Arguments

[optional: focus area, goal, constraint, or time horizon]

Idea Wizard

Convert open-ended brainstorming into ranked, implementation-ready improvements.

Input

Focus: $ARGUMENTS

If focus is missing, infer from the current repository context and recent user intent. If focus is ambiguous, ask the minimum clarifying questions needed to proceed.

Workflow

1. Define Target and Constraints

Establish:

• Desired outcome (speed, reliability, revenue, UX, risk reduction, etc.)
• Time horizon (quick-win, this-quarter, long-bet)
• Constraints (team size, tech limits, compliance, no-migration limits)

If constraints are unknown, proceed with explicit assumptions.

2. Generate 30 Candidate Ideas

Produce exactly 30 one-line ideas before any filtering.

Use this category quota:

• Reliability and bug prevention: 5
• UX and accessibility: 5
• Performance and cost: 5
• Security and privacy: 4
• Developer experience and testing: 4
• Product and growth impact: 4
• Architecture and technical debt: 3

Each idea must include:

• A concrete action
• A target surface (feature, component, workflow, infra layer)
• A measurable expected result

3. Run Kill Gates

Reject ideas that fail any of these gates:

• Duplicate or near-duplicate of another idea
• No clear measurable outcome
• Effort is high with weak impact (effort >= 4 and impact <= 2)
• Depends on unavailable prerequisites
• Violates explicit user constraints

Capture rejections with concise reasons.

4. Score Survivors with Weighted Model

Use the scoring model in references/scoring-rubric.md.

For each surviving idea:

• Assign 1-5 scores for impact, reach, confidence, time-to-value, strategic fit, effort, and risk
• Compute weighted score
• Add confidence percentage and reasoning quality

5. Select and Balance Finalists

Pick the top 5 by weighted score, sorted highest first.

Then apply balance checks:

• Include at least 2 quick-win ideas when available
• Include at least 1 structural or foundational idea when it is competitive
• Remove redundant finalists that attack the same bottleneck in nearly identical ways

6. Deepen Finalists into Actionable Plans

For each finalist, provide:

• What: specific change
• Why: expected outcome and metric movement
• How: concrete implementation path with touched files/components
• Code sketch: minimal snippet when it clarifies approach
• Test strategy: unit or e2e coverage needed
• Downsides: complexity, risk, maintenance cost
• Confidence: 0-100% plus rationale

7. Present Results

Display the full report inline using the format from assets/idea-wizard-report-template.md.

Include:

• Candidate summary (30 generated, how many rejected, how many survived)
• Rejection ledger (idea + reason)
• Ranked finalists with full detail
• Recommendation for what to do next

8. Offer to Save

After presenting the report, ask whether to save it to docs/discovery/YYYY-MM-DD-<slug>-idea.md (derive <slug> from the focus area as short kebab-case). Create the directory if needed. Add frontmatter (summary + read_when) per the docs-frontmatter skill convention.

Guardrails

• Do not reward novelty over value.
• Prefer boring, high-leverage, low-regret improvements.
• Avoid speculative architecture unless evidence justifies it.
• Tie every recommendation to expected measurable outcomes.
• Never auto-implement. Ask which finalist to execute next.

Next-Step Options

After presenting finalists, offer:

1. Save report to docs/discovery/ (if not already saved)
2. Turn a finalist into a PRD with /dev-brainstorm
3. Draft a phased plan for top 3 ideas
4. Re-run wizard with a narrower focus