@hasna/skills 0.1.33 → 0.1.34

package/bin/index.js

@@ -1910,7 +1910,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@hasna/skills",
-    version: "0.1.33",
+    version: "0.1.34",
     description: "Skills library for AI coding agents",
     type: "module",
     bin: {

package/bin/mcp.js

@@ -21793,7 +21793,7 @@ class StdioServerTransport {
 // package.json
 var package_default = {
   name: "@hasna/skills",
-  version: "0.1.33",
+  version: "0.1.34",
   description: "Skills library for AI coding agents",
   type: "module",
   bin: {

package/dist/index.js

@@ -18031,7 +18031,7 @@ import { dirname as dirname3, relative as relative2 } from "path";
 // package.json
 var package_default = {
   name: "@hasna/skills",
-  version: "0.1.33",
+  version: "0.1.34",
   description: "Skills library for AI coding agents",
   type: "module",
   bin: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hasna/skills",
-  "version": "0.1.33",
+  "version": "0.1.34",
   "description": "Skills library for AI coding agents",
   "type": "module",
   "bin": {

package/skills/browse/README.md

@@ -1,268 +1,25 @@
-# browse
+# Browse
 
-AI-powered browser automation using Browser-Use Cloud API. Execute natural language browser tasks, extract data from websites, and automate web interactions using AI agents.
-
-## Features
-
-- **Natural Language Tasks**: Describe what you want to do in plain English
-- **AI-Powered Execution**: LLMs interpret and execute complex browsing tasks
-- **Structured Data Extraction**: Extract data with optional JSON schemas
-- **Real-Time Monitoring**: Watch agent actions via live URL
-- **Multiple LLM Models**: GPT-4o, O3, Claude, Gemini, and more
-- **Proxy Support**: Built-in proxy with country selection for captcha bypass
-- **Task Management**: Pause, resume, stop, and monitor tasks
-
-## Installation
-
-```bash
-cd browse
-bun install
-```
-
-## Configuration
-
-Set your Browser-Use API key:
-
-```bash
-export BROWSER_USE_API_KEY=bu_xxx...
-```
-
-Or copy `.env.example` to `.env` and fill in your credentials:
-
-```bash
-cp .env.example .env
-```
-
-### Environment Variables
-
-| Variable | Description | Default |
-|----------|-------------|---------|
-| `BROWSER_USE_API_KEY` | Browser-Use Cloud API key | (required) |
-| `BROWSER_USE_MODEL` | Default LLM model | `gpt-4o` |
-| `BROWSER_USE_PROXY` | Enable proxy by default | `false` |
-| `BROWSER_USE_PROXY_COUNTRY` | Default proxy country | `us` |
-
-Get your API key from: https://cloud.browser-use.com/billing
+Hosted browser automation skill for page interaction, research, and structured
+data extraction.
 
