vassiliylakhonin/agenda-intelligence-md
vassiliylakhonin
JavaScript8updated 29d ago
JavaScript8updated 29d agoDeterministic evidence-packet checks for AI-generated claims: references, quotes, numbers, and support gaps. CLI, Python, MCP, and review adapters.
ClaudeCodexCursorMCP
a2aai-agentsclaim-verificationevidence-validationhuman-in-the-loopjson-schemallm-evaluationmcpmcp-servermodel-context-protocolpythonsanctions-compliance
README
Agenda Intelligence MD
Product runtime and evidence-discipline layer for strategic intelligence agents. One core service layer behind four delivery surfaces — MCP server, HTTP API, A2A adapter, and a deployable Cloudflare Worker baseline — plus structured per-product contracts, geography-routed reasoning, schema validation, evidence audit, and scoring. Ships with four vertical workers: Middle Corridor Deal Risk Gate, CIS Secondary-Sanctions Exposure, Agentic Interaction Trust Gate, and Gulf Maritime Exposure Gate.
Read it as a trust-routing layer for agent-mediated decisions: it turns a partial evidence pack into a structured routing decision — allow, step up, escalate, or not-decision-ready — surfaces the specific missing evidence, and always requires human review for high-stakes actions. It does not approve, clear, or make a factual determination, and does not verify factual truth. Live retrieval is off by default and opt-in per worker profile (see the CIS worker and ADR 0014). Evaluate any worker in a few minutes with live curl calls: docs/agenstry/demo-pack.md.