<!-- Copyright (c) 2026 JG Systems Consulting Ltd. — MIT License (see LICENSE). See LICENSE for terms. -->

JGS Goal-to-Spec Kit

Copyright (c) 2026 JG Systems Consulting Ltd. — MIT License (see LICENSE).

A small, composable suite of Claude Code skills that turn a rough idea into a hardened engineering specification — ready to execute. Three skills chain together:

flowchart TD
    A([Rough idea]) --> B["/goal-formatter"]
    B -->|"structured, testable goal"| C["/goal-spec"]
    C -->|"complete spec — gaps resolved by targeted questions"| D["/spec-review"]
    D -->|"hardened spec — autonomous multi-discipline review"| E([Execute with your executor of choice])
    classDef skill fill:#16171a,stroke:#3a3e46,color:#e8ebf0;
    classDef bookend fill:#f4f2ec,stroke:#f4f2ec,color:#0a0a0b;
    class B,C,D skill;
    class A,E bookend;

The pack's job is to get you to a spec you can trust enough to build from. The skills do not implement the spec themselves — execution is a separate, deliberate next step (see "Then execute it" below). Authoring once and executing many ways is the point.

[!NOTE] These skills are tool-agnostic: no external dependencies, no MCP server — pure prompt-engineering workflows that work with any model or agent framework.

Install with your AI agent

Copy everything in the block below and paste it into your coding agent (Claude Code, Cursor, etc.). It will read this repo and install the pack for you.

You are installing jgs-goal-spec, an open-source (MIT) Claude Code skills pack
(JG Systems Consulting Ltd.) that turns a rough idea into a hardened engineering
spec via three chained skills (goal-formatter → goal-spec → spec-review).
Repository: https://github.com/jgsystemsconsulting/jgs-goal-spec (version 1.2.0). Do this in order:

1. Read README.md, docs/skill-usage.md, and CHANGELOG.md so you understand what
   you are installing. There are NO external prerequisites — these skills call no
   MCP server and need no licence tier at runtime.
2. Run `python install.py --dry-run` and show me the skill list and target
   (~/.claude/skills/jgs-goal-spec/). If it looks right, run `python install.py`.
3. Verify: list the installed skills under the target and confirm the count matches SKILLS.md.
4. Tell me to restart Claude Code, then start with `/goal-formatter`.
5. This is MIT-licensed open-source software. Do not edit anything outside the install target.

What's in this distribution

LICENSE, COPYRIGHT, NOTICE             Legal & attribution
README.md                              This file
SKILLS.md                              Skill index (auto-generated at release time)
CHANGELOG.md                           Version history
install.py / install.sh / install.ps1  Installer scripts
skills/goal-formatter/                 Idea → structured, testable goal (+ optional draft spec)
skills/goal-spec/                      Goal → complete spec (asks gap-filling questions)
skills/spec-review/                    Spec → hardened spec (autonomous multi-discipline review)
docs/
  └── skill-usage.md                   How to invoke and chain the skills, and how to execute the result

Prerequisites

1. Claude Code (or another skill-capable client). spec-review spawns reviewer subagents,

so a client that supports subagents gets the full autonomous loop; without subagents it can run the personas inline (one model voicing each), which is faster but less independent.

1. Nothing else. No MCP bridge, no external service, no licence-tier check at runtime.

Installation & Invocation

Option A — Plugin marketplace (recommended):

/plugin marketplace add jgsystemsconsulting/jgs-goal-spec
/plugin install jgs-goal-spec@jgs-goal-spec

Skills are namespaced when installed this way — invoke them as /jgs-goal-spec:goal-formatter, /jgs-goal-spec:goal-spec, /jgs-goal-spec:spec-review.

Option B — Installer script:

python install.py        # Claude Code (default); or ./install.sh / .\install.ps1

Installs into ~/.claude/skills/jgs-goal-spec/; invoke flat as /goal-formatter, /goal-spec, /spec-review. Restart Claude Code after installing.

Use with other agents

These skills aren't Claude-only. Because they're pure prompt workflows in the portable SKILL.md format, the installer can target several agents — some read SKILL.md natively, others get an automatic format transform:

python install.py --list-agents       # show every target and where it installs
python install.py --agent openclaw     # SKILL.md-native: copied unchanged
python install.py --agent gemini       # transformed to that agent's format
python install.py --agent all          # all user-global agents

Transform targets keep the full workflow: any reference files (e.g. goal-spec's spec template) are inlined so each generated prompt is self-contained. Full details, paths, and limitations: see docs/other-agents.md.

Quick usage

In any Claude Code conversation after installation:

1. /goal-formatter — describe your idea in plain language; get back a structured, testable

goal. Add "and a spec" (or any spec-style ask) to also get a draft specification.

1. /goal-spec — turn that goal into a complete engineering spec. It asks you targeted

clarifying questions for anything the goal left undecided, grounded in what you said you want.

1. /spec-review — harden the spec autonomously: a roundtable of engineering-discipline

reviewers (architecture, security, test/QA, ops, product) critiques it round after round until no critical or major issues remain — no check-ins required.

Full index: see SKILLS.md. Usage guide: see docs/skill-usage.md.

Then execute it

The hardened spec is the input to execution, not the execution itself. Once spec-review reports convergence, hand the goal + hardened spec to an execution agent:

• Claude Code `/goal` — paste the goal; point it at the spec as the detailed design.
• An autonomous loop (autopilot / ralph-style runner) — feed it the spec as the source of

truth and let it implement to the spec's Success Criteria.

• An `executor` subagent — dispatch with the spec inline; verify its output against the

spec's Success Criteria.

The spec's Success Criteria are the acceptance gate: execution is done when every criterion is observably met. That's how the loop closes on a built outcome, not just a written one.

Support & Security

For licensing enquiries, commercial licences, or custom-skill development: JG Systems Consulting Ltd.

To report a security issue, see SECURITY.md or email [email protected] privately; please do not open public issues for vulnerabilities.

Legal

This software is released under the MIT License — you are free to use, copy, modify, and distribute it, subject to the licence terms.

"JG Systems Consulting" is a trademark of JG Systems Consulting Ltd.