phData Agent Marketplace

Claude Code plugin marketplace for phData's coding-agent extensions — plugins that package phData Toolkit CLI workflows (skills, preflight scripts, hooks) so engineering repos don't each carry their own .claude/skills/.

Plugins

Plugin	Description
toolkit-core	Toolkit CLI foundation: install, PATH setup, project init, JDBC connection/datasource configuration. Required by all other toolkit plugins.
toolkit-pipeline	Dataspec-driven pipeline generation via toolkit agent discovery and toolkit agent pipeline-build: SQL, dbt, or PySpark transforms plus data-quality tests and a synthetic-data spec.
informatica-etl-conversion	Informatica PowerCenter conversion: survey the mappings (collect, lineage, ds migration-plan) and convert them wave by wave via agent etl-extract, discovery with a grouped review queue, and pipeline-build — or convert a handful of mappings in a single pass.

Install — Claude Code

Inside Claude Code:

/plugin marketplace add phdata/agent-marketplace
/plugin install toolkit-core@phdata
/plugin install toolkit-pipeline@phdata
/plugin install informatica-etl-conversion@phdata

Dependencies install automatically from this marketplace: toolkit-pipeline pulls in toolkit-core, and informatica-etl-conversion pulls in both.

For team/repo-wide auto-install, check a .claude/settings.json into the consuming repo — see docs/team-setup.md.

Install — Cortex Code

Cortex Code consumes the same plugin format but not the marketplace catalog, so install each plugin by its subdirectory:

cortex plugin install github:phdata/agent-marketplace/toolkit-core
cortex plugin install github:phdata/agent-marketplace/toolkit-pipeline
cortex plugin install github:phdata/agent-marketplace/informatica-etl-conversion

Notes:

• Append @<branch> to pin a branch (e.g. github:phdata/agent-marketplace/toolkit-core@feature/initial_plugin while testing pre-release changes); the subdirectory path goes before the @.
• Cortex does not auto-install dependencies — install toolkit-core explicitly alongside toolkit-pipeline, and all three for informatica-etl-conversion.
• cortex plugin update re-fetches from the pinned source; cortex plugin list shows what's installed and enabled.
• Teams can also ship these via a Snowflake connection profile — see docs/team-setup.md.

Prerequisites

The plugins detect a missing or misconfigured Toolkit CLI and fix it with you (/toolkit-core:setup), delegating installation to the official phData install scripts for macOS, Linux, and Windows — see the install docs. Most toolkit commands also need an auth token (toolkit auth); some features are license-tiered (free vs pro).

Contributing

• One directory per plugin, manifest at <plugin>/.claude-plugin/plugin.json.
• No top-level bin/ directory. Plugins distributed through claude.ai organization settings are rejected at sync if they ship one, because bin/ is added to the Bash PATH without appearing on the admin approval surface. Put executables in scripts/ instead. They are then not on PATH, and ${CLAUDE_PLUGIN_ROOT} resolves only in hook commands — never in the Bash tool — so a skill must be told the absolute path. toolkit-core's SessionStart hook prints its scripts' paths into every session for exactly this reason.
• Run claude plugin validate <plugin-dir> before opening a PR.
• Local development: claude --plugin-dir ./<plugin-dir>, then /reload-plugins to pick up edits.
• Releases are git tags of the form <plugin-name>--v<version>, pushed with claude plugin tag --push.