Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →miosa-osa/synthetic-data
Generate diverse synthetic test inputs via dimension-based tuple generation. Defines variation dimensions, enumerates combinations, filters for relevance, and produces labeled test cases. For LLM eval pipelines, training data augmentation, and stress testing.
Passed 19 of 21 test cases with the skill, and 8 of 21 without it.
and 19 more in the full report
13 fixed8 passed either way1 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 12 test cases the skill fixed. Same model, same prompt, with and without the skill.
Construct a single synthetic test case record for a math topic and hard difficulty where input prompt is 'Solve 2x + 5 = 15', answer is 'x = 5', and label confidence score is 0.95.
| Checked | Without skill | With skill |
|---|---|---|
| Uses expected as the key for ground truth answers | Missed | Passed |
| Includes a confidence key set to 0.95 | Missed | Passed |
| Includes a dims object containing dimension key-value pairs | Missed | Passed |
The instructions your model reads when the skill loads.
Generate diverse synthetic test inputs through dimension-based combinatorics.
Generate synthetic test inputs by defining variation dimensions (topic, difficulty, format, length, persona, edge case type, etc.), computing the combinatorial product, filtering for meaningful combinations, and producing labeled test cases. Ensures eval datasets cover the full capability surface rather than clustering around easy cases. Supports seeded generation for reproducibility.
# Generate from dimension spec
/synthetic-data --dimensions dimensions.yaml --count 200
# Quick generation with inline dimensions
/synthetic-data --dim "topic:math,science,history" --dim "difficulty:easy,medium,hard" --count 50
# Generate to fill gaps identified by eval-audit
/synthetic-data --gaps eval-audit-report.md --count 100
# Generate with constraints
/synthetic-data --dimensions dims.yaml --count 200 --constraint "not (topic=math and difficulty=easy)"
# Append to existing dataset
/synthetic-data --dimensions dims.yaml --count 50 --append evals/dataset.jsonl
# Reproducible generation
/synthetic-data --dimensions dims.yaml --count 200 --seed 42
| Flag | Type | Default | Description |
|---|---|---|---|
--dimensions | string | — | Path to YAML dimension specification file |
--dim | string[] | — | Inline dimension definition (repeatable): "name:val1,val2,val3" |
--count | int | 100 | Number of test cases to generate |
--gaps | string | — | Path to eval-audit report to fill identified gaps |
--constraint | string | — | Boolean expression to exclude combinations |
--append | string | — | Append to existing dataset file |
test cases passed, vs 8 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.