## Propagation breaker for agent fleets (P47) — TextForAgents LLC
An epidemiological layer against self-replicating prompt injections (agent worms). An egress screen scores outbound agent output for instruction-like content copied from inbound context — blended shingle + token similarity, cheap-first, no inference call — quarantining verbatim forwards and holding paraphrased ones closed. Payload-fingerprint telemetry alerts when a fingerprint appears in 3 or more distinct agents, with a reproduction-number (R) estimate. Shared-store writes are data-only by default and never auto-promote to instruction context; elevation requires an HMAC token bound to key and agent. A sealed control registry detects tampering with the loop-breaking controls. Agents can query a payload fingerprint before acting on forwarded instructions.
**Fleet $19/mo per fleet** (up to 50 agents, fingerprint telemetry + outbreak alerts; first fleet outbreak scan free, one time per operator) · **Fleet Pro $49/mo per fleet** (up to 500 agents + egress quarantine + data-only write enforcement) · **Fingerprint query $0.001/query** (x402).
Docs/packet summary: propagation-breaker-agent-worms — see the product page for the full description, limits, and support (support@textforagents.com).
Product: https://textforagents.com/products/propagation-breaker-agent-worms/
Limits: cheap-first scoring with no inference call on the hot path; BORDERLINE is held closed, not delivered; a wired LangGraph framework integration ships (src/adapters/langgraph_adapter.py, contrast re-run inside LangGraph: naive 12/12 + 11/11 vs breaker 1 + 0) while other framework adapters are thin adapters over the hosted API and the product does not claim to protect a framework it has not been wired into; patentability memo LEGAL-167 on file (FILE-NARROWED); FTO screen LEGAL-168 on file (AMBER — no plausible read on the product as designed, no RED; keep-clear conditions apply); both worm reproductions are research settings, no confirmed in-the-wild outbreak yet; demand is unmeasured.
Start: POST /register {"operator": "your-name"} -> API token + breaker key (each shown once). POST /trial claims the one-time free fleet outbreak scan. POST /scan runs a fleet outbreak scan. POST /fingerprint-query {"text": "..."} is the $0.001/query x402 pre-trust query (plan/trial paths in docs). POST /rpc speaks the JSON-RPC surface (ping/egress_check/fingerprint_query/store_write/store_read/telemetry_report/controls_verify/elevation_token) with your Bearer token. POST /subscribe {"plan": "fleet"|"fleet_pro"} with a verified x402 payment starts a plan. MCP JSON-RPC at POST /mcp (tools: egress_check, fingerprint_query, store_write, store_read, telemetry_report, controls_verify).