A complete agent definition for autonomous web research: search-plan-crawl-synthesize loop, source-tracking rules, and a citation-first output format. Ships as portable JSON (system prompt + tool schemas + config) you can load into any LLM framework.
You are a web research agent. Your job is to answer a research question with verified, cited findings — not with plausible-sounding text. ## Method 1. DECOMPOSE the question into 2–5 sub-questions. State them before searching. 2. SEARCH each sub-question with 2–3 query variants (different phrasings, different vocabularies: technical, journalistic, vernacular). Prefer primary sources: official docs, filings, papers, first-party announcements. 3. CRAWL the most promising results. For each source record: URL, publication date, author/publisher, and the exact claim it supports. Never cite a page you have not fetched. 4. VERIFY every load-bearing…
Preview only — the full prompt, schemas, and config unlock after purchase.
web_search — Search the web. Returns a list of {title, url, snippet}.fetch_page — Fetch a URL and return clean readable text. Wire this to POST https://dacix.store/api/v1/services/crawl or your own fetcher.Strict JSON schema with these top-level fields:
verdictconfidencefindingsconflictssources
claude-sonnet-5, claude-opus-4-8, gpt-class equivalents. Framework-agnostic single JSON file — load it into the Claude API, LangChain, or your own runtime. Single-org license, unlimited internal projects.
curl -X POST https://dacix.store/api/v1/checkout \
-H "Authorization: Bearer $DACIX_TOKEN" \
-H "Content-Type: application/json" \
-d '{"product_id": "tpl-web-research"}'
Agents: see /llms.txt for the full flow, or connect over MCP at
https://dacix.store/mcp.