sboghossian Tool — Web Search Orchestrator
What it does
The Web Search Orchestrator manages outbound web queries for legal research tasks where the pre-indexed corpora (firm KB, public legal corpus) do not have current information. It enforces a source allowlist, batches multiple sub-queries efficiently, and returns structured results with provenance so that downstream analysis can cite sources accurately.
This is the tool that asks: "Is this question about a recent change that my training data or pre-indexed corpus might have missed?" If yes, it goes to the web.
When to invoke
Invoke when:
- Freshness required: the question is about a regulatory change, court decision, or law amendment from the last 6–12 months
- Official current text needed: the user needs the current version of a regulation or statute that may have been amended recently
- Regulator bulletin: SAMA circular, CBUAE guidance note, DFSA consultation paper, SDAI decision — published in the last year
- Breaking news with legal implications: a major court judgment, a legislative development, a sanctions designation just announced
- The local KB is silent: neither firm KB, personal KB, nor the public legal corpus returned a relevant result
Do not invoke for:
- Statute text that is well-established and unlikely to have changed (use [[tool-rag-public-legal-corpus]] or [[tool-legifrance-fr]])
- Generic legal concepts (limitation periods, consideration, force majeure) — the model already knows these
- Boilerplate clause drafting — no web access needed
- Queries the firm KB already answered comprehensively
- Fact-finding that requires accessing paywalled content (Westlaw, LexisNexis) — use those tools directly
Source allowlist
The orchestrator enforces a strict source priority hierarchy. Sources are only queried or returned if they fall within the allowlist defined per tenant ([[tool-web-search-source-allowlist]]):
Tier 1 — Official primary sources (highest authority)
- Official Gazettes: Umm Al-Qura (KSA — https://www.uqn.gov.sa), UAE Federal Gazette (https://moj.gov.ae/ar/legislations), Journal Officiel (France — via Legifrance), al-Jarida al-Rasmiyya (Lebanon), Official Gazette of Egypt
- Court databases: DIFC Courts (difccourts.ae), ADGM Courts (adgmcourts.com), BAILII (UK), CourtListener (US), EUR-Lex (EU)
- Regulator websites: SAMA (sama.gov.sa), CBUAE (centralbank.ae), BDL (bdl.gov.lb), CMA (cma.org.sa), SDAIA (sdaia.gov.sa), DFSA (dfsa.ae), FSRA (fsra.adgm.com), DIFC (difc.ae), ADGM (adgm.com)
Tier 2 — Established legal publishers
- Practical Law (UK/Int'l) — uk.practicallaw.thomsonreuters.com
- LexisNexis news and analysis — www.lexisnexis.com
- IFLR (International Financial Law Review)
- Kluwer Arbitration Blog
- Out-Law (Pinsent Masons legal news)
- AL-Bayan Legal (Arabic-language legal analysis)
- Financier Worldwide
- Bloomberg Law News
- Law Society Gazette (UK)
Tier 3 — Bar associations and professional bodies
- Law Society of England and Wales
- American Bar Association
- DIFC Dispute Resolution Authority
- Saudi Bar Association
Tier 4 — Secondary / commentary (verify with primary)
- Major law firm client alerts (Clifford Chance, Freshfields, Al Tamimi, Baker McKenzie, Hadef & Partners, Dentons, etc.)
- Academic institutions' legal research blogs (Harvard Law Review, Oxford Human Rights Hub)
Never use as primary source:
- Wikipedia — wrong for legal authority
- Reddit / legal forums — unverifiable
- News aggregators (Google News snippets, Yahoo News) — not primary
- LinkedIn posts — not authoritative
Batching logic
Efficient web search batches related sub-queries:
User: "What are the latest UAE cybercrime law amendments and TDRA guidance?"
Decompose to:
sub-query 1: "UAE Federal Decree-Law cybercrime amendments 2024 2025" → site:moj.gov.ae OR site:tdra.gov.ae
sub-query 2: "TDRA UAE cybersecurity guidance circular 2024" → site:tdra.gov.ae
sub-query 3: "UAE cybercrime law enforcement cases 2024" → site:adgmcourts.com OR site:difccourts.ae
Execute all 3 in parallel → merge results → deduplicate → rank
Maximum sub-queries per orchestrated call: 6. Beyond 6, apply iterative round — run the most important 6, evaluate gaps, then run the next round.
Capabilities
Structured query execution
Input: {
queries: [
{ q: "SAMA AML circular 2025", sites: ["sama.gov.sa"], dateFrom: "2025-01-01" },
{ q: "SAMA AML penalties enforcement", sites: ["sama.gov.sa", "out-law.com"] }
],
jurisdiction: "KSA"
}
Output: {
results: [
{
url, title, snippet, datePublished,
sourceType: "regulator" | "court" | "publisher" | "bar" | "secondary",
sourceTier: 1 | 2 | 3 | 4,
relevanceScore: 0.88
}
]
}
Gazette scraping
For MENA official gazettes, the orchestrator can extract recent legislative instruments:
Input: { gazette: "umm-al-qura", dateFrom: "2025-01-01", topic: "company law" }
Output: [ { title, issuedDate, arabicText, summary } ]
Regulator bulletin feed
For regulators that publish RSS feeds or structured announcement pages:
Input: { regulator: "cbuae", since: "2025-06-01" }
Output: [ { title, date, url, type: "circular" | "guidance" | "press_release" } ]
Output format
Per result returned:
{
"url": "https://www.sama.gov.sa/...",
"title": "SAMA AML/CFT Annual Progress Report 2024",
"snippet": "...",
"datePublished": "2025-03-15",
"sourceType": "regulator",
"sourceTier": 1,
"relevanceScore": 0.91,
"language": "ar",
"translationRequired": true,
"verifiedPrimary": true
}
The verifiedPrimary flag is set to true when the result is from a Tier 1 source (official government or regulator) and the URL directly matches the expected domain.
Anti-patterns
| Anti-pattern | Why it's wrong |
|---|---|
| Citing Wikipedia for a statute | Wikipedia is not a primary source; statutes must come from official gazettes or Legifrance |
| Using news aggregator snippets as legal authority | News snippets are second-hand; always trace to the primary source |
| Forums as legal authority | Unverifiable; legally irrelevant |
| Not checking date | A result from 3 years ago may be superseded; always report and check datePublished |
| Accepting paywalled content snippet as full analysis | The snippet may not reflect the full picture; note truncation |
Failure modes
| Failure | Symptom | Resolution |
|---|---|---|
| No Tier 1 results | All results from Tier 3–4 only | Flag to user; primary source may not be indexed online |
| Outdated results | All results > 18 months old | Flag; report gap; recommend direct regulator contact |
| Arabic-only source | Result is Arabic text, no English | Flag; route to [[tool-ocr-arabic]] or translation if critical |
| CAPTCHA / bot detection | Web request blocked | Retry with backoff; route to official API if available |
| Paywalled result | 403 or login wall | Skip; report URL to user for manual access |
Related skills
- [[tool-web-search-source-allowlist]] — per-tenant configuration of trusted source domains
- [[tool-legal-data-hunter]] — orchestrator that calls this tool plus database tools
- [[tool-rag-public-legal-corpus]] — pre-indexed corpus; check here before invoking web search
- [[tool-legifrance-fr]] — direct official API for French law (preferred over web scraping)
- [[tool-lexisnexis]] — premium database with more complete content than web search for supported jurisdictions