developer skill pack

Abstract

Complete development toolkit — UX testing, multi-agent coordination, TDD workflow, GitHub templates, research spikes, and repo scaffolding. Works with Claude Code and OpenAI Codex CLI.

Developer Skill Pack – Technical Briefing

The Problem

Teams that rely on AI‑assisted coding often lack a single, repeatable workflow that ties together UX testing, multi‑agent coordination, TDD, and repo scaffolding. The repository supplies a set of “skills” that aim to close those gaps, but the plumbing and execution model are not obvious from the file list alone.

What This Does

The pack bundles UX testing (web and native), a coordinator that manages a swarm of agents, TDD helpers, and scaffolding tools. Each skill is a self‑contained directory with a SKILL.md that describes its intent and a set of supporting files (schemas, templates, scripts). For example, skills/ux-walker contains SKILL.md, a references folder with audit scripts, and a templates/latest-report-template.md for output. The skills/coordinator directory hosts the core orchestration logic: claude-coordinator (shell driver), coord-validate (schema checker), and coordinator-settings.json (run‑time config).

How It Is Wired

Entry Point – bin/claude-coord The binary is a wrapper that delegates to skills/coordinator/claude-coordinator. This script pulls coordinator-settings.json, validates it against skills/coordinator/schemas/coordinator-output.schema.json, and then launches each agent defined in the agents directory.

Agent Invocation – skills/coordinator/claude-coordinator The coordinator script iterates over the list of agents (e.g., agents/briefer.md, agents/issue-groomer.md). Each agent is a Markdown spec that the coordinator interprets and, where applicable, invokes a corresponding shell or JavaScript helper. For instance, agents/issue-groomer.md triggers the issue-groomer-settings.json and the test script skills/coordinator/test/issue-groomer-agent.test.sh.

External Effects –

Filesystem – The coordinator writes reports to a .worktreeinclude‑controlled directory, creating JSON output per agent schema.
Network – Agents that interact with GitHub (e.g., agents/issue-implementer.md) hit the GitHub API to create or update issues and PRs.
Command‑line – The bin/claude-coord script accepts --help and --config flags, though no explicit usage is documented in the repo.

Responsibility Map

File	Responsibility	External Impact
`bin/claude-coord`	CLI entry point	Spawns coordinator
`skills/coordinator/claude-coordinator`	Orchestrator	Drives agents, writes output
`skills/coordinator/coordinator-settings.json`	Runtime config	Determines agent list
`agents/*`	Agent logic	GitHub interactions, file generation
`skills/coordinator/schemas/*.schema.json`	Validation	Ensures output correctness

The graph shows a single hub (claude‑coordinator) with a fan‑out to agents, minimizing cycles. A change to the coordinator would ripple to all agents; an agent change is isolated.

How To Use It

Setup No build system is present. The repo relies on shell scripts and the installed claude CLI. Clone with:

git clone https://github.com/moses-y/developer-skill-pack

Configuration Create a .env file at the repo root (not committed) with:

CLAUDE_API_KEY=your-key
GITHUB_TOKEN=your-token

Place a coordinator-settings.json in skills/coordinator if you want custom agent lists.

Running Invoke the coordinator:

./bin/claude-coord --config skills/coordinator/coordinator-settings.json

The script will validate the JSON, launch each agent, and write results under skills/coordinator/.worktreeinclude.

Testing Run the provided shell tests:

./test/launcher.test.sh
./test/plugin-manifest.test.sh

These exercise the CLI wrapper and manifest generation.

Real‑World Use

A product team can add skills/ux-walker to its nightly CI (e.g., GitHub Actions) by calling ./bin/claude-coord --config skills/coordinator/coordinator-settings.json. The output JSON feeds into a dashboard that flags visual regressions, while agents/issue-implementer.md automatically opens GitHub PRs to fix them.

Code Health & Issues

Missing CI – No .github/workflows or GitHub Actions configuration.
No automated tests – Only shell tests exist; no unit tests for JSON schemas.
Documentation gaps – README.md lists skills but does not detail CLI flags or environment variables.
License – MIT (present in LICENSE).

The Bottom Line

The repo delivers a modular, agent‑driven workflow that can be wired into existing CI pipelines. Its strengths lie in the breadth of UX‑centric skills and the clear separation of configuration from execution. However, the lack of automated pipeline configuration, unit tests, and explicit documentation for environment variables limits out‑of‑the‑box usability. Teams comfortable with shell scripting and AI‑driven tooling will find it useful, while others may need to add CI and richer documentation to adopt it fully.