A multi-agent pipeline for Gemini CLI designed to build strict, logic-based taxonomies. Whether you're training Computer Vision models (Visual Taxonomy) or organizing an e-commerce catalog (Functional Taxonomy), this tool automates the process from brainstorming to final attribute definition.

Most taxonomies fail because they mix logic (e.g., grouping by "Color" and "Usage" in the same level). Taxonomy Architect enforces a "Lens" or "Razor":

• Visual Lens: If it looks the same (silhouette/topology), it stays together.
• Functional Lens: If it solves the same problem, it stays together.

Install directly via Gemini CLI:

gemini extension install https://github.com/Kwentar/taxonomy-architect.git

The process is divided into stages, each handled by a specialized agent:

1. Initialize (/taxonomy:start): Sets up your project in taxonomies/<domain>.
2. Explode (/taxonomy:explode): Generates a massive list of potential items.
3. Razor (/taxonomy:razor): Filters and merges items based on the chosen Lens.
4. Cluster (/taxonomy:cluster): Groups items into logical folders/nodes.
5. Attributes (/taxonomy:attributes): Defines differentiating attributes for each cluster.
6. Build (/taxonomy:build): Runs the entire pipeline in one go.

1. Start a new project:

   gemini taxonomy start shoes visual

   This creates taxonomies/shoes/spec.md. Review the filtering rules there.

2. Run the full pipeline:

   gemini taxonomy build shoes

3. Check output: Your final taxonomy will be generated in taxonomies/shoes/ as configured in output_config.yaml.

• commands/: CLI command definitions.
• skills/: Specialist agent instructions (Architect, Brainstormer, Critic, etc.).
• taxonomies/: (Ignored by git) Your working projects.

MIT - see LICENSE for details.