content-pipeline-init

Scaffold the content-pipeline OS into the current project. This is the one-time setup that lets the /content-pipeline orchestrator run the project afterward. Genre- and format-agnostic: the project's specifics live in project.md, never in the skills.

Step 0 — don't clobber an existing pipeline

Check whether the project already has pipeline.md or project.md. If either exists, STOP and ask — this project may already be scaffolded. Never overwrite a populated dashboard.

Step 1 — gather the minimum direction (one question at a time)

Ask only what the config needs, one question at a time, each with a recommended answer. Don't over-interview — deep creative direction belongs to the stage skills later, not here. Collect:

1. project slug — short machine name (e.g. my-channel).
2. niche — one line: what this project makes.
3. format — short or long. This selects the creative skill family the orchestrator dispatches to; if the needed family skills don't exist yet, note that and proceed (they can be built later).
4. budget rails — per-item and total USD ceilings (recommend $5 / $200).
5. stage graph — start from the template's default graph; adjust stage ids, the skill: for each, and gate: values (none/pr/release) to match this project. Confirm which stages need a human gate.

Step 2 — write the three files

Copy the templates from this skill's templates/ directory into the project root, filling placeholders:

• project.md — replace PROJECT_SLUG, PROJECT_NAME, ONE_LINE_NICHE, ONE_LINE_VOICE, personas, budget, and the stages: graph with the confirmed values. Keep the frontmatter valid YAML (the stages entries are flow mappings: { id: ..., skill: ..., gate: ... }). Fill commands.* with the real shell command for every run-* stage ({item} = the item's directory), or delete keys for stages the project doesn't have.
• pipeline.md — replace PROJECT_SLUG, set STAGE_SEQUENCE to a human- readable arrow chain of the stage ids with gates marked (e.g. ideate → script(GATE) → storyboard(GATE) → generate → build(GATE) → publish), and set THRESHOLD (stockpile launch target). Leave the table empty — the orchestrator adds rows as items enter the pipeline.
• context.md — copy as-is unless the user already has decisions to record.

Step 3 — verify

• project.md frontmatter parses as YAML. Sanity-check by reading it back; every stages entry must have id, skill, and gate.
• pipeline.md's stage sequence matches the ids in project.md stages.
• No placeholder tokens (ALL_CAPS_UNDERSCORE) remain in any of the three files.

Report the three files written and tell the user the next step: run /content-pipeline to add the first content item and start advancing stages.