Skip to content
/
Phase B Batch 1

Batch 01 Notes

10 frameworks in this batch

Batch 01 Notes

Roster

# slug display_name stars distribution_type tier key_differentiator
1 openspec-skills openspec-skills (chyiiiiiiiiiiii) 6 skill-pack A 3-skill SKILL.md pack (proposal, apply, archive); zero runtime deps; most minimal openspec implementation in corpus
2 openspec-mcp openspec-mcp (Lumiaqian) 24 npm-mcp-server A ~31 MCP tools + web dashboard (React+Fastify+WebSocket) on port 3000; approval state machine
3 openspec-zed OpenSpec-Zed 14 zed-extension A Only Rust/WASM32-WASI implementation; only Zed-native; only GPL-3.0 in batch; prompt bridge only (no execution due to WASM sandbox)
4 openspec-for-copilot openspec-for-copilot 18 vscode-extension A 23 VS Code commands; React+Vite webview; targets GitHub Copilot Chat; builds-prompts.js compiles Markdown→TypeScript at build time; forked from kiro-for-codex-ide
5 openspec-ui openspec-ui (ToruAI) 18 standalone-app A Pure passive observer — Rust/Axum backend + React/Tailwind frontend; no AI interaction; watches N openspec/ dirs via SSE; genuinely novel archetype
6 superspec SuperSpec (danielhanold) 8 openspec-schema A Pure meta-framework — OpenSpec custom schema.yaml wiring OpenSpec + Superpowers seeds; zero new code; only framework that composes two seeds
7 ralphy-openspec ralphy-openspec 186 npm-cli-plus-skills A Highest stars in batch; SQLite state.db; Ralph Loop continuous execution; <promise>TASK_COMPLETE</promise> signal; budget tracking; 9 CLI subcommands; 8 slash command files (4×Claude Code + 4×Cursor)
8 openspec-reviewed-workflow openspec-reviewed-workflow 0 template-bundle A Mandatory evidence-driven review gate (OPTIMAL/IMPROVABLE/RETHINK verdict) between proposal and specs; schema-level gate with no hooks or runtime
9 goopspec GoopSpec 29 opencode-plugin A 16-agent multi-model Conductor+Specialists; 2 mandatory human contract gates (confirm/accept); 22 MCP tools; 7 hooks; wave-based execution; local-embedding semantic memory; 1,660 passing tests
10 spec-compare spec-compare (cameronsjo) 39 methodology-doc C Not a framework — research comparison of 12 SDD tools; SDD Maturity taxonomy (Spec-First/Spec-Anchored/Spec-as-Source); highest stars of any non-framework in this space

Intra-Batch Patterns

Five of the ten frameworks (openspec-skills, openspec-mcp, openspec-zed, openspec-for-copilot, openspec-reviewed-workflow) are direct openspec derivatives — they either extend the openspec lifecycle or deliver the same skills/schema for a different host IDE, revealing that openspec has the strongest "fork gravity" in the SDD ecosystem. There is a clear split between two architectural philosophies: schema/skill-level frameworks that enforce process through artifact dependency graphs (superspec, openspec-reviewed-workflow, openspec-skills) and execution-layer frameworks that add runtime machinery (openspec-mcp with MCP tools, ralphy-openspec with SQLite+Ralph Loop, goopspec with 16 agents+hooks). The batch contains one genuinely novel archetype not covered by the five seeds — openspec-ui as a passive observer dashboard — confirming that the SDD space is generating non-framework tooling that orbits it. Star counts are a weak signal in this batch: ralphy-openspec (186 stars, sophisticated execution engine) and spec-compare (39 stars, no code) both outrank goopspec (29 stars, 1,660 tests, most complete implementation); the disconnect reflects discovery/SEO effects rather than quality. Two frameworks (openspec-reviewed-workflow with 0 stars, ralphy-openspec with BSD-3-Clause) represent personal experiments with no external validation despite interesting ideas.

Most Interesting Finds

ralphy-openspec — The Ralph Loop continuous execution pattern (<promise>TASK_COMPLETE</promise> as machine-parseable stop signal + SQLite state.db + budget tracking) is a distinct paradigm not represented in any seed framework. Its 186 stars suggest strong practitioner validation despite being a 1-contributor project. The combination of 8 slash command files (4 for Claude Code, 4 for Cursor) with a dedicated CLI binary is the most complete cross-tool distribution strategy in the batch.

goopspec — The executor tier system (4 executor agents with different default LLMs assigned by task complexity: low→claude-sonnet-4-6, medium→kimi-k2p5, high→gpt-5.3-codex, frontend→claude-opus-4-6) is the most sophisticated cost-optimization strategy in the corpus. The two typed confirmation gates ("confirm" and "accept") are architecturally unique — no other framework prevents AI self-completion at the protocol level.

Tier C Items

slug reason
spec-compare Pure methodology documentation (13 Markdown files); no commands, skills, agents, or executable components; comparison resource not a deployable framework

Cross-References

  • openspec-skills is the minimal reference implementation of the openspec seed (Archetype 4); use it as the baseline when comparing openspec derivatives
  • superspec depends on both openspec and superpowers seeds coexisting in the target project; requires both to be installed separately
  • ralphy-openspec implements the Ralph Loop pattern independently documented in spec-compare's docs/tools/ralph-loop.md
  • goopspec is the OpenCode-native equivalent of ralphy-openspec's execution-layer approach; both add persistent state + CLI but via different architectures (Conductor+Specialists+SQLite vs. single-agent+SQLite+Ralph Loop)
  • openspec-reviewed-workflow is the only schema-level gate in the corpus — comparable in intent to superpowers' Iron Law guardrails but applied at the artifact-dependency level instead of the prompt level
  • spec-compare SDD Maturity taxonomy (Spec-First / Spec-Anchored / Spec-as-Source) is directly applicable to all frameworks in this batch: openspec derivatives are Spec-Anchored; goopspec/ralphy are Spec-First (specs used during execution then archived); superspec spans both