Skills
tewarishishir-dataforge

agensi

Manifest-driven data pipeline toolkit: four cooperating skills that discover a warehouse, enforce pipeline standards, run every query through one platform interface, and build metric pipelines from a GitHub issue to a pull request. Supports Databricks and DuckDB. No business values are hardcoded; everything project-specific is read from one generated manifest. Triggers on "dataforge", "initialize dataforge", "setup dataforge", "build metric", "new metric", "metric from issue", "metric pipeline", "check standards", "lint pipeline", "run SQL", "run job", "databricks", "duckdb".
stars3
forks2
watches3
updated2026-10-09 01:06:43

dataforge

Entry point for the dataforge bundle. It routes a request to one of four skills in skills/ and holds no rules of its own. Read the routed skill in full before acting; each one links its own references.

Route the request

The user wants toRead
Set up dataforge, or refresh the manifest after the warehouse or codebase changedskills/forge-init/SKILL.md
Run SQL, explore catalogs, log in to a zone, trigger or monitor a jobskills/forge-dbx/SKILL.md
Change or review pipeline code, SQL, schema, or job configurationskills/forge-lint/SKILL.md
Build a metric pipeline from a GitHub issue or a plain-language requestskills/forge-builder/SKILL.md

If the request fits none of these, say so rather than picking the nearest.

Before any route

  1. Manifest first. Every skill reads dataforge.manifest.resolved.yaml. If the user's project has a dataforge.manifest.yaml at its root, that is the seed. Otherwise the seed is assets/dataforge.manifest.yaml in this folder, documented by assets/dataforge.manifest.seed.example.yaml. With no resolved manifest, run forge-init before any other skill.
  2. Keep the layout. The skills link each other and the manifest by relative path (../forge-lint/SKILL.md, ../../assets/). Do not move or rename files inside this folder.
  3. Script paths. forge-builder runs python skills/forge-builder/verify.py and keeps build state in skills/forge-builder/.build-state/. Both paths are relative to this folder, so resolve them against it rather than against the user's project.

What the skills execute

The skills run local command-line tools and nothing else: databricks or duckdb (whichever platform.engine names), gh for issue intake and pull requests, git, and python for the bundled verify.py. They make no network calls of their own beyond what those tools do. Each skill marks the points where the user must confirm before continuing.

Try it without a cloud account

examples/support-desk/ is a runnable DuckDB project. Install duckdb, run python seed.py in that directory, open it as the workspace, and say initialize dataforge. DEMO.md walks from a GitHub issue to a pull request.

Source, issues, and releases: https://github.com/tewarishishir/dataforge