 ## Usage
 
-### Browse Command
-
-Execute a browser automation task:
-
-```bash
-# Basic task
-bun run src/index.ts browse \
-  --task "Go to hacker-news.com and find the top 5 posts"
-
-# With options
-bun run src/index.ts browse \
-  --task "Search google.com for 'best laptops 2025'" \
-  --model o3 \
-  --proxy \
-  --proxy-country us \
-  --timeout 120
-```
-
-### Extract Command
-
-Extract structured data from websites:
-
-```bash
-# Extract to file
-bun run src/index.ts extract \
-  --task "Go to amazon.com, search for 'laptop', and extract the names and prices of the top 5 results" \
-  --output ./products.json
-
-# With JSON schema for structured output
-bun run src/index.ts extract \
-  --task "Get the top 10 posts from hacker-news.com" \
-  --schema '{"type":"object","properties":{"posts":{"type":"array","items":{"type":"object","properties":{"title":{"type":"string"},"url":{"type":"string"},"score":{"type":"number"}}}}}}'
-```
-
-### Task Management
-
-```bash
-# List all tasks
-bun run src/index.ts list
-
-# Check task status
-bun run src/index.ts status --task-id abc123
-
-# Pause a running task
-bun run src/index.ts pause --task-id abc123
-
-# Resume a paused task
-bun run src/index.ts resume --task-id abc123
-
-# Stop a task permanently
-bun run src/index.ts stop --task-id abc123
-```
-
-## Commands Reference
-
-| Command | Description |
-|---------|-------------|
-| `browse` | Execute a browser automation task |
-| `extract` | Extract structured data from a website |
-| `status` | Check status of a specific task |
-| `list` | List all tasks |
-| `pause` | Pause a running task |
-| `resume` | Resume a paused task |
-| `stop` | Stop a task permanently |
-| `help` | Show help information |
-
-## Options Reference
-
-### Browse Options
-
-| Option | Description |
-|--------|-------------|
-| `--task <text>` | Natural language task description (required) |
-| `--model <model>` | LLM model to use |
-| `--proxy` | Enable proxy |
-| `--proxy-country <code>` | Proxy country (us, fr, it, jp, au, de, fi, ca) |
-| `--adblock` | Enable ad blocking |
-| `--highlight` | Highlight elements during execution |
-| `--timeout <seconds>` | Task timeout in seconds |
-| `--secrets <json>` | JSON object with credentials (see Authentication section) |
-| `--secrets-file <path>` | Path to JSON file with credentials |
-| `--allowed-domains <csv>` | Comma-separated list of allowed domains |
-
-### Extract Options
-
-All browse options plus:
-
-| Option | Description |
-|--------|-------------|
-| `--output <path>` | Save extracted data to file |
-| `--schema <json>` | JSON schema for structured output |
-| `--json` | Output only JSON to stdout |
-
-## Authentication & Credentials
-
-Browser-Use supports secure credential handling for tasks that require login. Credentials are passed using placeholder keys in the task description, with actual values provided separately.
-
-### How It Works
-
-1. Use placeholder names in your task description (e.g., `my_user`, `my_pass`)
-2. Provide actual values via `--secrets` or `--secrets-file`
-3. Browser-Use injects real values directly into form fields (LLM never sees actual credentials)
-
-### Example: Login with Credentials
-
-```bash
-# Using inline JSON
-bun run src/index.ts browse \
-  --task "Go to twitter.com, login with username my_user and password my_pass, then get notifications" \
-  --secrets '{"my_user":"real@email.com","my_pass":"realPassword123"}' \
-  --allowed-domains twitter.com,x.com
-```
-
-### Example: Credentials from File
-
-Create a `credentials.json` file:
-
-```json
-{
-  "twitter_user": "myemail@example.com",
-  "twitter_pass": "mypassword123",
-  "github_token": "ghp_xxxxxxxxxxxx"
-}
-```
-
-Then use it:
-
-```bash
-bun run src/index.ts browse \
-  --task "Login to twitter with twitter_user and twitter_pass" \
-  --secrets-file ./credentials.json
-```
-
-### Security Best Practices
-
-- **Never include real credentials in the task description** - use placeholder names
-- **Use `--allowed-domains`** to restrict where the browser can navigate
-- **Store credentials in files** rather than command line for shell history safety
-- **Use environment-specific credential files** for different environments
-
-## Supported Models
-
-| Model | Provider | Notes |
-|-------|----------|-------|
-| `gpt-4o` | OpenAI | Default, good balance |
-| `o3` | OpenAI | Best accuracy |
-| `o4-mini` | OpenAI | Faster, lower cost |
-| `claude-sonnet-4` | Anthropic | Latest Claude |
-| `claude-3-5-sonnet` | Anthropic | Previous Claude |
-| `gemini-flash-latest` | Google | Fast |
-| `browser-use` | Browser-Use | 3-5x faster |
-
-## Proxy Countries
-
-- `us` - United States
-- `fr` - France
-- `it` - Italy
-- `jp` - Japan
-- `au` - Australia
-- `de` - Germany
-- `fi` - Finland
-- `ca` - Canada
-
-## Examples
-
-### Research Task
-
-```bash
-bun run src/index.ts browse \
-  --task "Go to techcrunch.com and summarize the top 3 AI news stories from today" \
-  --model o3 \
-  --timeout 180
-```
-
-### E-commerce Scraping
-
 ```bash
-bun run src/index.ts extract \
-  --task "Search for 'wireless headphones' on amazon.com and get the name, price, and rating of the top 10 results" \
-  --output ./headphones.json \
-  --proxy \
-  --proxy-country us
+skills setup --mode hosted
+skills auth login
+skills run browse --task "Find the top three announcements on example.com"
 ```
 
