When a single AI reviews your code, it plays every role at once — security engineer, pragmatist, craftsman, adversary — and negotiates them internally before it speaks. The result is balanced, reasonable, and slightly soft. BMAD 6.9.0 ships five tools that break that pattern. Each one gets its own post below.
The five tools — each with a deep-dive
Code Review · Idea Pressure-Testing
Why Your Code Reviews Are Too Nice (And How to Fix That)
Five adversarial reviewers — Vex (security), Grumbal (adversary), Boundary (edge-cases), Yui (craftsman), Dana (pragmatist) — each thinking independently with
--mode subagent. Plus bmad-forge-idea: two voices per branch, three possible verdicts.Read → /blog/bmad-code-review-crew/
Persistent Memory
Party Memory: The AI Room That Remembers Your Codebase
Memory is per-group and append-only. Named groups default off — opt in with
memory = true. The room accumulates knowledge across sessions. Open-cast walk-ons can be saved at wrap-up. How it actually works (and what it doesn't do).Read → /blog/bmad-party-memory/
Brainstorming · Creative Sessions
108 Ways to Break Out of an Obvious Idea
A structured facilitation workflow across 108 techniques and 13 categories. Three modes: Facilitator, Creative Partner, Ideate for me. Ends with a hand-crafted HTML keepsake — no template, every session looks different. The magic happens around idea 50.
Read → /blog/bmad-brainstorming-facilitator/
Open-Cast Party Rooms · Custom Personas
The AI Room That Casts Itself
No fixed roster — just a scene describing a universe. The cast assembles based on what you're talking about. Ask about navigation and Hera shows up. Switch to weapons systems and Zeb takes the seat. Three saved examples included: Engineering Legends, YC Advisors, Enterprise Bridge.
Read → /blog/bmad-open-cast-rooms/
Forensic Debugging · Code Exploration
Debug Like a Detective: Evidence-Graded Investigation
Three evidence tiers: Confirmed (directly observed, cited path:line), Deduced (reasoning chain from Confirmed), Hypothesized (must state what would confirm or refute it — never deleted). Works for bug investigation and for building a mental model of unfamiliar code.
Read → /blog/bmad-investigate/
Start here if you're new
💬 Claude Code Chat · Best first runfive independent reviewers, blind to each other
/bmad-party-mode --party code-review-crew --mode subagent
How to update
npx bmad-method@latest install --action update --tools claude-code
Comments
Loading comments…
Leave a comment