Repo Prep Skill for Claude

Repo Prep

Interactively scaffold the legal, metadata, community, and compliance files a

repository needs before publication — consistently, from a central author

profile, with author choices respected per repo.

When to use

• Preparing a new or existing repo for GitHub / open-sourcing.
• Adding missing LICENSE, NOTICE, AUTHORSHIP, README sections, or package metadata.
• Writing a defensible AUTHORSHIP record for AI-assisted work (the wizard).
• Adding community docs or .github/ templates.
• Checking EU/Germany compliance obligations for a repo.

Operating principles

• Interactive, not assumed. Ask the user for choices (see *Asking questions*

below); respect the author profile's defaults but confirm per repo. Don't

blindly dump templates.

• Idempotent. Detect what already exists; skip or refresh, never clobber

silently. Show a plan before writing.

• Evidence over assertion (especially authorship): mine the repo for facts.
• Confirm outward-facing actions (anything via gh, pushes) before running.
• Follow the user's git conventions — commit only when asked.

Asking questions (interview transport)

Prefer cenno when it is installed — it gives a nicer panel, progress dots,

and voice input (great for the reflective authorship answers).

• Detect cenno once per run: ToolSearch for

mcp__cenno__ask_sequence. If the tools resolve, use cenno; otherwise fall

back to AskUserQuestion (choices) / plain conversational prompts (free text).

• When using cenno:
• Multi-question flows (the authorship wizard, scope selection) →

mcp__cenno__ask_sequence with flow: "question"; it auto-fills progress dots.

• Single questions → mcp__cenno__ask_user.
• Use input.kind: "choice" + choices for pick-one; "confirm" for yes/no;

"voice_text" for reflective free-text (Decisions, Judgment, Direction, Art

Direction); "scale" for ratings.

• Parse the returned JSON ({answers: [...]} / {answer, ...}). Handle

{answered: false} (timeout) by retrying or falling back.

• If cenno is wanted but not installed, mention it can be set up via the

cenno skill; do not block — fall back gracefully.

Central author profile (single source of truth)

Identity, promo links ("ads"), and defaults live in

~/.config/personal-os/repo-prep/profile.toml (override with REPO_PREP_PROFILE).

• On first run, if it doesn't exist, copy assets/templates/profile.example.toml

there and walk the user through filling it (name, email, github, url, promo

projects, defaults). It is pre-seeded for this author.

• Read it at the start of every run to resolve {{author}}, {{email}},

{{github}}, the promo block, and default choices.

Workflow

• Detect — inspect the repo: manifest/ecosystem (pyproject.toml,

package.json, Cargo.toml, go.mod), existing files, git remote, default

branch, whether it's already on GitHub. Read the author profile.

• Scope — confirm which components to do via AskUserQuestion (core /

community / .github / promo / compliance / gh remote). Pre-fill from profile

defaults.

• Resolve placeholders — once, per references/checklist.md.
• Generate — for each chosen component, create files from

assets/templates/ and assets/github/, filling placeholders. Use

scripts/fetch_license.py for LICENSE. Edit package metadata inline.

• Authorship — if doing AUTHORSHIP.md, run the wizard (below), don't just

fill the template.

• Compliance — if requested or the author is EU-based, run the gating

questions in references/eu-germany-compliance.md and add only what applies.

• gh remote (optional) — confirm, then create/configure the repo.
• Verify — run the test suite (metadata edits can break builds); show the

diff; commit only if asked.

Full item-by-item detail, placeholder table, and ecosystem detection:

references/checklist.md (read it before generating).

The authorship wizard

A first-class feature. To produce a defensible AUTHORSHIP.md for AI-assisted

work, read and follow references/authorship-wizard.md. It:

• gathers external evidence first (ADRs, git log, changelog, AI session logs,

CLAUDE.md/AGENTS.md, design tokens) and offers a draft to edit;

• asks the jurisdiction first, then interviews across five sections —

**Decisions, Exercise of Judgment, Goal Setting & Direction, Art Direction,

AI Implementation**;

• writes a Legal & Copyright section with jurisdiction-aware framing

(US human-authorship bar, EU/Germany "own intellectual creation", UK CGW);

• always states it is not legal advice.

Output template: assets/templates/AUTHORSHIP.md.

EU / Germany compliance

references/eu-germany-compliance.md covers CRA (SBOM, vuln reporting), AI Act

Art. 50 transparency, German Impressum (DDG), GDPR, the Product Liability

Directive, and the EAA. Obligations are conditional — ask the gating

questions and add only what applies. Not legal advice.

Resources

• assets/licenses/ — bundled MIT, BSD-3-Clause (offline); others fetched.
• assets/templates/ — AUTHORSHIP, CODE_OF_CONDUCT, CONTRIBUTING, SECURITY,

NOTICE, CHANGELOG, profile.example.toml.

• assets/github/ — PR + issue templates, dependabot, CI workflows (python/node).
• scripts/fetch_license.py — write a filled LICENSE for any SPDX id.
• references/ — checklist.md, licenses.md, authorship-wizard.md,

eu-germany-compliance.md.