# Agenda Intelligence MD

Primary interface: a deterministic evidence-packet linter for claim-backed AI output. `agenda-intelligence check <packet.json>` validates claim/source references, checks declared quotes verbatim, reports lexical-support gaps and unmatched numbers, and returns `packet_complete`, `source_review_required`, or `packet_incomplete`. `factuality_status` is always `not_assessed`; human review is always required. Contracts: schemas/v1/evidence-packet-request.schema.json and schemas/v1/evidence-packet-response.schema.json.

Compatibility surfaces remain available: the strategic-intelligence product shell, MCP server, HTTP API, A2A adapter, Cloudflare Workers, reasoning modules, and vertical profiles. They are not the primary product narrative and do not establish buyer traction.

Product MCP tools:
- check_evidence_packet — primary deterministic preflight for claim/source packets; checks references, declared quotes, lexical support, and unmatched numbers over caller-supplied source text. No retrieval or factuality determination; human review required.
- analyze — accepts an AgendaRequest (schemas/v1/agenda-request.schema.json), routes geography to bundled regional references, assembles a system prompt from the Global Think Tank Analyst method plus matched regional lenses, optionally calls the Anthropic API when ANTHROPIC_API_KEY is set, and returns a memo validated against schemas/v1/agenda-memo.schema.json.
- validate_memo — validates a third-party memo against agenda-memo.schema.json and returns errors plus a score where implemented.
- check_memo_quality — checks schema validity plus post-hoc evidence-readiness quality guardrails for a third-party memo; catches schema-valid approval/clearance overreach, hidden evidence gaps, generic monitoring, weak owner actions, confidential project-room alias leaks, and evidence-mode discipline failures.
- list_signals / get_signal — read access to the vendored Global Think Tank Analyst signal archive.
- deep_dive — reserved stub; returns a v2 placeholder pointing callers at `analyze` with depth: scenario or red_team.
- middle_corridor_deal_risk — Kazakhstan / Middle Corridor deal-risk gate; structured deal_risk_request (middle-corridor-deal-risk-request.schema.json) in, triage recommendation + decision-readiness + evidence gaps out. Pre-compliance triage; human review required.
- cis_secondary_sanctions_exposure — CIS counterparty secondary-sanctions exposure triage (cis-secondary-sanctions-request.schema.json) for EU/UK/UAE/Singapore EDD. Local stdio runs on user-supplied evidence only; a name match is not identity verification.
- agentic_interaction_trust — trust-evidence triage for an agent-mediated interaction (agentic-interaction-trust-request.schema.json) before a high-stakes action. Evidence triage only; human review required.
- gulf_maritime_exposure — maritime sanctions + chokepoint-disruption exposure triage (gulf-maritime-exposure-request.schema.json) for a vessel/voyage transiting Strait of Hormuz / Persian-Arabian Gulf / Gulf of Oman / Bab-el-Mandeb / Red Sea. No live retrieval; does not resolve vessel ownership or verify identity; human review required.

Vertical worker on HTTP + A2A (no MCP stdio tool):
- kazakhstan_market_entry_readiness — Kazakhstan market-entry readiness gate (market-entry-readiness-request.schema.json + market-entry-readiness-response.schema.json) for distribution / import / service / showroom / EPC / renewable-energy / infrastructure / technology-transfer / partner-entry files; company + project + Kazakhstan objective + counterparties + supplied sources in, gate decision + readiness label + evidence gaps + claim audit + owner actions + watch-next out. Service function + HTTP POST /v1/market-entry/readiness + A2A profile; deployed Cloudflare Worker live at https://kazakhstan-market-entry-readiness-a2a.vassiliy-lakhonin.workers.dev. Source taxonomy: source-requirements/kazakhstan-market-entry-readiness.json; use-case doc: docs/use-cases/kazakhstan-market-entry-readiness.md; contract examples: examples/kazakhstan-market-entry-readiness/contract/. Evidence triage only; not legal, tax, IP, data-compliance, customs, or sanctions advice; human review required.

Commercial discovery discipline:
- There is no active default commercial wedge in this repository. Do not point users to removed discovery teardowns, profile packs, or public-signal corpus artifacts.
- Public profiles should be crawlable, claim-addressable evidence packets: source IDs, buyer questions, project/system claims, missing evidence, owner actions, readiness decision, boundary notes, and follow-up signal.
- Agent-readable concept map: okf/index.md, entitymap.json, and docs/agent-entity-map.md. Live worker copies are available at https://agenda-intelligence-a2a.vassiliy-lakhonin.workers.dev/okf/index.md and https://agenda-intelligence-a2a.vassiliy-lakhonin.workers.dev/entitymap.json. Use these to understand core concepts before interpreting vertical workers.
- Agentic Resource Discovery catalog: .well-known/ai-catalog.json, live at https://agenda-intelligence-a2a.vassiliy-lakhonin.workers.dev/.well-known/ai-catalog.json. Worker API catalog and OpenAPI are live at https://agenda-intelligence-a2a.vassiliy-lakhonin.workers.dev/.well-known/api-catalog and https://agenda-intelligence-a2a.vassiliy-lakhonin.workers.dev/api/openapi.json. These advertise capabilities/resources, not buyer traction.
- Confidential project-room workflow: docs/trust/confidential-project-workflow.md plus the live public contract at https://agenda-intelligence-a2a.vassiliy-lakhonin.workers.dev/profiles/confidential-project-room and synthetic redacted example at https://agenda-intelligence-a2a.vassiliy-lakhonin.workers.dev/profiles/confidential-project-room/redacted-example.json. Use aliases and source IDs; never expose client names or private source text in public artifacts.
- Success is buyer behavior: redacted file, second profile request, paid concierge review interest, budget-owner intro, or concrete workflow correction. Traffic, citations, likes, and "interesting" feedback are not traction.

