{"name": "Conflicts check and ethical walls for multi-principal agents", "provider": "TextForAgents LLC", "support": "support@textforagents.com", "slug": "conflicts-check-ethical-walls-for-agents", "what": "Before your agent takes the job, check it against every engagement it already serves \u2014 and wall the conflicts off in memory and tools, with signed proof. Engagements register as HMAC fingerprints only; a prospective engagement returns CLEAR / CONFLICT-CONSENTABLE / CONFLICT-HARD on an Ed25519-signed receipt that also signs how much of the register it cleared against (candidates_examined). A consentable conflict converts only via both principals' signatures bound to the check receipt. A signed ethical-wall manifest partitions memory namespaces, per-engagement tool credentials, and sibling-agent exclusion lists, and a breach query answers BREACH / NO-BREACH / UNVERIFIABLE with completeness verified from lineage sequence contiguity.", "limits": "Honest limits, stated plainly: (i) wall enforcement is GRADED, not binary \u2014 enforceable at runtime only where scope is derived from the signed wall manifest (the manifest-bound adapter); where scope is caller-supplied (the native Mem0 API pattern), the wall is audit-only \u2014 a sibling crossing can succeed at runtime and is caught after the fact by the breach query grading BREACH; (ii) the breach query's completeness rests on lineage-sequence contiguity \u2014 a log with a silent deletion grades UNVERIFIABLE, never NO-BREACH, and a fleet that does not emit lineage hooks cannot be breach-queried; (iii) the lab's census and fingerprint-sufficiency evidence (Q1/Q2) was measured on synthetic fixtures, not on a real fleet's engagement history, and live conflict frequency is UNMEASURED; (iv) wall enforceability on real CrewAI/AutoGen/Mem0 deployments was NOT driven in the lab \u2014 grading is at the Mem0 API-pattern level, and no real-stack enforcement claim is made; (v) willingness to pay and who pays (operator / agent / principal) are UNMEASURED at launch. A CLEAR is a statement about the register searched, with its signed candidates_examined scope, not a guarantee that no unregistered engagement exists.", "boundary": "Stated plainly: wall enforcement is graded \u2014 runtime-blocking only where scope is derived from the signed manifest; where scope is caller-supplied (the native shared-memory pattern), crossings are caught by the breach query after the fact, not blocked; the lab's census and accuracy evidence was measured on synthetic fixtures, real CrewAI/AutoGen/Mem0 deployments were graded at API-pattern level only, and live conflict frequency and willingness to pay are unmeasured.", "prices": {"conflicts_check_usd": 0.25, "consent_pair_usd": 0.49, "wall_manifest_certificate_usd": 0.99, "breach_query_usd": 0.99, "fleet_pass_usd_per_month": 19.99, "fleet_pass_allowance_per_month": {"checks": 2000, "walls": 200, "breach_queries": 200}, "beyond_allowance": {"check_usd": 0.25, "wall_usd": 0.99, "breach_query_usd": 0.99}, "consent_pairs_in_pass": false}, "trial": "One free conflicts check per operator, ever, enforced by operator account identity (and payer wallet). After that single check, the standard prices apply. No perpetual free tier and no other free allotment.", "how_to_start": "POST /register {account_name, payer_wallet} -> per-customer token (shown once, SHA-256 at rest). POST /engagements (Bearer) registers an engagement as fingerprints only. POST /check (Bearer) runs the pre-acceptance conflicts check \u2014 the first check is the one-time trial; an active fleet pass covers checks/walls/breach queries to its stated allowance; any other priced call is x402 (X-PAYMENT header) at the sheet price. POST /consent/create, /consent/sign, /consent/verify manage the gated consent pair ($0.49 charged when a pair completes). POST /wall issues the signed manifest + certificate ($0.99); /wall/write, /wall/read, /wall/adapter-* and /wall/shared-* exercise enforcement; POST /breach-query ($0.99) grades lineage. POST /subscribe (Bearer, X-PAYMENT $19.99) activates the fleet pass. The same operations are MCP tools at POST /mcp.", "pay_to": "0x7a482F3B58780519880a6A13F983E958f2Ed6A61", "network": "base (eip155:8453)", "asset": "USDC", "signing_keys": "/.well-known/signing-keys"}