AI Native Dev

A repo-native workflow loop for AI agent development.

Align context, resolve uncertainty, review the plan, validate behavior, and preserve commit intent in the repository instead of losing it in chat.

AI Native Dev workflow loop: project context, spec, spec review, plan, plan review, test evidence, commit, then Repo; prototype sits above the main flow and loops back to spec review when requirements are unclear.

Install

Install the repo protocol layer locally.

The local path gives teams a concrete way to apply the protocol without an account, a hosted backend, or a public GitHub repository requirement.

Install in Codex

npx skills add https://skill.ferryman.app --global --yes --agent codex

Install for Claude Code

npx skills add https://skill.ferryman.app --global --yes --agent claude-code

Inspect protocol actions

npx skills add https://skill.ferryman.app --list

First step

Initialize project instructions with vibe-project.

After installing, open a new Codex thread or reload your agent session, then run this in the target repo.

Copy first-run prompt
Use vibe-project to create or update project instructions for this repo. For Codex, update AGENTS.md; for Claude Code, update CLAUDE.md.
  • Codex AGENTS.md
  • Claude Code CLAUDE.md

Product model

From shared context to reviewable engineering work.

01

Repo-Native Artifacts

Project instructions, specs, prototypes, plans, reviews, validation evidence, and commit intent live where agents and engineers can both read them.

02

Organization Skills

Reusable workflow units define how context is aligned, unknowns are prototyped, specs and plans are reviewed, tests are evidenced, and commits preserve intent.

03

Trusted Workflow Events

Future hooks and CI can emit compact, auditable events tied to specs, reviews, plans, tests, commits, outcomes, and human decisions.

04

Engineering Observability

Teams can eventually see adoption, spec lead time, review gates, test evidence, and commit quality without reading every chat.

Engineering observability

Measure flow and quality, not AI usage theater.

The current product is a local skill bundle. The longer-term value is that the same artifacts can become trusted workflow events, so teams can see whether work moved from intent, to reviewed spec, to reviewed plan, to tested and reviewable commits.

Workflow step

  • Spec creation
  • Spec readiness review
  • Implementation and validation planning
  • Plan readiness review
  • Development
  • Test evidence
  • Code and evidence review
  • Commit preparation
  • CI coverage

Adoption signal

  • skill name
  • event type
  • artifact path
  • plan_ref
  • outcome
  • timestamp
  • evidence_ref

Quality / efficiency question

  • Was the spec protocol used?
  • How long until the spec was ready?
  • Was implementation and validation planned and reviewed before development?
  • Did development follow the plan?
  • Is coverage or evidence present?
  • Was code reviewed before commit?
  • Did the commit preserve intent?

Protocol execution layer

Make the development loop preserve context by default.

Each workflow action turns fragile chat context into a repeatable repo-native step: align project instructions, define the feature, prototype unclear points, review specs and plans, capture test evidence, review code, hand off when needed, preserve commit intent, and prepare release notes.

Workflow skill

vibe-project

Create project-level agent instructions so coding agents share the same operating rules and repository context.

Workflow skill

vibe-spec

Turn product intent, issue notes, screenshots, prototypes, and tradeoffs into a repository-native feature spec.

Workflow skill

vibe-review

Review specs, plans, code diffs, docs, or workflow artifacts with a target-specific readiness gate.

Workflow skill

vibe-plan

Turn a ready spec into vertical implementation slices with validation, architecture checks, and human-review points.

Workflow skill

vibe-prototype

Answer one uncertain product, UI, backend, agent workflow, prompt, or architecture question with disposable evidence.

Workflow skill

vibe-test

Inspect coverage, run test cases, capture validation evidence, and recommend test improvements.

Workflow skill

vibe-screenshot

Generate store, release, and marketing screenshots from real app UI with composed frames and visual QA.

Workflow skill

vibe-handoff

Move active work between agent sessions with concise artifact pointers, validation status, blockers, and next steps.

Workflow skill

vibe-commit

Preserve intent, changed surface, and validation evidence in commit messages instead of losing them in chat.

Workflow skill

vibe-release

Create concise release documents with bilingual release notes, shipped changes, validation, risks, rollback, and links.

Local-first

The protocol layer runs inside your project.

Install once; the workflow lives in your repo, where every coding agent reads the same context.