菜单
Author or validate a DiscoveryConfig dataspec — the YAML describing a target table (columns, grain, business rules, load strategy) consumed by toolkit agent discovery. Use when the user wants to define a new pipeline target, has a dataspec to check, or asks what a dataspec needs to contain.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | spec |
| description | Author or validate a DiscoveryConfig dataspec — the YAML describing a target table (columns, grain, business rules, load strategy) consumed by toolkit agent discovery. Use when the user wants to define a new pipeline target, has a dataspec to check, or asks what a dataspec needs to contain. |
A dataspec is one YAML file per target table. Full field reference: the plugin's
references/spec-schema.md (two levels up from this skill's directory); complete examples per
tooling under references/examples/. Read the schema reference before writing anything.
Read it and validate against the schema reference:
content present and substantive (target table + columns + grain + rules)?tooling/materialization combination legal (per-tooling enums)?loadStrategy keys spelled right (type, change_detection, watermark_column, ...)?targetPlatform consistent with tooling (pyspark → databricks unless they know better)?targetRequirement/resolvedTransformation (producer-only fields)?Report problems with concrete fixes; small gaps (missing grain, vague rules) matter more than formatting — discovery quality tracks spec quality.
Gather, in order (skip what the user already said):
sql, dbt, or pyspark — what should pipeline-build emit?snowflake or databricks (pyspark defaults to databricks).format: ddl.format: natural_language: columns (name, source or "derived from X",
nullability), the grain ("one row per ..."), and business rules (derivations, filters,
SCD expectations). Push for the grain and rules — they drive the generated tests.materialization: merge + loadStrategy.type: incremental_merge with
change_detection: watermark; append-only facts: incremental_append; small/reference
tables: full_refresh/ctas.filters block in
toolkit.conf, that bounds what discovery can see at all.Write the file as specs/<target_table>.yaml in the working directory (one spec per target
table), show it to the user, and point at /toolkit-pipeline:discover as the next step.
Add or verify a JDBC connection and datasource in toolkit.conf for any supported type — Snowflake, SQL Server, Oracle, Postgres, MySQL, Hive, Databricks, Athena — including JDBC driver installation for non-bundled types and verification via a scoped ds scan. Use when configuring data sources, fixing connection errors, or when toolkit-check reports no datasource.
Configure the optional llmClient block in toolkit.conf so toolkit agent commands (discovery, pipeline-build, fix-sql, codegen ask) can call an LLM — Amazon Bedrock, OpenAI, or Anthropic. Use when agent commands fail for lack of an LLM, when switching providers or models, or when onboarding to the toolkit agent workflows.
Install and configure the phData Toolkit CLI end to end — verify Java, get the toolkit binary on PATH, authenticate, initialize a toolkit project, and extract Data Source tool defaults. Use when the toolkit is missing, toolkit-check fails, or a user is onboarding to phData Toolkit workflows.
Show a doctor-style report of the phData Toolkit CLI installation — binary, Java, project, datasources, and optional connectivity. Use when the user asks whether the toolkit is set up, why a toolkit command might be failing, or wants a health check before starting toolkit work.
Run toolkit agent pipeline-build to generate pipeline code from approved data contracts — SQL, dbt, or PySpark transforms plus data-quality tests and a synthetic-data (datagen) spec. Use after discovery produces contracts, or when the user wants to generate, regenerate, apply, or test pipelines from contracts.
Run toolkit agent discovery to resolve a dataspec against a live datasource into a reviewable data contract, then walk the human-review resolve loop until the contract is approved. Use when the user has (or needs) a dataspec and wants source-to-target mappings, or mentions data contracts, discovery, or humanReviewItems.