solo-research

Deep market research — competitor analysis, user pain points, SEO/ASO keywords, naming/domain availability, and TAM/SAM/SOM sizing. Use when user says "research this idea", "find competitors", "check the market", "domain availability", "market size", or "analyze opportunity". Do NOT use for idea scoring (use /validate) or SEO auditing existing pages (use /seo-audit).

# /research

Deep research before PRD generation. Produces a structured `research.md` with competitive analysis, user pain points, SEO/ASO keywords, naming/domain options, and market sizing.

## MCP Tools (use if available)

If MCP tools are available, prefer them over CLI:
- `kb_search(query, n_results)` — search knowledge base for related docs
- `web_search(query, engines, include_raw_content)` — web search with engine routing
- `session_search(query, project)` — find how similar research was done before
- `project_info(name)` — check project details and stacks
- `codegraph_explain(project)` — architecture overview of an existing project (stack, patterns, deps)
- `codegraph_query(query)` — raw Cypher queries against code graph (find shared packages, dependencies)
- `project_code_search(query, project)` — semantic search over project source code

MCP `web_search` supports engine override: `engines="reddit"`, `engines="youtube"`, etc.
If MCP tools are not available, use WebSearch/WebFetch as primary. If MCP web_search tool is available, use it for better results.

### Reddit Search Best Practices

- **Max 3 keywords** in reddit queries — more keywords = fewer results
- Good: `"product hunt outreach launch"` — Bad: `"product hunt scraper maker profiles linkedin outreach launch strategy"`
- `include_raw_content=true` rarely works for Reddit — use fallback chain below

### Reddit Content Access — Fallback Chain

When a search finds a relevant Reddit post, reading its full content requires a fallback chain:

```
1. MCP Playwright (old.reddit.com)     ← BEST: bypasses CAPTCHA, full post + comments
2. PullPush API (api.pullpush.io)      ← search by query/subreddit/author/score/date
3. MCP web_search include_raw_content   ← sometimes works, often truncated
4. WebFetch / WebSearch snippets        ← last resort, partial data only
```

**Method 1: MCP Playwright** (recommended for full post content)
- Use `browser_navigate("https://old.reddit.com/r/...")` — old.reddit.com loads without CAPTCHA
- `www.reddit.com` shows CAPTCHA ("Prove your humanity"), always use `old.reddit.com`
- Snapshot contains full post text + comments in structured YAML
- Example: `old.reddit.com/r/indiehackers/comments/abc123/post_title/`

**Method 2: PullPush API** (for search/discovery)
- Endpoint: `https://api.pullpush.io/reddit/submission/search`
- Params: `q`, `subreddit`, `author`, `score` (e.g. `>10,<100`), `since`/`until` (unix timestamps), `size` (max 100)
- Rate limits: soft 15 req/min, hard 30 req/min, 1000 req/hr. Sleep 4 sec between requests.
- Returns JSON with full `selftext`, author, score, created_utc
- Comment search: `/reddit/comment/search` (same params)
- Can use via curl:
```bash
curl -s "https://api.pullpush.io/reddit/submission/search?q=product+hunt+launch&subreddit=indiehackers&size=10"
```

**Method 3: Reddit .json endpoint** (often blocked)
- Append `.json` to any Reddit URL: `reddit.com/r/sub/comments/id.json`
- Returns raw JSON with full post + comments
- Frequently blocked (403/429) — use as opportunistic fallback only

**Method 4: PRAW** (Reddit Official API, for live search/user profiles)
- [praw-dev/praw](https://github.com/praw-dev/praw) — Python Reddit API Wrapper
- OAuth2 auth, built-in rate limiting, sync/async support
- Best for: live subreddit search, user profiles, comment trees
- `pip install praw` / `uv add praw`

