# Generate Ideas

> Generates a wide range of ideas and explores multiple approaches before converging on the most promising direction using a diverge-then-converge process.

- Skill: `auto-skiller/generate-ideas` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add auto-skiller/generate-ideas`
- Raw SKILL.md: https://api.skillmd.com/api/skills/auto-skiller/generate-ideas/raw
- Safety review: PASS (external: skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Tags: Brainstorming, Diverge Converge, Ideation, Problem Framing
- Author: Auto-Skiller (https://skillmd.com/u/auto-skiller)
- Updated: 2026-08-22
- Page: https://skillmd.com/skills/auto-skiller/generate-ideas

---


# Generate Ideas

This skill allows the agent to generate a wide range of ideas and explore multiple approaches before converging on the most promising direction. It uses a deliberate diverge-then-converge process to avoid premature narrowing.

## Steps

1. **Frame the problem.** Restate the problem statement in your own words to confirm you understand it. Identify what a successful outcome looks like.
2. **Inventory constraints.** List hard constraints (non-negotiable) and soft constraints (preferences). Hard constraints narrow the solution space; soft ones inform prioritization.
3. **Diverge — generate freely.** Produce at least 5 distinct ideas without filtering. Apply multiple lenses: conventional, unconventional, hybrid, minimal, and maximal approaches.
4. **Categorize.** Group generated ideas by theme, approach type, or effort level.
5. **Converge — score and filter.** Apply a quick feasibility vs. impact filter. Eliminate ideas that violate hard constraints.
6. **Surface the top 1–3 candidates.** For each, provide a one-line rationale for why it's worth exploring.
7. **Output** the full divergent list + categorized view + recommended direction.

## Output Format

```yaml
ideation_result:
  problem_restatement: "<your restatement>"
  hard_constraints:
    - "<constraint>"
  ideas:
    - id: 1
      idea: "<idea>"
      category: "<theme>"
      feasibility: high | medium | low
      impact: high | medium | low
  recommended_direction: "<top 1-3 ideas with rationale>"
```

## Notes
- Never skip the diverge phase — the best solution is often the 4th or 5th idea generated.
- If the problem space is unclear, invoke `analyze-context` first to sharpen the framing.
- Use the `benchmarking` toolbox after this skill to formally compare shortlisted ideas.