Agenda Intelligence MD

Decide whether another agent's claim-backed output is safe to relay onward. Use before forwarding, publishing, or acting on a downstream agent's answer that cites evidence: it reports which claims are grounded, which are unsafe to relay, and which evidence references are orphaned. Pass audit_json matching evidence-audit.schema.json (claims, evidence records, optional unsupported_claims). Returns a relay verdict with per-claim findings and owner actions. Evidence-readiness only: it does not verify factual truth, fetch or validate cited sources, or authorize an action.

Triage the trust evidence for an agent-mediated interaction (identity, operator or principal authorization, tool scope, session authentication, action intent) before a high-stakes action executes. Pass a structured trust_request (actor, target_surface, requested_action, dated_sources, risk_question, decision_stage) matching agentic-interaction-trust-request.schema.json. Returns a triage recommendation, trust signal, decision-readiness score, and the specific missing trust evidence. Evidence triage only: not cybersecurity monitoring, identity verification, or authorization; human review is required.

Generate an auditable strategic-risk memo from a structured Agenda request. Use for sanctions, regulatory, geopolitical, trade, corridor, or policy-risk questions where the agent needs a memo with assumptions, scenarios, evidence discipline, and regional routing. Pass request matching agenda-request.schema.json. Returns a validated agenda-memo; when ANTHROPIC_API_KEY is unset it returns the assembled system_prompt for the host model to complete. No live source retrieval and no legal, compliance, financial, or investment advice.

Append a claim and its sources to an evidence pack and re-validate the result. Use when building an evidence pack incrementally as sources are read, instead of assembling the whole document by hand. Omit pack_json to start a new pack (topic is then required). A claim whose text already exists gains the new sources instead of being duplicated, and unsupported_claims is kept consistent with per-claim support_status. support_status is never inferred as supported: omitted it defaults to unsupported (no sources) or partially_supported (sources supplied). Returns the updated pack to the caller; it does not write files, fetch URLs, verify quotes, score source reliability, or verify factual truth.

Validate a claim-level evidence audit and summarize support quality. Use after drafting or receiving a memo to check whether important claims point to evidence IDs with explicit support levels, uncertainty hooks, and risk-if-wrong notes. Pass audit_json matching evidence-audit.schema.json. Returns validity, support-level distribution, orphan evidence references, and unsupported-claim counts. It does not verify factual truth or source reputation.

Check a caller-provided evidence packet before human review. Use when an AI output declares claims, source IDs, optional verbatim quotes, and the full supplied source text. Returns packet_complete, source_review_required, or packet_incomplete per claim and overall, with broken references, quote mismatches, lexical-support gaps, unmatched numbers, and owner actions. Deterministic and local-text only: it does not retrieve sources, score source authority, assess factual truth, or authorize an action.

Check a schema-shaped Agenda memo against post-hoc evidence-readiness quality guardrails. Use after validate_memo or on any external model memo to catch schema-valid but unsafe output: approval/clearance overreach, hidden evidence gaps, generic monitoring, weak owner actions, or evidence-mode discipline failures. Returns schema_valid separately from ok; it does not verify factual truth.

Triage secondary-sanctions exposure for a CIS-domiciled counterparty (Kazakhstan, Uzbekistan, Kyrgyzstan, Tajikistan, Turkmenistan, Georgia, Armenia, Azerbaijan, Moldova) for EU / UK / UAE / Singapore enhanced due diligence against OFAC EO 14114, the EU sanctions package, UK OFSI, and FATF / EAG typologies. Pass a structured exposure_request (counterparty, exposure_facets, jurisdiction_review_scope, dated_sources, risk_question, decision_stage) matching cis-secondary-sanctions-request.schema.json. Returns a triage recommendation, decision-readiness score, exposure dimensions, evidence gaps, and minimum sources before review. Local stdio runs on user-supplied evidence only (no live retrieval); a name match is not identity verification; human review is required.

Evaluate project finance bankability and IFI covenants (EBRD, ADB, EU Global Gateway) for Trans-Caspian and Middle Corridor infrastructure projects. Evaluates minimum DSCR floor (1.20x), non-sovereign margin (1.30x), leverage ceiling (<=80%), Caspian hydrological water-level constraints (-1.20m Baltic datum), and FX currency mismatch. Returns a free Decision Teaser with covenant pass/fail matrix, bottleneck analysis, and an x402 micropayment invoice to unlock the full 15-year debt waterfall model.

