Idea Spark

Generate project ideas based on a domain or interest. Use when the user asks for project ideas, wants inspiration, says 'what should I build', 'give me ideas...

Generates actionable project ideas by scanning real pain points from Hacker News, Reddit, and GitHub, then optionally validates them with idea-check.

Workflow

1. Extract the domain

From the user's message, extract:

Field	Notes
domain	Area of interest (e.g. "developer tools", "health tracking", "AI agents")
count	Number of ideas requested (default: 5)
type	Project type if specified: CLI tool, API, bot, app, library (default: any)

If the domain is vague, ask for clarification before proceeding.

2. Research pain points

Run 3 searches in parallel using the built-in web_search tool to find real problems people are complaining about or requesting:

{ "tool": "web_search", "query": "site:news.ycombinator.com \"I wish\" OR \"someone should build\" OR \"why isn't there\" <domain> 2025 2026" }

{ "tool": "web_search", "query": "site:reddit.com \"looking for\" OR \"is there a\" OR \"frustrated with\" <domain> tool" }

{ "tool": "web_search", "query": "github.com trending <domain> OR \"help wanted\" OR \"good first issue\" <domain>" }

3. Synthesize ideas

From the search results, extract recurring themes and unmet needs. For each idea, produce:

Field	Format
name	Short project name (2-4 words)
pitch	One-line description of what it does
pain	The real problem it solves (with source: HN/Reddit/GitHub)
type	CLI, API, bot, app, library, MCP server, OpenClaw skill
effort	low / medium / high

Generate count ideas, ranked by how specific and actionable the pain point is.

4. Validate with idea-check (if available)

For each idea, run a quick validation:

{
  "tool": "exec",
  "command": "mcporter call idea-reality.idea_check idea_text=\"<pitch>\" depth=quick"
}

If mcporter or idea-reality is not available, skip this step and note that validation was not performed.

Add the reality_signal score to each idea. Flag ideas with signal > 70 as "crowded".

5. Present the results

Format — strict:

💡 Idea Spark — <domain> — <count> ideas

1. <name>
   <pitch>
   Pain: <pain> (source)
   Type: <type> · Effort: <effort> · Signal: <score>/100 <status>

2. ...

Status mapping:

• Signal < 30: "open field"
• Signal 30-70: "some competition"
• Signal > 70: "crowded"
• Not checked: "not validated"

6. Follow-up

After presenting, suggest:

• "Want me to deep-check any of these?" → run idea-check with depth=deep
• "Want me to start building #N?" → proceed with development

7. Error handling

• If web searches return no results → broaden the domain, try without site filters
• If all ideas score > 70 → tell the user the space is saturated, suggest narrowing the niche
• If mcporter is not available → present ideas without validation, mention idea-check for later
• If the domain is too broad (e.g. "tech") → ask the user to narrow down

Examples

User says	Domain	Action
"Give me 5 project ideas for developer tools"	developer tools	Search HN/Reddit/GitHub, generate 5 ideas, validate each
"What should I build in the AI agent space?"	AI agents	Search for pain points, generate ideas, validate
"I want to build something with MCP servers"	MCP servers	Focus on MCP ecosystem gaps, generate ideas
"Startup ideas around health tracking"	health tracking	Search for unmet needs, generate ideas, flag crowded ones