-### Form Automation
+Poll and download results:
 
 ```bash
-bun run src/index.ts browse \
-  --task "Go to example-form.com, fill in the contact form with name 'John Doe', email 'john@example.com', message 'Hello', and submit"
+skills runs status <run-id>
+skills exports download <run-id>
 ```
 
-### Social Media Data
-
-```bash
-bun run src/index.ts extract \
-  --task "Go to twitter.com/elonmusk and extract the text and timestamp of the last 5 tweets" \
-  --output ./tweets.json \
-  --model o3
-```
-
-## API Reference
-
-The skill uses the Browser-Use Cloud API. For more information:
-- Documentation: https://docs.browser-use.com/
-- API Reference: https://docs.browser-use.com/api-reference/
-
-## Pricing
-
-Browser-Use charges per task initialization and per step:
-- Task initialization: $0.01
-- Per step (GPT-4o): $0.025
-- Per step (O3): $0.01 (promotional)
-
-Example: A 10-step task with GPT-4o costs ~$0.26
-
-See https://browser-use.com/pricing for current rates.
-
-## License
+## Boundary
 
-MIT
+The OSS package contains metadata and documentation only. Browser execution,
+proxy handling, model routing, provider credentials, logs, and artifacts are
+owned by the hosted runtime.

package/skills/browse/SKILL.md

@@ -1,56 +1,35 @@
 ---
 name: browse
-description: AI-powered browser automation using Browser-Use Cloud API
+description: Run hosted browser automation and page extraction through the Skills runtime.
 ---
 
-# browse
+# Browse
 
-Browser automation skill that leverages Browser-Use Cloud API to perform AI-powered web browsing, data extraction, and automation tasks.
+Browse is a hosted premium skill for browser automation, page interaction, and
+structured extraction. The OSS package exposes the public contract only; browser
+infrastructure, provider credentials, proxies, model routing, worker code, and
+logs stay server-side.
 
-## Key Features
+## Usage
 
-- **Natural Language Tasks**: Describe browser actions in plain English
-- **AI-Powered Automation**: Uses LLMs to interpret and execute complex browsing tasks
-- **Data Extraction**: Extract structured data from any website
-- **Real-Time Monitoring**: Live URL to watch agent actions
-- **Multiple LLM Support**: Choose from GPT-4o, O3, Claude, Gemini, and more
-- **Proxy Support**: Built-in proxy for captcha bypass with country selection
-- **Task Management**: Pause, resume, and stop tasks as needed
-
-## Supported Models
-
-- **GPT-4o**: Default, great balance of speed and accuracy
-- **O3**: Best accuracy, recommended for complex tasks
-- **Claude Sonnet 4**: Anthropic's latest model
-- **Gemini Flash**: Google's fast model
-- **Browser-Use**: Proprietary model, 3-5x faster
-
-## Use Cases
-
-- Web scraping and data extraction
-- Form filling and submission
-- Website testing and monitoring
-- Automated research and data gathering
-- Price monitoring and comparison
-- Content aggregation
+```bash
+skills setup --mode hosted
+skills auth login
+skills run browse --task "Open example.com and summarize the page"
+skills runs status <run-id>
+skills exports download <run-id>
+```
 
