Discover with a purpose
Organize candidates around a defined task instead of collecting prompts without context.
Evaluate the evidence
Use representative cases, explicit rubrics, and failure review before promoting a candidate.
Version the workflow
Preserve input contracts, exact prompt revisions, and evaluation history.
A different kind of mining
On MinerAPI.com, AI prompt mining means the workflow of discovering, organizing, and evaluating prompt candidates. It is not cryptocurrency mining. Its useful outputs are task-specific prompt records and inspectable evaluation results, not hashes or blocks.
Begin with success criteria before comparing outputs. Claude's evaluation documentation describes measurable criteria and different ways of assessing results. Read the evaluation design reference.
Move from a candidate to a dependable record
Define required input variables, the expected output shape, and how the application should handle missing information. Store prompt identity separately from the current revision. Keep dataset and evaluator versions attached to each run so another developer can understand what was actually tested.
A high aggregate pass rate is not the entire decision. Inspect failure cases, important input groups, and resource use separately. Keep human judgment visible where the task requires it, and document unresolved limitations before moving a candidate into an application workflow.
Try the public record shape
The prompt JSON fixture contains a fictional prompt record and sample evaluation metadata. Fetching it does not call a language model, evaluate user content, or connect a provider account. It is useful for building a consumer and testing its rendering before an authorized execution service is introduced.
Use the evaluation guide first, then the versioning guide. Together they describe a small release process with immutable revisions, bounded execution, output validation, and a clear relationship between a result and the contract that produced it.
