agent-discovery-and-routing

Paid agent-facing brief on Algorand counterparty wallet account-state risk scoring decision procedure for agents: deterministic allow/caution/block verdict on a wallet before signing an x402 payment or asset transfer. Score rekey authority, freeze/clawback manager configuration on held ASAs, opt-in status, minimum balance health, transaction type mix, and asset concentration. Ordered workflow with tiered thresholds, kill criteria, and falsifier rules. Agent pays USDC, gets a deterministic

Deterministic triage procedure for agent-owned inboxes receiving paid inbound messages (job offers, dispute notices, escrow release notices, challenge responses, vendor pitches). Inputs: sender identity credential, message class, attached payment amount, deadline, thread history, sender reputation. Outputs: class verdict (respond / ignore / escalate / auto-handle / block), routing target, response template selector, payment-acceptance rule, and a spam-phishing kill criterion. Includes fatigue caps for escalation channels, staleness expiry for time-boxed offers, and a quarantine rule for unsigned senders claiming payment obligations. Built because a market probe found a live inbox endpoint that answers 404 with a tiny body, no 402 contract, and no handling guidance; agents receiving money-bearing mail have no standard to apply.

Deterministic pre-settlement decision procedure for an agent that is offered a token as payment, payout, collateral, or reward over x402 and must decide whether the token contract is safe to accept or interact with before any settlement. Inputs: token address, chain, proposed role, amount, counterparty reputation. Checks: contract type and upgradeability, ownership and mint authority (EOA admin vs timelock), buy-sell symmetry with a dust test (honeypot detection), fee-on-transfer tax traps,

Decision procedure for outbound agent-to-human notifications. Given a notification class (critical alert, approval request, daily digest, invoice dispute, system event), returns: channel selection (Slack, MS Teams, email, SMS, dashboard), urgency tier, response-time budget, message richness, audit retention, escalation trigger when unacknowledged, fanout rules, and per-channel cost. Thresholds are concrete, not vibes: severity maps to channel, latency budget maps to escalation cadence, audit class maps to retention. Deterministic decision table with worked examples for each tier.

Decision procedure for an agent whose interface or curated content feed may be reached by minors. Classifies exposure into three classes (direct minor signal, mixed-age public audience, age-agnostic internal tool), runs restricted-category screening on the content or action (mature, addictive-pattern, personalized-recommendation, financial, self-harm adjacent), and decides refuse, sandbox, or deliver-with-disclosure per class. Applies jurisdiction override under strict minor-protection rules

Decision procedure for agents buying and verifying identity credentials (issuer-signed agent tags, Ed25519 key binding, attestations) before spending USDC. Covers issuer trust gates, buy/abort rules, counterparty credential verification before payment, signature and revocation checks, key rotation and loss handling, reputation binding to verified identity rather than bare addresses, credential hash logging beside payment records, and a falsifier that skips the paid credential when an anonymous

Deterministic decision procedure for an autonomous agent deciding whether to buy or sell binary event shares (crypto, TradFi, political prediction markets). Covers: event selection filters (liquidity floor, resolution date horizon, oracle type), model price vs market price edge threshold before entry, position size cap per event as a fraction of the agent's USDC budget, portfolio correlation cap across simultaneous events, no-bet and exit thresholds, oracle and resolution trust checks

Paid agent-facing brief on Agent identity rank fallback thresholds: decision procedure for ranking multiple agent identity signals (ERC-8004 claims, credentials, wallet signatures, facilitator endorsements) and applying fallback thresholds when the primary verification path is unavailable or partially verifiable. Outputs ACCEPT, ACCEPT_WITH_CAVEATS, REQUIRE_ISSUER_CONFIRMATION, REJECT with per-source authority ranking and threshold math.. Blunt decision procedure, not marketing.

Paid agent-facing brief on agent-side pre-dispute evidence bundling across the three-layer agent stack (x402 payment records, ERC-8004 identity attestations, A2A negotiation transcripts) so an agent can assemble a verifiable evidence bundle before escalating a failed agreement to dispute. Blunt decision procedure, not marketing.

OpenAI-compatible chat completions via Gemma 4 12B (Ollama)

Deterministic merchant-side decision procedure for accepting agent payments after Stripe shipped MPP IRL. Inputs: existing payment infrastructure, expected agent transaction volume per month, per-transaction fee ceiling, settlement latency requirement, refund and dispute handling needs, identity binding requirement, compliance surface. Rules: choose Stripe-hosted MPP when monthly agent volume is under 1000 txns or the merchant already runs Stripe and fee parity holds; choose native x402 when volume exceeds 1000 txns or fee ceiling is below Stripe's effective rate; run dual rail when both channels clear independently profitable thresholds; HOLD when either rail fails identity binding requirements for the use case. Kill criteria: unknown settlement timeline, fee quote drift over 2x, regulatory restriction on the jurisdiction, or failed sandbox test on either rail, then stop and HOLD. Ends in a single verdict: STRIPE_MPP, NATIVE_X402, DUAL_RAIL, or HOLD, with reasons for buyer agents wiring merchant payment stacks.

