solution-design（解决方案设计）

Solution Design — Orchestrator

*Strategy — Step 2 of 4. Brainstorms targeted solutions and ranks them with evidence-backed scoring.*

Core Question: "What's the highest-impact thing we can do about this?"

Critical Gates — Read First

1. Every initiative hypothesis MUST name the root cause. Generic growth ideas produce low-impact, unfocused initiatives. If the hypothesis "because" clause doesn't reference the confirmed root cause, the initiative is untethered.
2. Force-rank BEFORE scoring — ranking prevents "everything is a 6." The forced ranking sets the ceiling for ICE scores. If you ranked it #1, its ICE should be highest.
3. Anti-generic test is mandatory — "would this help ANY company?" Delete the root cause reference from an initiative. If it still makes sense for any company, it's generic. Rewrite.
4. ≤3 initiatives above cut line — force the constraint. More than 3 active initiatives means none get full attention. Parked initiatives have their turn after the current batch ships.

Inputs Required

• Root cause from .agents/problem-analysis.md

Output

• .agents/solution-design.md

Chain Position

Previous: problem-analysis | Next: funnel-planner | Cross-stack: system-architecture (from product-skills — for initiatives requiring technical builds)

Re-run triggers: When root cause changes (re-diagnosis), when all "Proceed" initiatives are shipped or killed, or when new constraints emerge.

Agent Manifest

Routing Logic

Route A: Full Analysis (Default)

For any non-trivial solution design. This is the default route.

Layer 1:        research-agent
                     ↓
Layer 1.5:      initiative-generator-agent + unconventional-agent (parallel)
                     ↓ merge
Layer 2:        ranking-agent → ice-scoring-agent → cut-line-agent → critic-agent

Route B: Quick Design

When the root cause is well-established, constraints are known, and speed matters.

research-agent → initiative-generator-agent → ranking-agent → ice-scoring-agent → cut-line-agent → critic-agent

Skip unconventional-agent. Use when:

• The user has already identified approaches and wants prioritization
• Time is critical and the user confirms skipping unconventional scan
• The problem space is well-understood with clear standard solutions

Dispatch Protocol

Pre-Dispatch: Context Gathering

Step 0: Product Context

Check for research/product-context.md. If missing: Strongly recommended: run icp-research (from hungv47/research-skills) first to create research/product-context.md — this skill works without it but produces significantly better analysis with it. If the user prefers not to, ask the user 8 product questions (what, who, problem, differentiator, proof points, pricing, objections, voice) and save to research/product-context.md.

If .agents/problem-analysis.md has a date field older than 30 days, recommend re-running problem-analysis before proceeding — the root cause landscape may have shifted. Tip: /navigate status (from meta-skills) gives a single-pass freshness report across all upstream artifacts.

Required Artifacts

Optional Artifacts

Root Cause Review

Read .agents/problem-analysis.md. Quote the root cause statement(s) and gap percentages. This anchor is necessary because generic growth ideas produce low-impact, unfocused initiatives.

Constraint Interview

Interview for:

• Root cause context: Confirm the root cause and gap size
• Constraints: Budget ceiling? Team size? Hard deadlines?
• Prior attempts: What's been attempted? What worked/didn't?

Route Selection

• Default → Route A (Full Analysis)
• If user requests speed OR problem has clear standard solutions → Route B

Single-Agent Fallback

If the full orchestration is unnecessary (simple problem, few initiatives, user already has a ranked list), you may run the analysis inline without dispatching agents. Apply the same quality gate (8-point checklist) before delivering.

