core-text

name	core_text
description	Reference documentation for DataFlow's core_text operators — 8 generators, 3 filters, 2 refiners and 5 evaluators. Read by generating-dataflow-pipeline when a task needs an operator beyond the six core primitives. This is a reference package, not a workflow: it is consulted, never invoked directly.

core_text operator reference

Per-operator API documentation for DataFlow's core_text family. Consult it when you need an operator's exact constructor and run() signature, its real row-processing behaviour, or the mistakes that break it.

Do not invoke this as a skill. There is no workflow here. The generating-dataflow-pipeline skill reads these files when it needs an operator beyond its six core primitives.

Layout

core_text/<category>/<operator>/
├── SKILL.md            English reference: constructor, run(), execution logic,
│                       mandatory rules, return-value semantics
├── SKILL_zh.md         Chinese translation
└── examples/
    ├── good.md         Best-practice pipeline usage
    └── bad.md          Common mistakes and how they fail

What is documented

Category	Path	Operators
Generate	generate/	PromptedGenerator, FormatStrPromptedGenerator, Text2MultiHopQAGenerator, BenchAnswerGenerator, ChunkedPromptedGenerator, EmbeddingGenerator, RandomDomainKnowledgeRowGenerator, RetrievalGenerator
Filter	filter/	GeneralFilter, KCenterGreedyFilter, PromptedFilter
Refine	refine/	PandasOperator, PromptedRefiner
Eval	eval/	BenchDatasetEvaluator, BenchDatasetEvaluatorQuestion, PromptedEvaluator, Text2QASampleEvaluator, UnifiedBenchDatasetEvaluator

How to use it

1. Find the operator's directory under its category.
2. Read its SKILL.md for the authoritative signature.
3. Check that operator's bad.md example before writing code — it documents the failure modes that come up most often.

Accuracy and scope

These are static reference docs, aligned to a specific DataFlow version (recorded in VERSION.md in this directory). They are not a live query.

When an MCP server is available, get_operator_detail_by_name is authoritative: it reflects the operators actually installed. Prefer it, and treat a disagreement with these files as this reference being out of date.

Adding an operator

1. Create core_text/<category>/<operator-slug>/ containing SKILL.md, its SKILL_zh.md translation, plus good.md and bad.md under examples/.
2. Add the operator to the category table above. Add a rule to ../generating-dataflow-pipeline/SKILL.md only when it changes operator selection policy; the planner reads this index on demand.