Every agent call runs in Standard mode only, every evaluation bundle is marked SYNTHETIC,
every person-facing interview or crisis decision stays strictly human-only, and every swarm task is
claimed once, finished once, and cryptographically receipted.
Find, quantify, and combat online human harm, internet scams, financial fraud, and exploitation across social media, websites, messages, documents, and payments while clearing honest look-alikes.
Donate unused AI tokens from OpenCode, Ollama Cloud, or Meta models to build, deterministically
verify, and independently review (drafter != reviewer) off-the-shelf skills, tools,
SOPs, and processes for NGOs.
Atomic work queue and compare-and-set storage for agent swarms: every task a verb tile with fencing tokens and receipts; every shared artifact a versioned noun over HTTP, MCP, CLI, and web.