Layer 1 — Research

Dispatch research-agent.

research-agent

• Input: Problem statement, root cause from problem-analysis.md, gap percentages, constraints, prior attempts
• References: references/hypothesis-framework.md, references/initiative-types.md
• Expected output: Validated root cause, 3-6 case studies with quantified results, unconventional tactic scan, constraint summary

Layer 1.5 — Parallel: Initiative Generation

After research-agent returns, dispatch initiative-generator-agent and unconventional-agent simultaneously.

initiative-generator-agent

• Input: Research-agent output (case studies, validated root cause, constraints)
• References: references/hypothesis-framework.md, references/initiative-types.md, references/initiative-planning.md, references/churn-playbook.md (if root cause is churn)
• Expected output: 5-10 initiatives with hypotheses, mechanics, effort sizing, anti-generic checks

unconventional-agent

• Input: Research-agent output
• References: references/initiative-types.md
• Expected output: 2-4 unconventional initiatives with risk assessments

Layer 1.5 Merge

After both agents return:

1. Combine standard and unconventional initiatives into a single numbered list
2. Verify effort mix across the combined list: ≥2 Small, ≥2 Medium, ≥1 Large
3. If effort mix fails, send feedback to initiative-generator-agent to adjust
4. Present the full list to the user: *"Which resonate? Any to cut immediately? Any that spark a different idea?"*
5. Incorporate user feedback before passing to Layer 2

Layer 2 — Sequential: Rank → Score → Cut → Critic

Step 1: ranking-agent

• Input: Merged initiative list + user feedback
• References: references/ice-scoring-rubric.md
• Expected output: Strict 1-through-N forced ranking with reasoning per rank
• Interview prompt: *"If you could only do ONE of these, which would it be? And after that one?"*

Step 2: ice-scoring-agent

• Input: Ranking-agent output + initiative list
• References: references/ice-scoring-rubric.md
• Expected output: ICE scores (1-10 each) with evidence citations, differentiation check

Step 3: cut-line-agent

• Input: ICE-scoring-agent output
• References: references/initiative-planning.md
• Expected output: Proceed/Park/Kill decisions, capacity assessment, Proceed validation
• Interview prompt: *"How many can you execute well simultaneously?"*

Step 4: critic-agent

• Input: Complete merged analysis
• Expected output: PASS or FAIL against 8-point quality gate

Critic Gate

Maximum 2 rewrite cycles. If the critic returns FAIL:

1. Read the critic's failure report — it names the specific gate, the fix, and the agent to re-dispatch
2. Re-dispatch ONLY the named agent(s) with the critic's feedback
3. Re-merge and send back to critic-agent
4. If FAIL again after 2 cycles: deliver the artifact with a "Known Issues" section listing unresolved gate failures

Out-of-Scope Persistence

After delivering the artifact, write killed initiatives to .agents/meta/out-of-scope/ so future sessions don't re-analyze them:

For each initiative marked Kill in the Decisions table:

# [Initiative Name]
**Decided:** [date]
**Context:** [root cause and goal that prompted this analysis]
**Decision:** Killed because [reason from Kill criteria]
**Revisit if:** [condition that would change the decision — e.g., "team grows to 5+", "root cause shifts to retention"]

Save as .agents/meta/out-of-scope/[kebab-case-name].md. Create the directory if it doesn't exist.

Why: Prevents re-debating settled decisions in future sessions. Navigate and discover read this directory before recommending workflows or asking about features already rejected.

Artifact Template

On re-run: rename existing artifact to solution-design.v[N].md and create new with incremented version.

---
skill: solution-design
version: 1
date: {{today}}
status: draft
---

# Solution Design

**Root Cause:** [from problem-analysis.md]

## Phase 1: Initiatives

### 1. [Name] — Effort: S
**Hypothesis:** If [action], then [outcome], because [root cause connection].
**Mechanic:** [Specific steps — what you'd actually do, in order]
**Target Metric:** [metric]
**Anti-generic check:** [Why this only works for our specific root cause]

### 2. [Name] — Effort: S
[Same format]

### 3. [Name] — Effort: M
[Same format]

[...5-10 total]

## Phase 2: Prioritization

### Forced Ranking (gut + evidence)

1. [Name] — most promising because [reason]
2. [Name]
3. [Name]
...

