Skip to content
/

Taxonomy

Twenty-one archetypes, grouped by what they do.

The taxonomy is the spine of this atlas. Phase A framed the field as spec-driven development (5 archetypes); Phase B uncovered an entire operational layer the seeds had no vocabulary for; Phase D added six more — cross-vendor routing, compiled enforcement, self-evolving plugins, token-efficient encoding, OS-level distros, and LLM-evaluated hooks. Below they're grouped by use-case, not by project phase.

§ methodology

Methodology & instruction

Frameworks that ship installable discipline — SKILL.md, slash commands, prose. Behavior comes from the prompt + a few hooks.

§ orchestration

Multi-agent & orchestration

Spawning N agents, routing across vendors, coordinating roles. From hive-mind queens to cross-vendor Claude+GPT+Gemini chains.

§ memory

Memory & context

Where state lives across sessions. File, SQLite, Neo4j, vector DBs, HTTP-proxy intercept, and JIT semantic loaders.

§ governance

Governance & safety

Hooks that block, audit ledgers, compiled invariants, LLM-evaluated policy gates. Where compliance enforcement lives.

§ substrate

Execution substrate

MCP servers, IDEs, daemons, sandboxes, OS distros — the runtime layer agents call into. Sits below the agent loop, not inside it.

§ specialized

Specialized patterns

Single-purpose archetypes that don't fit the bigger families — UI passthroughs, harness generators, self-evolving plugins.

§ unclassified

Unclassified

Pure stubs from broken or marketing-only repos. Kept in the corpus for completeness.

§ a99 6 frameworks
Unclassified

Pure stubs from broken or marketing-only repos. Kept in the corpus for completeness.