init

name	init
description	Initialize or bootstrap a repository with Devflow Native presets by calling devflow init and interpreting the result.

Devflow Init

Use this skill when a maintainer wants to set up Devflow Native in a new or existing repository, especially when they mention init, bootstrap, preset, review policy, CI, or native Codex/Claude harness files.

This skill is a thin wrapper over the Devflow CLI. The CLI owns policy, file generation, review config, gate inference, CI workflow output, and native harness file installation. The skill chooses a preset, calls devflow init, and explains the result.

Preset Choice

• Use solo-product for product repositories where feature changes should normally use PR review, review.required should be enabled, CI should run, and AGENTS.md should record direct-main exceptions.
• Use research for paper, experiment, benchmark, or evaluation repos where evidence logs, fixtures, reproducibility, and bench gates matter most.
• Use content-site for docs, blog, or marketing-content repos where small writing edits can go direct main but build, lint, link, design, SEO, route, or feature changes still need gates and review.

Steps

1. Run devflow doctor --json and apply the local execution contract.
2. Inspect package scripts and existing instructions with devflow status --json plus targeted file reads when needed.
3. Pick the narrowest preset that matches the repo.
4. Render a dry run first, for example:

devflow init --preset solo-product --targets codex,claude --ci github --review required --json

5. Explain planned writes, skipped files, inferred gates, CI choice, and whether AGENTS.md will be created or augmented.
6. Only run the confirmed write when the maintainer asked to proceed or the workflow is already confirmation-gated:

devflow init --preset solo-product --targets codex,claude --ci github --review required --confirm --json

7. Verify with devflow health --json. If native agent files were installed, also run devflow harness health --targets codex,claude --json.

Output

Report:

• chosen preset and why
• files written, updated, skipped, and ignored
• inferred gates and CI workflow
• review policy
• next command, such as health, harness health, or finish evidence recording