/create-lookalike

name: create-lookalike
description: |
  Create a look-alike audience from a seed audience and a candidate
  population dataset. Classifies Rosetta Stone attributes, generates
  the same materialized-view scoring pipeline Lookalike Studio emits
  (Naive-Bayes categorical weights + Gaussian continuous similarity),
  gates on approval, submits via `narrative_workflows_create`, and
  monitors the build to completion.
  Use when: "create a lookalike audience", "find more users like this
  segment", "expand my seed audience to 500k similar users", "score
  the population against my customers", "build a look-alike of
  dataset X".
  (narrative-audience)
license: MIT
compatibility: >-
  Requires the narrative-mcp MCP server and local file Read. Recommends
  AskUserQuestion (a Claude Code primitive; prose fallback in
  references/HARNESS_FALLBACK.md), the narrative-knowledge-base MCP
  server, and a shell with python3 (3.8+) to run
  scripts/lookalike_state_tag.py for UI re-edit support. Uses the harness waiting tools (job_monitor / wait_for / sleep) when
  present, and paced status checks when not. Portable to
  any agentskills.io-compliant harness via the documented fallbacks.
metadata:
  version: 0.1.3
  narrative:
    args:
      - name: "--seed"
        value: "<id|name>"
        required: false
        description: >-
          The seed audience dataset — the users the output should look
          like. Numeric ID or datasetName. If omitted, the skill asks.
      - name: "--population"
        value: "<id|name>"
        required: false
        description: >-
          The candidate population dataset to score and select from.
          Numeric ID or datasetName. If omitted, the skill asks.
      - name: "--size"
        value: "<N>"
        required: false
        description: >-
          Output mode "size": keep the top N highest-scoring candidates.
          Mutually exclusive with --min-score.
      - name: "--min-score"
        value: "<0..1>"
        required: false
        description: >-
          Output mode "score": keep every candidate whose similarity
          probability is at or above this threshold (clamped to
          0.001–0.999 and converted to log-odds internally). Mutually
          exclusive with --size.
      - name: "--include-seed"
        required: false
        default: false
        description: >-
          Union the seed members back into the output audience with a
          perfect score of 1.0. Off by default — the output is
          look-alikes only.
      - name: "--name"
        value: "<audience name>"
        required: false
        description: >-
          Unique name for the output audience dataset. Slugified to form
          the pipeline prefix. Must not collide with an existing dataset.
      - name: "--data-plane"
        value: "<id>"
        required: false
        description: "UUID of the data plane to target. Skips data-plane resolution."
      - name: "--no-trigger"
        required: false
        description: >-
          Create the workflow without triggering a run. Default is
          trigger_immediately: true — the audience builds right away.
      - name: "--no-state-tag"
        required: false
        description: >-
          Skip the _nio_lookalike_serialization wizard-state tag on the
          output audience. Default is to include it (via
          scripts/lookalike_state_tag.py) so Lookalike Studio can reopen
          the audience for editing.
      - name: "--dry-run"
        required: false
        description: >-
          Render and explain the full pipeline but do NOT submit.
          Implies --show-spec.
      - name: "--show-spec"
        required: false
        description: >-
          Include the full rendered workflow YAML in the approval
          preview. Off by default; the plain-English summary is enough
          for most users.
      - name: "<free-text tail>"
        required: false
        description: >-
          The user's intent (e.g., /create-lookalike 100k users like my
          premium_subscribers from the acme_population dataset).
    requires:
      tools:
        - Read
      mcp-servers:
        - narrative-mcp
      mcp-tools:
        - narrative_context_get
        - narrative_datasets_search
        - narrative_datasets_describe
        - narrative_dataset_get_column_stats
        - narrative_attributes_search
        - narrative_nql_validate
        - narrative_data_planes_list
        - narrative_workflows_create
        - narrative_workflow_runs_list
        - narrative_jobs_search
        - narrative_jobs_describe
    recommends:
      skills:
        - narrative-common:generate-rosetta-stone-mappings
      tools:
        - AskUserQuestion
        - Bash
        - Write
      mcp-servers:
        - narrative-knowledge-base
      mcp-tools:
        - narrative_context_search_companies
        - narrative_context_set_company
        - search_narrative_i_o_knowledge_base

<!-- AUTO-GENERATED from SKILL.md.tmpl — do not edit directly --> <!-- Regenerate: bun run gen:skill-docs -->

Create Lookalike

Persona

You are an audience modeler who turns "find me more users like these" into a deterministic scoring pipeline. You optimize for:

1. Pipeline fidelity — every materialized view is rendered from the fixed stage templates in `references/PIPELINE.md`, the same shapes Lookalike Studio generates. You substitute names and attributes into the templates; you do not redesign the statistics. 2. Defensible attribute selection — features enter the model only when the classification rules say they're eligible, and the user sees and approves the feature set before anything is built. 3. Transparency before submit — the user approves a plain-English description of the pipeline, the output configuration, and the data plane before anything is created server-side.

You never invent an attribute, column, or dataset name, never submit without approval, and never claim the aud