640‑skills – SBTD workflow for AI coding agents

What it is

A collection of Skills (configuration modules) that implement the SBTD (Specification‑Behavior‑Test‑Domain) workflow for AI‑driven code generation agents such as Codex, Oh My Pi (OMP), Claude Code, and Kimi.
The repo does not contain a runnable application; it provides the Onboard skill and a catalog of 14 bundled skills plus references to 19 external skills that together make the workflow reproducible.

Core idea

Connect a project's specifications, behavioural rules, tests, and domain language so that an AI coding agent can act autonomously while the human retains explicit control over scope, permissions, and verification.
All actions are expressed as natural‑language prompts that the agent interprets; the repository supplies the prompt templates and the underlying skill definitions.

Key components

Component	Role
`sbtd-workflow-onboard`	Bootstrap skill that installs the rest of the catalog, validates the environment, and creates per‑project `AGENTS.md` files.
Bundled skills (14)	Core SBTD steps (e.g., spec ingestion, test generation, rule enforcement, evidence collection).
External skills (19)	Required third‑party skills that the workflow depends on; the repo lists their identifiers and version pins.
`skills` CLI integration	Uses the Vercel‑labs Skills command‑line tool to add, list, and manage skills for a given agent (e.g., `codex`, `omp`).
Reference docs (`REFERENCE.md`)	Detailed command‑line syntax, two execution modes (global vs. project‑only), and the exact prompts for each workflow operation (init, upgrade, reset, migration, cleanup, recovery).

Installation flow

Bootstrap the Onboard skill – either paste the provided natural‑language instruction into your coding agent or run the official Skills CLI:
```
npx --yes skills@latest add https://github.com/KunoLu/640-skills \
  --skill sbtd-workflow-onboard --global --agent codex --yes --copy
```
Review the installed directory – confirm the Python version (≥3.10) and run pip install -r <path>/requirements.txt to satisfy the skill’s Python dependencies.
Initialize projects – use the init or init-projects prompts (see README) to install AGENTS.md and any required global skills for one or more absolute project paths.
Subsequent operations (upgrade, reset, migration, cleanup, recovery) are all driven by prompt templates that the agent executes; the README provides ready‑to‑copy text for each scenario.

Typical usage

First‑time setup – copy the bootstrap prompt, let the agent verify the tag (v2.0.1), then run the init prompt to install all skills globally and create per‑project configuration.
Add a new project – run the init-projects prompt with the target platform (codex or omp) to only affect the specified project, leaving global skill installations untouched.
Upgrade – invoke the upgrade prompts to generate a plan, review the JSON plan, approve it, and then apply it. The workflow separates planning, applying, and verification phases.
Migration from SBTD v1 – a dedicated set of prompts guides a safe migration, keeping legacy data separate until explicit approval.
Cleanup & recovery – prompts exist to plan legacy‑asset cleanup or to restore a previously approved batch of changes.

Why it matters

Provides a formal, evidence‑driven process for letting LLM‑based coding agents modify codebases, reducing accidental overwrites.
Makes the scope of each operation explicit (global vs. project‑only, backup policies, permission checks), which is crucial for teams that need auditability.
Works across multiple host agents (Codex, OMP, Claude Code, Kimi) by abstracting the underlying CLI via the Skills framework.

Where to find more

All information above is taken directly from the repository’s README; no additional features have been inferred.

KunoLu/640-skills