sc-research 1.0.4 → 1.0.5

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sc-research",
-  "version": "1.0.4",
+  "version": "1.0.5",
   "description": "Headless Social Media Research Data Provider for AI Agents",
   "type": "module",
   "main": "dist/index.js",
@@ -45,8 +45,7 @@
     "dist",
     "templates/base",
     "templates/platforms",
-    "README.md",
-    "CLAUDE.md"
+    "README.md"
   ],
   "publishConfig": {
     "access": "public"

package/templates/base/skills/using_social_media_research.md

@@ -21,10 +21,11 @@ Treat this as the entrypoint skill when user requests map to social-media resear
 
 ## Core Routing Contract
 
-- Run exactly one worker by default.
+- For analysis intents, run exactly one worker by default.
 - Run multiple workers only when the user explicitly asks for multi-view output (for example: "full analysis", "all views", "run everything").
 - Use deep fetch for every worker analysis.
-- Use quick fetch only for explicit quick-answer requests.
+- Use quick fetch for explicit quick-answer requests and ambiguity fallback.
+- If intent is still ambiguous after routing rules, use quick-answer mode (no classified file).
 - Reuse existing raw data only if it still matches topic, source, and date range; otherwise refetch.
 
 ## Step 1: Resolve Intent (Strict Precedence)
@@ -45,7 +46,7 @@ Apply rules top-to-bottom:
 4. **Inferred strongest intent**
    - Map by primary question intent (keywords table below).
 5. **Fallback**
-   - If still ambiguous, default to `social_media_rank`.
+   - If still ambiguous, default to quick-answer mode (no classified file).
 
 ## Step 2: Map Intent to Route
 

package/CLAUDE.md

@@ -1,67 +0,0 @@
-# Social Media Research Skill
-
-> **"The Skill provides the catch; the Agent cooks the meal."**
-
-This project is a headless social media research tool designed for **Claude Code first**. It fetches raw discussion data from Reddit and X (Twitter), classifies it into multiple analysis templates, and visualizes results in a local dashboard.
-
-## Canonical Source
-
-- **Main source of truth:** `templates/base/` (skills + commands body templates)
-- **Platform rules source:** `templates/platforms/*.json` (paths/format behavior)
-- `.claude/`, `.cursor/`, `.agent/`, and other platform folders are rendered outputs from `sc-research init`.
-
-## ⚡️ Quick Start
-
-This project uses `bun`.
-
-### Environment Variables
-Required in `.sc-research`:
-- `OPENAI_API_KEY`: For Reddit URL discovery and general LLM ops
-- `XAI_API_KEY`: (Optional) For X/Twitter data
-
-### Available Commands
-- `/research <topic>` - Research + quick answer flow
-- `/quick <topic>` - Fast Reddit-only quick answer
-- `/deep-research <topic>` - Deep research + route to best template
-- `/rank` - Generate ranking report (`classified_rank.json`)
-- `/sentiment` - Generate sentiment report (`classified_sentiment.json`)
-- `/trend` - Generate timeline report (`classified_trend.json`)
-- `/controversy` - Generate controversy map (`classified_controversy.json`)
-- `/discovery` - Generate discovery clusters (`classified_discovery.json`)
-- `/visualize` - Launch dashboard (auto-detects available `classified_*.json`)
-- `/test-research` - Run fixed-link pipeline test
-
-### CLI Init Targets
-- `sc-research init --ai claude` - Install Claude templates
-- `sc-research init --ai claude,cursor` - Install multiple targets in one run
-- `sc-research init --ai all` - Install all supported targets (`claude`, `cursor`, `windsurf`, `antigravity`)
-
-## 🏗️ Architecture
-
-- **Orchestrator**: `using_social_media_research` (The routing brain)
-- **Workers**: 
-  - `social_media_fetch` (Data provider)
-  - `social_media_rank` (Ranking analysis)
-  - `social_media_sentiment` (Sentiment analysis)
-  - `social_media_trend` (Timeline analysis)
-  - `social_media_controversy` (Debate analysis)
-  - `social_media_discovery` (Viral topic clustering)
-  - `social_media_visualize` (Dashboard launcher)
-  - `social_media_research_test` (Legacy/fixed-link test helper)
-
-## 🧪 Testing Rules
-
-> **IMPORTANT**: Automated tests MUST NOT invoke paid AI APIs.
-> Use `bun run test:fixed` for manual verification or the `/test-research` command.
-
-## 🔒 Project Rules
-
-- Always use `bun` for runtime and scripts.
-- Do not run paid AI API calls in automated tests.
-
-## 📂 Project Structure
-
-- `src/`: Core TypeScript logic
-- `.claude/skills/`: AI Skill definitions
-- `.claude/commands/`: User slash commands
-- `web/`: React/Vite visualization app