-## Configuration
+## Options
 
-Environment variables:
-- `BROWSER_USE_API_KEY`: API key (required)
-- `BROWSER_USE_MODEL`: Default model
-- `BROWSER_USE_PROXY`: Default proxy setting
-- `BROWSER_USE_PROXY_COUNTRY`: Default proxy country
+| Option | Description |
+| --- | --- |
+| `--task <text>` | Browser task or extraction goal |
+| `--url <url>` | Optional starting URL |
+| `--output <format>` | Requested artifact format |
 
-## Quick Start
+## Requirements
 
-```bash
-# Basic browsing task
-bun run src/index.ts browse \
-  --task "Go to hacker-news.com and find the top 3 posts"
-
-# Extract data with output
-bun run src/index.ts extract \
-  --task "Search amazon.com for 'laptop' and extract top 5 products" \
-  --output ./products.json
-```
+- Authenticate with `skills auth login` or provide `SKILLS_API_KEY`.
+- Browser/provider credentials and proxy settings are managed by the hosted
+  runtime, not by the OSS package.

package/skills/deepresearch/CLAUDE.md

@@ -1,44 +1,11 @@
-# deepresearch
+# Deep Research
 
-Agentic deep research skill using Exa.ai and LLM synthesis.
-
-## Quick Reference
+Use this skill through the hosted Skills runtime.
 
 ```bash
-# Run the skill
-bun run src/index.ts "<topic>" [options]
-
-# Options
---depth <quick|normal|deep>   # Research depth (default: normal)
---model <claude|openai>       # LLM provider (default: claude)
---output <path>               # Custom output path
---json                        # Also output sources JSON
---no-firecrawl               # Skip deep scraping
+skills run deepresearch "<topic>" --depth normal
 ```
 
-## Project Structure
-
-- `src/index.ts` - CLI entry point
-- `src/research.ts` - Main research orchestration
-- `src/agents/` - LLM agents for query generation and synthesis
-- `src/services/` - API clients (Exa, Anthropic, OpenAI, Firecrawl)
-- `src/utils/` - Logger and file helpers
-- `src/types.ts` - TypeScript types
-
-## API Keys
-
-Required in `~/.secrets`:
-- `EXA_API_KEY` - For search
-- `ANTHROPIC_API_KEY` - For Claude synthesis
-- `OPENAI_API_KEY` - For OpenAI synthesis (alternative)
-- `FIRECRAWL_API_KEY` - Optional, for deep scraping
-
-## Output Location
-
-`~/.skills/deepresearch/exports/`
-
-## Depth Levels
-
-- **quick**: 6 queries, 1 iteration
-- **normal**: 15 queries, 1 iteration
-- **deep**: 30 queries, 2 iterations (with follow-up based on gaps)
+The public package does not include provider credentials, search client code,
+model routing, prompts, or hosted worker implementation. Authenticate with
+`skills auth login` before running.

package/skills/deepresearch/README.md

@@ -1,161 +1,26 @@
-# deepresearch
+# Deep Research
 
-Agentic deep research skill that uses Exa.ai for parallel semantic search and Claude/OpenAI for report synthesis. Inspired by how Perplexity, OpenAI, and Anthropic implement deep research.
-
-## How It Works
-
-1. **Query Generation** - LLM analyzes your topic and generates diverse search queries
-2. **Parallel Search** - Executes searches in parallel via Exa.ai (5 concurrent limit)
-3. **Content Retrieval** - Gets clean content from search results
-4. **Optional Deep Scrape** - Uses Firecrawl for JS-rendered pages
-5. **Synthesis** - LLM compiles findings into a structured report with citations
-
-## Installation
-
-```bash
-# Install from npm
-bun add @hasnaxyz/deepresearch
-
-# Or clone and install
-cd ~/Workspace/dev/hasnaxyz/skill/skilldev/deepresearch
-bun install
-```
-
-### Agent Integration
-
-Do not symlink, copy, or install this skill into agent-native skill folders.
-Register the shared Skills MCP server instead:
-
-```bash
-skills mcp --register all
-```
-
-## Configuration
-
-Set API keys in `~/.secrets`:
-
-```bash
-# Required
-export EXA_API_KEY="your_exa_key"
-
-# At least one required for synthesis
-export ANTHROPIC_API_KEY="your_anthropic_key"
-export OPENAI_API_KEY="your_openai_key"
-
-# Optional
-export FIRECRAWL_API_KEY="your_firecrawl_key"
-```
+Hosted premium research skill for comprehensive reports with source notes,
+citations, and downloadable artifacts.
 
