jsapartners

claude-essentials

A Claude Code plugin with opinionated agents and skills for shipping code.

Table of Contents

• Background
• Install
• Usage
• Resources
• Maintainers
• Contributing
• License

Background

This plugin codifies how JSA+Partners ships software with Claude Code. Every skill is human-gated. The workflow enforces a disciplined path from idea to pull request: draft a user story, decompose it into single-commit units, then implement, review, document, and commit each unit before shipping the PR.

The review pipeline runs five specialized agents in parallel (complexity, architecture, technical, scope, and language idioms), then passes all findings through an adversarial skeptic agent to eliminate false positives.

Install

Dependencies

• Git
• GitHub CLI (gh) for /essentials:pr
• A CLAUDE.md in your project root (run /essentials:init to generate one)

/plugin marketplace add JSA-Partners/claude-essentials
/plugin install essentials@jsapartners

Usage

The core workflow chains skills together. You review the output at every step and run /clear before moving on. This keeps the context window focused and puts a human in the loop at every stage.

flowchart LR
    Z("/essentials:discuss") -.optional.-> A("/essentials:story")
    A --> B("/essentials:decompose")

    subgraph unit ["for each unit"]
        direction LR
        C("/essentials:implement") --> D("/essentials:review") --> E("/essentials:document") --> F("/essentials:commit")
        F -- next unit --> C
    end

    B --> C
    F --> G("/essentials:pr")

Core Loop

Start with /essentials:decompose story.md to break a story into unit files stored in ~/.cache/claude-essentials/. Then work through each unit one at a time:

1. /essentials:implement <unit-file> to build it. Scan the changes, then /clear.
2. /essentials:review <unit-file> to run the review agents. Address findings, do a manual diff review, then /clear.
3. /essentials:document <unit-file> to capture any learnings. Review the docs, then /clear.
4. /essentials:commit to wrap up the unit.

Repeat for the next unit. When all units are done, run /essentials:pr.

Every command accepts a full unit file path. You can also pass a partial name like auth or unit-02 and the skill will search ~/.cache/claude-essentials/ to find it. Commands work ad-hoc too: /essentials:review src/ reviews a directory, /essentials:document auth-patterns documents a topic.

Skills

Skill	Purpose
/essentials:commit	Generate commit messages matching project patterns. Defaults to conventional commits.
/essentials:decompose	Decompose a plan or user story into single-commit implementation units.
/essentials:discuss	Explore a problem through discussion and research, then produce a prompt for /essentials:story.
/essentials:document	Capture learnings from implementation into project docs, memories, or CLAUDE.md.
/essentials:implement	Implement one decomposed unit with idiomatic patterns and quality gates.
/essentials:init	Generate a CLAUDE.md for the current project based on codebase analysis.
/essentials:meta	Update the /essentials:sharpen skill with new resources.
/essentials:playbook	Inject behavioral plays and execution protocols.
/essentials:pr	Create a GitHub PR matching project conventions. Requires gh CLI.
/essentials:review	Review code with parallel specialized agents, adversarial verification, and human approval.
/essentials:sharpen	Systematically improve skills and agents using proven patterns.
/essentials:story	Create user stories through agent collaboration.

Agents