## Search Strategy: Hybrid (MCP + WebSearch)

Use **multiple** search backends together. Each has strengths:

| Step | Best backend | Why |
|------|-------------|-----|
| **Competitors** | WebSearch + `site:producthunt.com` + `site:g2.com` | Broad discovery + Product Hunt + B2B reviews |
| **Reddit / Pain points** | MCP `web_search` with `engines: reddit` (max 3 keywords!) + MCP Playwright for full posts | PullPush API, selftext in content |
| **YouTube reviews** | MCP `web_search` with `engines: youtube` | Video reviews (views = demand) |
| **Market size** | WebSearch | Synthesizes numbers from 10 sources |
| **SEO / ASO** | WebSearch | Broader coverage, trend data |
| **Page scraping** | WebFetch or MCP `web_search` with `include_raw_content` | Up to 5000 chars of page content |
| **Hacker News** | WebSearch `site:news.ycombinator.com` | HN discussions and opinions |
| **Funding / Companies** | WebSearch `site:crunchbase.com` | Competitor funding, team size |
| **Verified revenue** | WebFetch `trustmrr.com/startup/<slug>` | Stripe-verified MRR, growth, tech stack, traffic |

### Search Availability

Use WebSearch/WebFetch as primary. If MCP `web_search` tool is available, use it for better results (supports engine routing and raw content extraction).

## Steps

1. **Parse the idea** from `$ARGUMENTS`. If empty, ask the user what idea they want to research.

2. **Detect product type** — infer from the idea description:
   - Keywords like "app", "mobile", "iPhone", "Android" → mobile (ios/android)
   - Keywords like "website", "SaaS", "dashboard", "web app" → web
   - Keywords like "CLI", "terminal", "command line" → cli
   - Keywords like "API", "backend", "service" → api
   - Keywords like "extension", "plugin", "browser" → web (extension)
   - Default if unclear → web
   - Only ask via AskUserQuestion if truly ambiguous (e.g., "build a todo app" could be web or mobile)
   - This determines which research sections apply (ASO for mobile, SEO for web, etc.)

3. **Search knowledge base and past work:**
   - If MCP `kb_search` available: `kb_search(query="<idea keywords>", n_results=5)`
   - If MCP `session_search` available: `session_search(query="<idea keywords>")` — check if this idea was researched before
   - Otherwise: Grep for keywords in `.md` files
   - Check if `research.md` or `prd.md` already exist for this idea.

4. **Check existing portfolio** (if MCP codegraph tools available):
   - `codegraph_explain(project="<similar project>")` — architecture overview of related projects in the portfolio
   - `project_code_search(query="<relevant pattern>", project="<sibling>")` — find reusable code, patterns, infrastructure
   - `codegraph_query("MATCH (p:Project)-[:DEPENDS_ON]->(pkg:Package) WHERE pkg.name CONTAINS '<relevant tech>' RETURN p.name, pkg.name")` — find projects using similar tech
   - This helps assess: feasibility, reusable code, stack decisions, and time estimates
   - If no MCP tools available, skip this step.

5. **Competitive analysis** — use WebSearch (primary) + MCP web_search (if available):
   - `"<idea> competitors alternatives 2026"` — broad discovery
   - `"<idea> app review pricing"` — pricing data
   - WebFetch or MCP `include_raw_content=true`: scrape competitor URLs for detailed pricing
   - MCP `engines: reddit` or WebSearch: `"<idea> vs"` — user opinions
   - `"site:producthunt.com <idea>"` — Product Hunt launches
   - `"site:g2.com <idea>"` or `"site:capterra.com <idea>"` — B2B reviews
   - `"site:crunchbase.com <competitor>"` — funding, team size
   - `"site:trustmrr.com <idea>"` or WebFetch `trustmrr.com/startup/<slug>` — Stripe-verified MRR, growth %, tech stack, traffic (24h/7d/30d)
   - For each competitor extract: name, URL, pricing, key features, weaknesses, verified MRR (if on TrustMRR)

6. **User pain points** — use MCP web_search / WebSearch + YouTube:
   - MCP `engines: reddit` or WebSearch: `"<problem>"` — Reddit discussions (**max 3 keywords!**)
   - If Reddit post found but content not available → open via MCP Playwright: `browser_navigate("https://old.reddit.com/r/...")` — old.reddit.com bypasses CAPTCHA
   - MCP `engines: youtube` or WebSearch: `"<problem> review"` — video reviews
   - `"site:news.ycombinator.com <problem>"` — Hacker News opinions
   - WebSearch: `"<problem> frustrating OR annoying"` — broader sweep
   - Synthesis: top 5 pain points with quotes and source URLs

7. **SEO / ASO analysis** (depends on product type from step 2):

   **For web apps:**
   - `"<competitor> SEO keywords ranking"` — competitor keywords
   - `"<problem domain> search volume trends 2026"` — deman