Créer une compétence dbt-lineage

---

description: Scaffold a new product skill under skills/dbt-lineage/ matching the repo's SKILL.md conventions. argument-hint: [skill-name] [one-line purpose]

Scaffold a new product skill for the dbt-lineage Claude Code plugin (the skills shipped to end users of the CLI — not contributor tooling).

Requested skill: $ARGUMENTS

Before writing — gather what you can't infer

Read one existing skill of the same type as a template (skills/dbt-lineage/skills/*/SKILL.md) and the plugin README (skills/dbt-lineage/README.md). Then determine — asking the user only if not already clear from the request:

1. Slug — kebab-case, becomes both the directory name and frontmatter name.
2. Type — one of:
   • command — wraps a specific CLI invocation, deterministic (e.g. trace-column).
   • agent — judgment-heavy, multi-step, may branch (e.g. change-safety-review).
   • helper — a precondition/utility other skills call (e.g. refresh-artifacts).
3. The question it answers in the user's words ("what breaks if…", "where does…").
4. CLI surface it wraps — the exact dbt-col-lineage command(s) and which JSON/ Markdown fields it reads. Verify the flags against the current CLI (poetry run dbt-col-lineage --help), not memory.
5. Sibling disambiguation — which existing skill is it most likely confused with, and how does its description draw the line?

Create skills/dbt-lineage/skills/<slug>/SKILL.md

Match the house style exactly:

---
name: <slug>
description: <what it does>. Use when the user asks "<phrase 1>", "<phrase 2>", or <situation>. <One clause distinguishing it from the closest sibling skill>.
---

# <Title Case Name>

<One or two sentences framing the job and, if relevant, pointing to the sibling
skill for the adjacent job.>

## Inputs
- `<name>` (required/optional, default …).

## Steps
1. Ensure artifacts are fresh (`refresh-artifacts`).   # include for any lineage-reading skill
2. <run the CLI invocation, in a fenced bash block>
3. <read these fields from the output>

## Output
- <the deliverable: what Claude should synthesize, and any ⚠️ flags to raise>

Conventions that matter:

• The description is the trigger — it's how Claude auto-selects the skill. Front-load what it does, list concrete user phrasings, and end with the disambiguation clause.
• Lineage-reading skills start with the refresh-artifacts precondition (stale artifacts are the #1 cause of wrong results).
• Prefer --format json for skills that compute; Markdown when the report is the deliverable.
• Any external side effect (posting a PR comment, rebuilding artifacts) → confirm with the user first.

Finish

1. Add a row to the Skills table in skills/dbt-lineage/README.md (| slug | type | answers |).
2. Bump skills/dbt-lineage/.claude-plugin/plugin.json version only if the user is cutting a plugin release (ask; don't assume).
3. Show the created file and the README diff, and suggest a feat(skills): add <slug> skill Conventional Commit — but do not commit unless asked.