Greenfield plugin
A Claude Code plugin that extracts behavioral specs, test vectors and acceptance criteria from a codebase, with a provenance trail
- 296 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Agents read all analyzed code and can take binaries apart or run the target program in a container
- Writes many files into the workspace and spends many tokens
Install
- Works in
Manual install
/plugin marketplace add prime-radiant-inc/prime-radiant-marketplaceThen /plugin install greenfield@prime-radiant-marketplace and restart Claude Code.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Agents read all analyzed code and can take binaries apart or run the target program in a container
- Writes many files into the workspace and spends many tokens
This is third-party code. Review the repository files before installing.
What it does
Greenfield reads source code, documentation, SDKs, runtime behavior and binaries and describes what the software does, leaving out how it is built inside. The output is behavioral specifications, test vectors, acceptance criteria and a full provenance trail. A team can build its own implementation from these materials without inheriting the original's structure. The /analyze command with a project path runs a seven-layer pipeline. First it discovers available sources and gathers evidence from each. Then it writes specs with citations and builds test vectors and acceptance criteria. Finally the specs are cleaned of implementation detail and go through a second review. Two agent types do the work: the analyzer reads evidence and writes specs, and the sanitizer rewrites raw specs into clean ones. Gates between layers check for contradictions, completeness, implementation leakage and detail lost in sanitization. The result lands in a three-part workspace, and the /sanitize command cleans it again when the audit finds leftovers of the original.
Who it is for. For teams that need to document the behavior of an existing product or their own legacy code in order to rewrite it from specifications.
Good fit when
- A legacy service must be rewritten and its behavior first captured as specs and tests
- You have a minified JavaScript bundle with no sources and the team must learn what it does
- You need acceptance criteria and test vectors to validate a new implementation
- You must keep a trail showing where each statement came from
- Specs must be cleaned of implementation detail before handing them to another team
Not a fit when
- You need a plain description of one function: a seven-layer pipeline is overkill
- The source or product is covered by a license or contract that forbids analysis and you lack permission
- You want the new implementation code itself: the plugin writes specs only
- You cannot run the target software or take its binaries apart on the work machine
Example request
/analyze ./legacy-billing: extract behavioral specs and test vectors, and list the sources in the provenance trailLimitations
The plugin runs inside Claude Code and uses its agents, so token spend on a large project is high. Whether you may recreate someone else's software depends on the original's license and contract, and you need to check that yourself.
Sanitization lowers the chance that traces of the original remain in the specs but does not remove it. The second pass and the /sanitize command exist for that. Observing a running program and taking binaries apart go through separate skills and may need containers.
How to disable. Run `/plugin uninstall greenfield@prime-radiant-marketplace` and restart Claude Code. Workspaces stay where you created them.
FAQ
How do I install Greenfield in Claude Code?
Run /plugin marketplace add prime-radiant-inc/prime-radiant-marketplace, then /plugin install greenfield@prime-radiant-marketplace and restart Claude Code.
What do I get after /analyze?
A workspace in three parts: raw with working material and citations, output with sanitized specs, test vectors and validation, and provenance with the citation trail.
What sources does Greenfield read?
Source code, documentation, SDKs, community discussion, runtime behavior, binaries, git history, tests, UI and contracts. The pipeline has separate paths for minified bundles and decompiled binaries.
Why sanitize the specs?
Sanitization removes details of the original's internals so the new team builds from a behavior description. Second-pass checks look for leftovers of the original's structure and content.
Does the plugin write new code?
No. Greenfield stops at specs, tests and acceptance criteria. A team works from the output folder to build the implementation.
Discussion
No questions or reviews yet.
Related
PluginsA Grid Dynamics plugin that loads a team's shared engineering standards, architecture and workflows into a coding agent session
A plugin for Claude Code and Codex: turns a feature request into a spec and executes it task by task with fresh context
14 skills that find the gaps in a task before the code: interviews, plans, assumption tests and proof that a bug fix works
A plugin with three skills for large development work: project analysis, phased tasks tracked as GitHub Issues, parallel agents and code review
A Claude Code and Codex plugin with skills for code research, tests, refactoring and independent review of finished changes
A CLI that wires a research-to-delivery pipeline with documents and quality gates into Claude Code, Cursor, Codex and other agents
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| Greenfield (this entry) | no | 296 | |
| OpenSpec | no | 73,368 | drives new features through proposals and specs and does not derive specs from existing code |
| Reverse Engineering Skills | no | 2,329 | a skill set for taking binaries apart, with no pipeline to specs and tests |
README in short
The README covers the plugin's purpose, marketplace installation and two commands. It then shows the workspace layout, the roles of the two agents and a table of the seven pipeline layers with gates. The license is Apache 2.0.