Coding agents author px workspaces well when they know the file model, the concept kinds, the datasource rules, and the schemas. px skill install gives them exactly that — generated from the installed px, so it always matches your version and needs no repo to clone. The flow is: install px → add the skill → profit.

​
What it installs

The skill is generated from the same bundled schemas and curated guide that px validate enforces, so it can’t drift from what the CLI accepts.
Writes ~/.claude/skills/prometheux/:
  • SKILL.md — the authoring guide + command reference
  • reference/schemas.md — the field-by-field schema reference
  • reference/*.schema.json — the raw JSON Schemas
Reload skills or restart Claude Code to pick it up.
Install several at once by repeating -t:

​
When it activates

The skill’s description triggers it whenever you work on Vadalog, Prometheux concepts / ontologies / datasources / apps, lineage-as-code, the context layer, or any px command. You don’t invoke it manually — the agent pulls it in when the task matches.

​
What the agent can then do

With the skill installed, an agent drives the same loop you would:
Keep the skill current after upgrading px: run px skill install --force again to regenerate it from the new version’s schemas.

​
Refreshing / removing

  • Refresh: px skill install --force overwrites an existing install.
  • Remove: delete the installed path (~/.claude/skills/prometheux/, ./.claude/skills/prometheux/, or ./.cursor/rules/prometheux.mdc).

​
Interactive access via MCP (optional)

The skill teaches file-based authoring. For interactive in-chat operations (list, run, or query concepts without files), you can also connect your agent to the platform’s MCP server — e.g. Claude Code or Cursor. The two compose: MCP for interactive ops, px (with the skill) for file-based authoring and deploy. Neither requires the other.