Assemble an agenda brief from supplied fields and report which required fields are still missing. Use when producing a brief inside the protocol instead of hand-building JSON: call it with whatever is known so far, read missing_required, and call again with the remaining fields. Call with no arguments to get an empty scaffold plus the required field list. evidence_mode defaults to reasoning_only. Returns the assembled brief object and its schema errors to the caller; it does not write files, retrieve sources, draft prose, or verify factual truth.

Reserved placeholder for a future Agenda Intelligence v2 deep-dive workflow. Do not use for current detailed analysis. For production work today, call analyze with request.depth set to scenario or red_team. This tool only returns a planned status message and performs no analysis.

Generate actionable self-correction instructions for an LLM agent from an evidence packet request. Inspects validation and claim-level issues (missing sources, misquoted excerpts, unmatched numbers, polarity/negation mismatches, weak lexical support) and formats a structured markdown prompt for the agent to revise its claims and citations.

Return the full markdown for one packaged regional or sector lens. Use after list_lenses when an agent needs the actual specialist context, such as the Central Asia/Caspian or sanctions lens, for a strategic-risk task. Pass lens_type and lens_id exactly as listed. Returns static markdown; it does not retrieve live events or decide which lens should be used.

Return packaged Agenda Intelligence protocol markdown. Use when an agent needs the reasoning contract, evidence-discipline rules, or operating instructions before producing strategic-risk analysis. Pass name='entrypoint' for the main protocol. Returns markdown text from the installed package; it does not analyze a question or validate user data.

Return a packaged Agenda Intelligence JSON Schema so an agent can construct a valid payload before calling validate_brief, validate_evidence, validate_memo, analyze, or a vertical worker. Pass name as the schema key (for example agenda_brief, evidence_pack, agenda_memo, middle_corridor_deal_risk_request), its file name, or its bare stem; omit name to list the available schema keys. Returns the schema document and its version. Contract discovery only: it does not validate data, fill in a template, or verify factual truth.

Return one packaged strategic-risk signal markdown file by ID. Use after list_signals when an agent needs the full text of a specific archived signal for context or examples. Pass signal_id without the .md extension. Returns static markdown from the installed package; it does not fetch live updates.

Check whether a caller-supplied corpus of source texts lexically supports each claim. Use when you have claims plus the full text of the sources they should rest on and need a deterministic grounded/weakly_grounded/ungrounded verdict per claim before human review. Pass request_json matching grounded-check-request.schema.json (claims with claim_id/claim_text and optional verbatim quotes, corpus documents with corpus_id/text). Returns per-claim grounding status, coverage, best-matching passage, unmatched numeric values, quote checks, and owner actions. Local-text only: no outbound requests, no source discovery, no source-reliability scoring, and no factual-truth verification — grounding in a wrong corpus does not make a claim true.

Triage maritime sanctions and chokepoint-disruption exposure for a vessel/voyage transiting the Strait of Hormuz, Persian/Arabian Gulf, Gulf of Oman, Bab-el-Mandeb, or Red Sea (Iran-oil, Russia price-cap, dark-fleet, STS transfer, flag-hopping, P&I gap, AIS manipulation). Pass a structured exposure_request (vessel/voyage, route, cargo, counterparties, dated_sources, risk_question, decision_stage) matching gulf-maritime-exposure-request.schema.json. Returns a triage recommendation, exposure signal, decision-readiness score, supplied vs. minimum-required sources, and evidence gaps. Pre-compliance evidence triage only: no live retrieval, does not resolve vessel ownership or verify identity, no legal or sanctions advice; human review is required.

Grade a Kazakhstan market-entry file (distribution, import, service, showroom, EPC, renewable-energy, infrastructure, technology-transfer, or partner-entry) against a staged source-requirement taxonomy before a launch, budget, or partner commitment. Pass a structured readiness_request (entry_mode, sector, market_entry_file, dated_sources, decision_stage) matching market-entry-readiness-request.schema.json. Returns a gate decision, readiness label, evidence gaps, claim audit, owner actions, and watch-next indicators. Evidence triage only: not legal, compliance, customs, tax, sanctions, or launch-authorization advice; no live retrieval; human review is required.

