Skip to content
/
Phase D Batch 5

Phase D Batch 05 — BMAD-family + research/spec hybrids

Phase D Batch 05 — BMAD-family + research/spec hybrids

Roster (10)

slug stars distribution cli_binary local_ui orchestration multi_model tier
bmad-at-claude 235 standalone-repo bmad-method (npx) none hierarchical no A
bmad-architecture-agent 15 standalone-repo none none sequential no A
claude-code-bmad-skills 435 skill-pack none (shell installer) none sequential no A
vibe-ceo-ing unknown methodology-doc none none none no C
research-kit-nguyenvanduocit 19 cli-tool (python) research (Python) none sequential no A
tars-gemini unknown unknown unknown unknown unknown unknown C
lit-ai-methodology unknown methodology-doc none none none no C
agentic-sdlc-12-factors 29 methodology-doc none Jekyll site sequential no A
agentic-agile-template-ms 60 template-bundle none none parallel-fan-out no A
vibecodex 4 npm-package vibecodex none none no A

Intra-Batch Patterns

BMAD convergence: Three repos (bmad-at-claude, bmad-architecture-agent, claude-code-bmad-skills) all port BMAD-METHOD to different execution contexts — each makes a different architectural tradeoff. bmad-at-claude adds hooks + JS message queue for multi-session management; bmad-architecture-agent adds domain-specialist agent content; claude-code-bmad-skills replaces persona-md with skill-md for token efficiency. The three together constitute a case study in how one methodology gets adapted across different runtime constraints.

Methodology-doc abundance: 4 of 10 entries are pure documentation (Vibe CEO'ing, Tars Gemini, LIT.AI, and agentic-sdlc-12-factors). Two of these (Vibe CEO'ing, Tars Gemini) have no public repository — they exist only as references in discovery documents. The methodology docs that do exist (12-factor SDLC, Microsoft template) are notable for their governance and measurement frameworks.

No multi-model routing anywhere: Zero of the 8 analyzable frameworks use multi-model routing. This contrasts with seed frameworks like taskmaster-ai (which added multi-model) and suggests BMAD-family ports prioritize methodology fidelity over model routing.

Research-kit as a research tool, not a code tool: The nguyenvanduocit/research-kit is the only framework in the full corpus (not just this batch) that adapts SDD methodology for academic/market research rather than software development. Its anti-refusal enforcement pattern in /research.execute is an unusual prompting innovation.

Microsoft's wave-based parallel model: The agentic-agile-template introduces a distinct coordination primitive — "waves" of independent stories with file-ownership enforcement — that no other framework in the corpus (seeds included) uses. It's the only framework that addresses team-level parallel execution with explicit merge-conflict prevention.

Most Interesting Find

Microsoft agentic-agile-template — the only framework with a formal 8-dimension quantitative evaluation framework for measuring human-agent partnership quality over time. The "wave-based parallel execution with file ownership" model is a concrete team coordination pattern not found in any seed or other analyzed framework.

claude-code-bmad-skills — the "helper pattern" (70-85% claimed token reduction over upstream BMAD by removing persona overhead) is the most direct engineering response to BMAD's well-documented token cost problem. The SessionStart hook that auto-detects project level from bmad/config.yaml is an elegant environment-injection pattern.

Items Written as Tier C

slug reason
vibe-ceo-ing Reddit reference only; no dedicated repository — maps to upstream BMAD-METHOD web UI path with a non-technical founder framing
tars-gemini Wave-3 discovery reference only; no GitHub repo found after 8 distinct search queries
lit-ai-methodology Blog-only (lit.ai/blog); domain reachable but port 8443 unreachable; no GitHub repo found

Cross-References Discovered

  • bmad-at-claude → depends on bmad-method@4.33.0 (direct npm dependency on the BMAD-METHOD seed)
  • bmad-architecture-agent → installed via npx bmad-method install (uses BMAD-METHOD's installer)
  • claude-code-bmad-skills → explicit attribution to bmad-code-org/BMAD-METHOD as upstream; adopts superpowers-style skill-md format
  • research-kit-nguyenvanduocit → explicit fork of spec-kit (seed); documented fork relationship
  • vibe-ceo-ing → maps to the upstream bmad-method seed (no separate codebase)
  • agentic-sdlc-12-factors → inspired by the Twelve-Factor App (heroku/12factor.net) and Agile Manifesto
  • agentic-agile-template-ms → credits Andrej Karpathy's Software 3.0 framework and Agile Manifesto as lineage