Geography routing terms (lowercased substring match against question, decision_context, geography). Modules union, not exclusive. Authoritative source: src/agenda_intelligence/product.py. Guard: tests/test_product_shell.py::test_routing_terms_documented_in_canon.

global-think-tank-analyst: always loaded as the core reasoning method.

central-asia-caspian (CA_CASPIAN_TERMS):
almaty, azerbaijan, baku, caspian, central asia, georgia, kazakhstan, kyrgyzstan, middle corridor, tajikistan, tashkent, tcita, tcitr, turkmenistan, uzbekistan.

gulf-middle-east (GULF_ME_TERMS):
arabian gulf, bab el mandeb, bab-el-mandeb, bahrain, gcc, gulf, hormuz, iran, iraq, ksa, kuwait, levant, middle east, oman, persian gulf, qatar, red sea, saudi arabia, strait of hormuz, uae, united arab emirates, yemen.

eu (EU_TERMS; plus exact-token geography match against EU/Europe):
brussels, cbam, cjeu, ecb, eu ai act, eu enforcement, eu regulation, european central bank, european commission, european council, european parliament, european union, gdpr, nis2, schrems.

sanctions (SANCTIONS_TERMS, sector module):
entity list, export control, export controls, ofac, sanctions, secondary sanctions.

Core rule: do not summarize the news. Identify what changed, who is affected, what incentives shifted, what remains unknown, and what to watch next.

Repository: https://github.com/vassiliylakhonin/Agenda-Intelligence-md
Skill path: skills/agenda-intelligence/SKILL.md

Recommended use:
- policy and geopolitical analysis;
- sanctions and compliance monitoring;
- trade and regulatory risk briefs;
- founder/investor operating-context notes;
- red-team checks on confident narratives.

Related full memo skill: https://github.com/vassiliylakhonin/global-think-tank-analyst
Use Agenda-Intelligence.md for lightweight agenda triage; use global-think-tank-analyst for full policy-risk memos.

Regional lens packs:
- Central Asia + Caspian: skills/agenda-intelligence/references/regional/central-asia-caspian.md
- Middle East: skills/agenda-intelligence/references/regional/middle-east.md
- European Union: skills/agenda-intelligence/references/regional/eu.md

Top-level protocol: Agenda-Intelligence.md
Role: sits beside files such as AGENTS.md as a conditional public-agenda reasoning layer.
Use it only for news, policy, regulation, sanctions, geopolitics, trade, markets, conflict, or strategic-risk analysis.

Before/after examples:
- examples/before-after/eu-ai-act.md
- examples/before-after/red-sea-shipping.md
- examples/before-after/sanctions-routing.md
- examples/before-after/evaluation-rubric.md

Adoption guide: ADOPTION.md

Sector lens packs:
- Sanctions: skills/agenda-intelligence/references/sector/sanctions.md

Signal lifecycle tracker:
- Protocol: skills/agenda-intelligence/references/signal-lifecycle.md
- Schema: schemas/v1/signal-tracker.schema.json
- Stages: detected → developing → escalated → stable → resolved → archived

AnalysisBank: analysis-bank/README.md
Reasoning memory format: analysis-bank/MEMORY_FORMAT.md
AnalysisBank lifecycle lint: agenda-intelligence memory-lint
AnalysisBank retrieval: agenda-intelligence memory-search <query> (active, unexpired lessons by default; use --include-inactive only for audits)
AnalysisBank retrieval bench: agenda-intelligence memory-search-bench tests/fixtures/analysis_bank_retrieval/manifest.json
AnalysisBank applicability bench: agenda-intelligence memory-applicability-bench tests/fixtures/analysis_bank_applicability/manifest.json
Analyze Reasoning Memory: analyze selects at most 3 active, retrieval-ranked, applicability-gated AnalysisBank lessons into a bounded prompt section and memo audit.reasoning_memory trace; they are guardrails, not factual evidence.
AnalysisBank hardening checkpoint: docs/product/analysisbank-hardening-checkpoint.md
Eval script: scripts/eval_before_after.py

Skills:
- skills/agenda-intelligence/SKILL.md — OpenClaw skill wrapper for agenda analysis
- skills/source-ingest/SKILL.md — normalize documents (PDF/DOCX/URL) into structured source records

Agent manifest: agent-manifest.json
Schemas:
- schemas/v1/agenda-brief.schema.json
- schemas/v1/evidence-pack.schema.json
- schemas/v1/evidence-audit.schema.json
- schemas/v1/memory-card.schema.json
- schemas/v1/lens-manifest.schema.json
- schemas/v1/signal-classification.schema.json
- schemas/v1/signal-tracker.schema.json
Packaged CLI: agenda-intelligence
Legacy script wrapper: scripts/agenda_intelligence.py
MCP server: agenda-intelligence-mcp
MCP docs: MCP.md and docs/integrations/mcp.md

Source Planning Layer:
- SOURCE_POLICY.md
- source-taxonomy.json
- source-requirements/*.json
CLI source/eval commands: start, source-categories, source-plan, source-coverage, validate-evidence, audit-claims, verify-quotes, check-memo-quality, memo-quality-bench, weekly-delta-bench, score, bench, doctor, mcp-config
Agent readiness: AGENT_READINESS.md (delegation-readiness checklist, MCP-first / A2A-later / payment-safe / human-escalated) + CLI validate-agent-card (static agent-card lint vs schemas/v1/agent-readiness-report.schema.json; no probing, no signature verification)

Boundary:
- Source-plan coverage is diagnostic before v1.0.
- Missing `must_check` coverage should be disclosed as an evidence gap, not treated as factual truth.
- These tools do not discover sources, score source reputation, or decide whether claims are true in the world.