### ICE Scoring

| Rank | Initiative | I | C | E | ICE | Key Evidence |
|------|-----------|---|---|---|-----|-------------|
| 1 | [Name] | [score] | [score] | [score] | [sum] | [One-sentence evidence for top-scoring dimension] |
| 2 | ... | ... | ... | ... | ... | ... |

### Decisions

| Initiative | Decision | Owner | Target Metric (Baseline) | Kill Criteria |
|------------|----------|-------|-------------------------|---------------|
| [Name] | Proceed | [Person] | [Metric]: [current] | Stop if [metric] < [threshold] after [duration] |
| [Name] | Park | — | — | — |
| [Name] | Kill | — | — | Reason: [why] |

**Cut line:** [N] initiatives proceeding. Team capacity: [context].

## Next Step

Run `funnel-planner` to set targets, then `experiment` to test.
If any "Proceed" initiative requires a technical build, also run `system-architecture` (from `hungv47/product-skills`) with these initiatives as context.

Worked Example

Root Cause: Signups declined because (1) ad targeting brought low-intent visitors (~55%) and (2) homepage redesign lost trust signals (~35%).

Layer 1: research-agent

Validates root cause, searches for case studies: "ad targeting quality case study", "homepage trust signals conversion case study". Finds 4 case studies with quantified results. Confirms constraints: 3-person growth team, no budget ceiling, prior attempts = none.

Layer 1.5: Parallel Dispatch

initiative-generator-agent produces 6 standard initiatives:

1. Restore + Refine Paid Targeting (S)
2. Restore Homepage Social Proof (S)
3. Landing Page per Ad Group (M)
4. Homepage A/B: Old vs New (S)
5. Intent-Qualifying Ad Creative (M)
6. Dedicated Onboarding for Paid Visitors (L)

unconventional-agent produces 1 unconventional initiative:

• U1. Pixel Sharing with Complementary SaaS (M)

Merge: 7 initiatives total. Effort mix: 3S, 2M, 1L + 1M unconventional = passes (3S ≥2, 3M ≥2, 1L ≥1).

Layer 2: Sequential

ranking-agent force-ranks 1-7. #1 = Restore Paid Targeting (biggest root cause, fastest fix).

ice-scoring-agent scores all 7. #1 gets ICE 25, #7 gets ICE 13. No more than 2 initiatives share a total.

cut-line-agent draws line: 2 Proceed (Restore Paid Targeting, Restore Social Proof), 2 Park, 3 Kill. Validates owners, baselines, kill criteria.

critic-agent: PASS — all 8 gates satisfied.

Final Artifact

# Solution Design

**Root Cause:** (1) Ad targeting quality (~55%), (2) Homepage trust/clarity (~35%)

## Phase 1: Initiatives

### 1. Restore + Refine Paid Targeting — Effort: S
**Hypothesis:** If we revert to original targeting and add a converters lookalike, then paid conversion returns to ≥3%, because the current broad targeting is bringing visitors outside our ICP.
**Mechanic:** Pull last 90d converters list → create 1% lookalike in Meta → A/B test current broad vs. lookalike targeting → 7-day test at $50/day per variant.
**Target Metric:** Paid visitor-to-signup rate
**Anti-generic check:** Only relevant because our specific targeting change caused the quality drop.

### 2. Restore Homepage Social Proof — Effort: S
**Hypothesis:** If we add back customer logos + testimonial above the fold, then homepage bounce rate drops to ≤40%, because the redesign removed the trust signals that were on the old page.
**Mechanic:** Screenshot old homepage via Wayback Machine → identify removed trust elements → add 3 customer logos + 1 short testimonial quote to new hero section → deploy.
**Target Metric:** Homepage bounce rate
**Anti-generic check:** Only relevant because our specific redesign removed these elements.

[...5 more initiatives]

## Phase 2: Prioritization

### Forced Ranking
1. Restore + Refine Paid Targeting — fastest fix for biggest root cause
2. Restore Homepage Social Proof — fast, addresses second root cause
3. Homepage A/B: Old vs New — definitive answer on redesign impact
4. Landing Page per Ad Group — higher effort but compounds with #1
5. Intent-Qualifying Ad Creative — interesting but less certain
6. Pixel Sharing with Complementary SaaS — unconventional, needs partner
7. Dedicated Onboarding for Paid — too complex for now

