Menu
Web search, read, and summarize tools with provider-based backend
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Long-running iterative development loops. Run arbitrarily-long tasks without diluting model attention. Triggers: ralph, ralph loop, iterative loop, long-running task, development loop.
Full compactor reference — tool parameters, anti-patterns, sandbox languages, context budget, workflows.
Diagnostics — validate config, session DB, runtimes, and troubleshoot compactor issues.
Context management — compact session, recall history, run code, search content.
Stats display — context savings, session metrics, compactions, sandbox and recall/search counters.
Helps you use the /unipi:btw side-conversation workflow effectively. Use when you want to think in parallel, ask side questions without interrupting ongoing work, or inject a side thread back into the main agent.
| name | web |
| description | Web search, read, and summarize tools with provider-based backend |
Use these tools to access web content. The read path uses a local smart-fetch engine by default — free, fast, and no API key required.
Search the web for information. Lower source = simpler/cheaper providers.
Parameters:
query (required): Search query stringsource (optional): Provider selection (1-5)Examples:
web_search(query: "TypeScript generics tutorial")
web_search(query: "latest AI research", source: 4) # Use Tavily
Read and extract content from URLs. Uses the smart-fetch engine by default (source=0 or omitted) — free, local, no API key required. Supports single URL or batch URLs.
Default behavior (source=0):
Parameters:
url (required): Single URL string or array of URLs for batchsource (optional): Provider selection (0=smart-fetch, 1=Jina Reader, 2=Firecrawl, 3=Perplexity)browser (optional): TLS fingerprint profile (default: chrome_145)os (optional): OS fingerprint (default: windows)format (optional): Output format — markdown, html, text, json (default: markdown)maxChars (optional): Maximum content characters (default: 50000)timeoutMs (optional): Request timeout in ms (default: 15000)removeImages (optional): Strip image references (default: false)includeReplies (optional): Include comments/replies (default: extractors)proxy (optional): Proxy URLbatchConcurrency (optional): Concurrent requests for batch (default: 8)verbose (optional): Include metadata header (default: true)Examples:
# Single URL (uses smart-fetch engine by default)
multi_web_content_read(url: "https://example.com/article")
# Batch URLs
multi_web_content_read(url: ["https://example.com/a", "https://example.com/b"])
# Use provider fallback (Jina Reader)
multi_web_content_read(url: "https://example.com/article", source: 1)
# Custom options
multi_web_content_read(url: "https://example.com/article", format: "json", maxChars: 10000)
Summarize URL with LLM. Higher cost (LLM tokens + provider).
Parameters:
url (required): URL to summarizeprompt (optional): Custom summarization promptsource (optional): Provider selection for content fetch (1-3)Examples:
web_llm_summarize(url: "https://example.com/long-article")
web_llm_summarize(url: "https://example.com/research", prompt: "Extract key findings")
source for auto-selection (smart-fetch engine for read, cheapest for search)source number for specific providerSearch providers:
Read providers: 0. Smart-Fetch Engine (free, local) — default
Summarize providers:
The smart-fetch engine is a local content extraction pipeline:
Features:
Configure defaults via /unipi:web-settings → "Smart Fetch Defaults"
Configure providers via /unipi:web-settings command.
Web content is cached for 1 hour by default.
/unipi:web-cache-clear