actually exists and the task requires it; do not assume assets that do not exist.

Do not test only one sample
Fix the model and parameters
Use a rubric for semantic results rather than fragile verbatim matching
Do not fabricate system behavior, fields, data, metrics, or root causes that do not exist in the input.
Key conclusions must be linked to evidence; when evidence is insufficient, mark them as assumptions and provide a verification method.
Priority must state the basis in terms of business impact, likelihood of occurrence, or detectability.
prompt-regression mode uses PRT-## discovery numbering, and must distinguish baseline, candidate version, expected behavior, observed behavior, differences, and verification method.
Version differences can only be based on comparable inputs and existing evidence; static analysis must not be written as the model having run or regression having passed.

Covered: instruction following, format, factuality, boundary inputs, adversarial inputs, multilingual, stability, regression, cost.
Facts, assumptions, gaps, and recommendations have been distinguished.
High-risk items have clear priority, evidence, owner, or next step.
The output contains verifiable judgment criteria rather than vague generalities.
The regression mode preserves dataset/test prompt identity, evidence status, differences, and Human decision boundaries.
No unauthorized production writes or destructive actions were performed.

Listing only checkpoints without stating input conditions, expected results, or evidence.
Marking everything as high priority, losing the value of trade-offs.
Replacing domain judgment with tool names or generic theory.
Rejecting outright when input is incomplete, or conversely pretending the conclusion is already certain.
Writing surface differences between a candidate version and the baseline directly as regression defects, or omitting comparability and evidence status.

Start with the paths most likely to cause business loss, safety issues, or release blockage.
Reduce uncertainty with the smallest verifiable experiment, and record the reproduction conditions.
Make the output something another engineer can directly execute and review.