### ICE Scoring

| Rank | Initiative | I | C | E | ICE | Key Evidence |
|------|-----------|---|---|---|-----|-------------|
| 1 | Restore Paid Targeting | 8 | 8 | 9 | 25 | I:8 — paid is 60% of traffic, old targeting converted at 3.5% vs 1.2% now |
| 2 | Restore Social Proof | 7 | 7 | 9 | 23 | C:7 — Wayback shows old page had logos+testimonial that new page removed |
| 3 | Homepage A/B | 6 | 9 | 7 | 22 | C:9 — direct comparison eliminates guessing about redesign impact |
| 4 | LP per Ad Group | 7 | 6 | 5 | 18 | I:7 — message match lifts are well-documented (CXL: 20-50% improvement) |
| 5 | Intent Ad Creative | 5 | 4 | 6 | 15 | C:4 — no precedent with our audience, hypothesis only |
| 6 | Pixel Sharing | 4 | 3 | 5 | 12 | C:3 — requires finding willing partner, untested |
| 7 | Paid Onboarding | 7 | 3 | 3 | 13 | E:3 — needs eng + dynamic content, 4+ weeks |

### Decisions

| Initiative | Decision | Owner | Target Metric (Baseline) | Kill Criteria |
|------------|----------|-------|-------------------------|---------------|
| Restore Paid Targeting | **Proceed** | Sarah | Paid conversion: 1.2% | Stop if <2% after 10 days |
| Restore Social Proof | **Proceed** | James | Homepage bounce: 52% | Stop if >48% after 7 days |
| Homepage A/B | **Park** | — | — | Do after #1 and #2 ship |
| LP per Ad Group | **Park** | — | — | Revisit if paid targeting fix isn't enough |
| Intent Ad Creative | **Kill** | — | — | Too uncertain; paid targeting fix addresses the same root cause more directly |
| Pixel Sharing | **Kill** | — | — | Requires partner development; revisit if standard approaches underperform |
| Paid Onboarding | **Kill** | — | — | Too complex for current team capacity |

**Cut line:** 2 initiatives proceeding. Team: 3-person growth team.

## Next Step

Run `funnel-planner` to set targets for Restore Paid Targeting and Restore Social Proof, then `experiment` to test.

Anti-Patterns

Generic initiatives — "Improve SEO" or "Optimize onboarding" are categories, not initiatives. INSTEAD: Every initiative must describe a specific mechanic tied to a specific root cause. Run the anti-generic test: delete the root cause — does the idea still work for ANY company?

Untethered ideas — Ideas that sound good but don't connect to the confirmed root cause. INSTEAD: Every hypothesis "because" clause must reference the root cause. Root cause anchoring prevents the brainstorm from drifting into a generic growth playbook.

All-small or all-large efforts — Skewing entirely toward quick wins misses structural fixes; all-large stalls momentum. INSTEAD: Enforce effort mix: ≥2 Small, ≥2 Medium, ≥1 Large.

"Everything is a 6" scoring — When all ICE scores cluster together, the ranking is meaningless. INSTEAD: Force-rank before scoring to create differentiation. If you ranked it #1, its ICE should be highest.

Naked scores without evidence — "Impact: 7" with no justification invites political scoring. INSTEAD: Every score needs a one-sentence citation: what data or reasoning supports this number.

Too many initiatives above cut line — More than 3 active initiatives means none get full attention. INSTEAD: Force the ≤3 constraint. Parked initiatives have their turn after the current batch ships.

Missing kill criteria — Initiatives without a "stop if" threshold run forever. INSTEAD: Every "Proceed" initiative needs a specific metric, threshold, and duration that triggers a stop.

Skipping the unconventional scan — Only considering standard approaches when the root cause may need asymmetric leverage. INSTEAD: Run unconventional-agent alongside initiative-generator-agent unless Route B is explicitly chosen.

References

• references/hypothesis-framework.md — If/Then/Because templates
• references/initiative-types.md — Hero vs Support classification
• references/ice-scoring-rubric.md — Detailed scoring calibration with scored examples
• references/initiative-planning.md — Detailed execution planning for complex initiatives
• references/churn-playbook.md — Cancel flow, dunning, health scores, exit-reason mapping (when root cause is churn)
• references/churn-cancel-flow-templates.md — Email templates for cancel flow and dunning sequences
• references/churn-health-score-guide.md — Health score implementation and calibration