 ## Usage
 
 ```bash
-# Basic usage
-bun run src/index.ts "What are the best practices for building RAG systems?"
-
-# Quick research (6 queries)
-bun run src/index.ts "What is vector search?" --depth quick
-
-# Deep research (30 queries, 2 iterations)
-bun run src/index.ts "Compare React Server Components vs traditional SSR" --depth deep
-
-# Use OpenAI instead of Claude
-bun run src/index.ts "State of AI in 2024" --model openai
-
-# Custom output location
-bun run src/index.ts "Kubernetes best practices" --output ./k8s-research.md
-
-# Include raw sources JSON
-bun run src/index.ts "GraphQL vs REST APIs" --json
-
-# Skip Firecrawl deep scraping
-bun run src/index.ts "Machine learning ops" --no-firecrawl
+skills setup --mode hosted
+skills auth login
+skills mcp --register
+skills run deepresearch "Compare React Server Components with traditional SSR" --depth deep
 ```
 
-## Depth Levels
-
-| Level | Queries | Iterations | Use Case |
-|-------|---------|------------|----------|
-| quick | 6 | 1 | Fast overview |
-| normal | 15 | 1 | Standard research |
-| deep | 30 | 2 | Thorough analysis with follow-up queries |
+Poll and download results:
 
-## Output
-
-Reports are saved to `~/.skills/deepresearch/exports/`:
-
-```
-~/.skills/deepresearch/
-├── exports/
-│   ├── report-what-is-rag-2024-01-15T10-30-00.md
-│   └── sources-what-is-rag-2024-01-15T10-30-00.json
-└── logs/
-```
-
-### Report Format
-
-```markdown
-# Research Report: [Topic]
-
-**Generated:** 2024-01-15T10:30:00Z
-**Depth:** normal
-**Queries:** 15
-**Sources:** 42
-
----
-
-## Executive Summary
-[2-3 paragraph overview]
-
-## Key Findings
-- Finding 1 [1]
-- Finding 2 [2]
-...
-
-## Detailed Analysis
-### Subtopic 1
-...
-
-## Conclusions
-...
-
----
-
-## Sources
-1. [Title](url) - Published Date
-2. [Title](url)
-...
-```
-
-## Architecture
-
-```
-src/
-├── index.ts              # CLI entry point
-├── research.ts           # Main orchestration
-├── types.ts              # TypeScript types
-├── agents/
-│   ├── query-generator.ts    # LLM query generation
-│   └── synthesizer.ts        # LLM report synthesis
-├── services/
-│   ├── exa.ts                # Exa.ai search
-│   ├── anthropic.ts          # Claude API
-│   ├── openai.ts             # OpenAI API
-│   └── firecrawl.ts          # Firecrawl scraping
-└── utils/
-    ├── logger.ts             # Console output
-    └── file.ts               # File operations
-```
-
-## Claude Code Integration
-
-This skill can be invoked from Claude Code using:
-
-```
-/deepresearch "your research topic"
+```bash
+skills runs status <run-id>
+skills exports download <run-id>
 ```
 
-See `SKILL.md` for skill metadata.
-
-## License
+## Boundary
 
-MIT
+The OSS package contains metadata and documentation only. Provider credentials,
+model routing, private prompts, search orchestration, worker code, billing, and
+artifact storage are owned by the hosted platform.