List packaged regional and sector lens IDs available to Agenda Intelligence. Use before get_lens when an agent needs to discover which geography or sector reference packs can be loaded. Optionally filter by lens_type='regional' or 'sector'. Returns metadata only; it does not return full lens markdown or run analysis.

List packaged strategic-risk signal records vendored from Global Think Tank Analyst. Use to discover available signal IDs before calling get_signal, or to show a static archive index inside an agent workflow. Returns the packaged signals/index.json snapshot. Read-only and offline: it does not fetch live news or update the archive.

List source requirement category slugs packaged with Agenda Intelligence. Use this first when you do not know which category to pass to source_plan or source_coverage. Returns category IDs and per-pack counts. Discovery only: it does not discover sources, validate coverage, or verify factual truth.

Screen a Kazakhstan / Middle Corridor (Trans-Caspian) trade deal for sanctions-adjacent and corridor risk before signature, shipment, insurer handoff, or committee review. Pass a structured deal_risk_request (route, cargo, counterparties, dated_sources, risk_question, decision_stage) matching middle-corridor-deal-risk-request.schema.json. Returns a triage recommendation, risk signal, decision-readiness score, supplied vs. minimum-required source categories, evidence gaps, and a high-risk-jurisdiction presence flag. Pre-compliance evidence triage only: no live retrieval, no factual-truth verification, no legal or sanctions advice; human review is required.

Route a caller-controlled action to continue, request_evidence, require_approval, or stop using caller-supplied claim evidence, risk tier, policy checks, and an optional external approval reference. Resubmit the same run_id after adding evidence or approval. Readiness only: the tool does not authenticate, authorize, enforce, persist state, or perform the action.

Score a before/after pair of agenda-analysis text with the bundled heuristic rubric. Use in evals or demos to compare whether an Agenda Intelligence rewrite improved structure, evidence labeling, uncertainty handling, and decision-readiness. Pass before_text and after_text as plain strings. Returns a heuristic score and breakdown; it is not a factuality, legal, compliance, or investment judgment.

Diagnose whether an evidence pack covers the must_check source types for a category. Use after collecting evidence to find source gaps before relying on a memo. Pass evidence_json and optionally category; if category is omitted, the tool uses evidence_json.source_category. Returns matched and missing source types. It does not discover new sources, verify truth, or change validate_evidence results.

Return required source categories for a strategic-risk evidence pack. Use before collection or review to know which source types should be checked for a domain such as sanctions, elections, conflict, cyber, or energy. Pass the source category slug as category. Returns a checklist of must_check and optional source types; it does not search the web, fetch documents, or validate an evidence pack.

Validate a caller-provided agenda brief against agenda-brief.schema.json. Use before running scoring, evidence audit, or publication steps to catch missing sections and schema drift. Pass the parsed brief object as brief_json. Returns validation status and schema errors only; it does not judge factual truth, retrieve sources, or improve the brief.

Validate a caller-provided evidence pack against evidence-pack.schema.json. Use when you need to confirm that claims, evidence IDs, provenance fields, and optional source_category metadata are structurally usable by Agenda Intelligence. Pass the parsed evidence pack as evidence_json. Returns schema validity and errors; it does not verify whether evidence is true, current, or sufficient.

Validate an Agenda memo against agenda-memo.schema.json. Use after a host model or external process drafts a memo and before treating it as an Agenda Intelligence artifact. Pass the parsed memo as memo_json. Returns validity and schema errors; it does not score truthfulness, retrieve sources, or rewrite the memo.

Issue a bounded factual Claim Verdict from caller-supplied evidence records. Evaluates freshness, authoritative source class, independent source groups, conflicts, jurisdiction, and exact subject identifiers as of a declared date. Returns verified, contradicted, partially_supported, unresolved, or not_verifiable. No source discovery or live retrieval; verified means the declared evidence threshold is met, not absolute truth. Human review remains required.

Check whether quoted fragments appear in caller-provided source text. Use when you have local excerpts and need to catch citation drift or misquoted snippets. Accepts an evidence pack (sources/evidence items with a quote) or an evidence-audit doc with claims[].supporting_quotes; span checks carry the originating claim_id. Pass pack_json plus texts mapping evidence_id to plain text. Returns present, absent, and missing_source_text results. Local-text only: it does not make outbound requests, discover sources, score source reputation, gather news, or verify factual truth.