Forge

STEP 2: Problem Restate + AskUserQuestion #1

Each member restates through their engineering lens. For --quick, skip AskUser.

Before presenting options, the Coordinator runs a silent pre-deliberation check:

• Is this a "help me think" or "validate my decision" question? (If the user already has a preferred answer, name it.)
• Is there a hidden constraint not mentioned? (team size, deploy deadline, legacy lock-in, budget)
• Is this a reversible or irreversible architectural decision? (Reversible: pick and iterate. Irreversible: deliberate hard.)
• Is the question actually two questions combined? (e.g., "should we refactor AND switch frameworks?")

AskUser #1 — Forge's three probing questions:

Surface these via AskUserQuestion with a note: "Before we start, three quick questions to make the panel more useful:"

1. "这个决定的时间压力是什么？"

   • "需要这周内给出答案" → Quick mode automatically, focus on decision
   • "有几周时间，想深入研究" → Full deliberation with evidence
   • "没有时间压力，想彻底想清楚" → Full mode, may include prototyping suggestions
   • "已经做了决定，想找人挑战" → Panel shifts to adversarial red-team mode (Popper leads)

2. "你自己倾向哪个方向？"

   • "我倾向于 X，但不确定" → Panel should challenge X specifically, not re-derive from scratch
   • "完全没有方向，都不清楚" → Panel derives recommendations independently
   • "两个方向都有人支持，内部有争议" → Panel explicitly maps both camps and arbitrates
   • "不想透露，想看客观分析" → Blind mode — members don't know user's lean

3. "这段代码/系统的维护者是谁？"

   • "我自己，一个人的项目" → Maintainability weight increases for solo dev
   • "小团队（<5人）" → Team communication cost factors in
   • "大团队，有专职 infra" → Enterprise patterns become relevant
   • "接手别人的代码，不熟悉" → Evidence gathering broadens; risk assessment heightens

If user's original message already answers some of these, skip those sub-questions.

Wittgenstein inline note: If the user's description contains vague technical terms (e.g., "性能不好" / "不够优雅" / "感觉不对"), Wittgenstein (if on panel) performs an F/D/Q decomposition inline during Step 2 — no extra interaction round required. The decomposition result is folded into the confirmed problem statement before proceeding.

AskUser #1 also surfaces Evidence Brief findings:

• "代码库分析完成，发现 X。这与你描述的问题一致吗？" → Yes / No, 实际情况是这样
• If discrepancy found: gather more evidence before proceeding

STEP 3: Round 1 — Informed Independent Analysis

All members analyze in parallel. Each must reference specific evidence from the Brief AND the user's stated constraints from Step 2.

STEP 4: Adaptive Depth Gate + AskUserQuestion #2

For Forge:

• HIGH consensus → likely a clear architectural winner emerged
• MEDIUM/LOW → engineering disputes with genuine tradeoffs worth Round 2

AskUser #2 — Don't just ask "go deeper?" Ask what's actually useful:

Present Round 1 summary (1 sentence per member), then ask:

"Round 1 完成了。在继续之前——"

1. "哪个视角让你最意外，或者最不舒服？"

   • 用户点名某个 agent → Round 2 中那个 agent 的论点成为 Antithesis 重点
   • "都没有" → 快速出结论，HIGH consensus 已足够
   • "Popper/Nietzsche 的挑战让我担心" → 专门 deep dive 风险维度

2. "有没有哪个成员完全误解了你的情况？"

   • 如果有 → 修正上下文，重新 Round 1（仅限误解的成员）
   • 没有 → 继续

3. 深度选择：

   • "够了，出结论" → Skip to Synthesis + Verdict
   • "有真正的分歧值得深挖" → Round 2
   • "补充一个你们没考虑到的约束" → User adds context, then Round 2

STEP 5: Round 2 — Hegelian Cross-Examination

Enforce synthesis requirement. In engineering context:

• Thesis = dominant technical recommendation
• Antithesis = strongest dissenting technical position
• Synthesis = the design that integrates both (not a compromise — a better design)

STEP 6: Coordinator Synthesis

Identify the Hegelian arc in engineering terms.

STEP 7: Forge Verdict (below)

Output Templates

Forge Verdict (Full Mode)

## Forge Verdict

### Problem
{Original engineering question}

### Panel
{Members convened, triad/mode, selection rationale}

### Evidence Summary
{3-5 bullet points from the Evidence Brief — what we actually know about the codebase}

### Architecture Decision
**Recommendation**: {Clear architectural recommendation}
**Rationale**: {Why — grounded in evidence, not just theory}
**Trade-offs accepted**: {What you're giving up with this choice}
**Trade-offs rejected**: {What alternative approaches were considered and why rejected}

### Implementation Path
**Phase 1** (immediate): {First concrete steps}
**Phase 2** (short-term): {Next steps within a sprint/week}
**Phase 3** (long-term): {Structural changes that require more time}

### Risk Assessment
| Risk | Likelihood | Impact | Mitigation |
|------|-----------|--------|------------|
| {risk} | H/M/L | H/M/L | {specific action} |

### Technical Debt Ledger
- **Debt created by this decision**: {new complexity introduced}
- **Debt paid by this decision**: {existing complexity resolved}
- **Debt deferred**: {known issues intentionally left for later}

### Dissenting Position
{The strongest argument against the recommendation and what would make it right}

### Confidence
{High / Medium / Low — with specific reasoning}

### 相关审议室
{E.g., "Also consider: /bazaar if this is a build-vs-buy decision, or /oracle if the architecture decision is tied to a career/team direction question"}

### 后续追踪
实施后回顾：这个架构决策有效吗？遇到了什么技术债？

Quick Forge Verdict

## Quick Forge Verdict

### Problem
{Engineering question}

### Panel
{Members and rationale}

### Recommendation
{Single concrete technical recommendation}

### Member Positions
- **Feynman**: {Core position}
- **Ada**: {Core position}
- ...

### Key Technical Risk
{The most important thing that could go wrong}

### Next Step
{Single most important first action}

Duo Forge Verdict

## Duo Forge Verdict

### Problem
{Engineering question}

### The Technical Dialectic
**{Member A}** ({their lens}) vs **{Member B}** ({their lens})

### What This Means for Your Decision
{How to use these opposing technical perspectives}

### {Member A}'s Position
{Core technical argument in 2-3 sentences}

### {Member B}'s Position
{Core technical argument in 2-3 sentences}

### Where They Agree
{Unexpected convergence on technical facts}

### The Core Technical Tension
{The irreducible engineering tradeoff}

### Recommended Reading of the Debate
{How a senior engineer should interpret this dialectic}

Example Usage

Architecture decision: /forge --triad architecture "Should we split our 50k-line monolith into microservices?" → Ada + Occam + Feynman convene, examine codebase structure, run 2-round deliberation, produce Forge Verdict with Implementation Path.

Quick debugging sanity check: /forge --quick "Is our N+1 query issue in the user/posts relationship worth fixing now?" → Auto-selects debugging triad, rapid 2-round analysis, Quick Forge Verdict.

Duo refactoring dialectic: /forge --duo "Should we incrementally refactor or do a full rewrite?" → Selects Torvalds vs Nietzsche (pragmatic fix vs creative destruction), 3-round dialectic.

Full panel review: /forge --full "Evaluate our entire API design before v2.0 launch" → All 6 members, full evidence gathering, complete 8-step deliberation.