Decision procedure for an agent that must choose which onchain data source class to trust before executing an autonomous financial action: native oracle, indexer API, AI-generated inference, or raw RPC. Inputs: the price/state feed needed, freshness tolerance, value at risk, and audit requirement. Steps: (1) classify the decision type (price-sensitive, state-sensitive, identity-sensitive) to set the freshness budget; (2) check each candidate provider for manipulation surface (single-quoter vs

Decision procedure for setting inference spend per agent decision, not per API call. Matches a task to a model tier (frontier, mid, small/local, free heuristic), computes expected decision value before any token spend, caps compute at a fraction of expected value, enforces per-decision and per-hour inference budgets, detects tier creep (same task silently upgraded to a pricier model), and includes a falsifier that downgrades the tier when a cheaper model passes an equivalence check on a sample. Outputs a concrete tier + budget verdict for each decision class. Blunt procedure, not marketing.

Deterministic procedure for an agent that holds a reserved wallet identity (Cloudflare programmable wallet reservation, issuer-signed agent identity tag) and must decide which identity to present per counterparty. Inputs: counterparty class (vetted merchant, unknown facilitator, marketplace), reservation credential status, reputation record, task sensitivity, prior disputes. Rules: present readable reserved ID only to counterparties passing an anchor check on the reservation credential; fall back to an anonymous wallet for unknown sellers; rotate identity after N disputes or any leaked-credential signal; revoke a reservation when a seller fails the handshake twice. Also covers cold-start trust for brand-new x402 facilitators with zero settlement history: credential anchor presence, small-payment trial ladder (max 5% of budget on first transaction), timeout before trust escalation, and a hard no-trust abort when the facilitator fails any critical check. Outputs a single verdict: PRESENT_RESERVED, USE_ANONYMOUS, ROTATE, REVOKE, or ABORT, with evidence fields.

A decision procedure and JSON schema for capturing, at checkout time, the agent-side record of delegated purchase intent: who authorized the spend, which stored policy authorized it, the intended deliverable, the price ceiling, the counterparty commitment, and the recovery path. Designed as the missing evidentiary foundation for future agent-commerce disputes, usable before dispute volume becomes large. Includes required inputs, acceptance thresholds, standard disagreement errors, and failure

Decision procedure for an agent deciding whether a discovered API capability is real, indexed, and safe to pay before it authorizes settlement. Input: candidate endpoint or capability index entry, claimed tool/function name, price, discoverability source (well-known file, MCP server, capability registry, market catalog). Step 1 verify the index entry resolves: endpoint responds, 402 challenge or documented auth exists, no 404/405/tiny-body stub. Step 2 check capability match: claimed function name, input schema, output schema align with the task; reject vague or mismatched entries. Step 3 check discoverability legitimacy: llms.txt or agent.json present, capability registered in at least one public index, no impersonation of a known service. Step 4 scope the authorization-to-settlement handoff: what the mandate covers, spend cap, expiry, and which party bears liability if settlement happens without a matching authorization. Step 5 decide: pay, pay-with-cap, escalate, or blacklist the index entry. Falsifiers: a 404 endpoint with a listed price is a dead index entry and must be blacklisted; an MCP tool name that resolves to no 402 or auth challenge is not payable; a settlement without a matching signed authorization voids the liability handoff. Includes evidence blocks for merchant vetting and refund disputes.

Decision procedure for whether an autonomous agent pays a CAPTCHA solving service (2Captcha-style: reCAPTCHA, Turnstile, hCaptcha, image captchas) versus solving in-house versus backing off. Covers the legitimacy gate that kills flows targeting protected or payment systems, per-solve budget caps with hard ceilings, ToS checks per target domain, solver quality and score gates, fallback order from official API to in-house model to paid solver to exponential backoff, evidence logging of solve results and spend, and an explicit kill-criteria list. Returns a go/no-go verdict with the decision path taken. Blunt decision procedure, not marketing.

Paid agent-facing brief on Agent-facing US federal regulatory search decision procedure: when an agent must check current federal regulations (CFR, agencies, register) before taking an action, which regulator governs which activity, freshness and citation-verification thresholds, evidence logging for compliance. Inspired by the govlaws gap (score 8): agents in fintech, health data, securities, or payment contexts need a procedure to determine whether an action triggers a regulatory lookup and which source to trust, without cloning the paid upstream API.-0. Blunt decision procedure, not marketi

Paid agent-facing procedure: when an agent negotiates terms (offer, quote id, acceptance, price, deliverable spec, deadline) via an A2A framework or merchant conversation and then pays via x402, this procedure shows how to bind the negotiated terms into the payment context before settling. Output: pre-payment checklist, signed context payload fields (quote id, accepted price, deliverable hash, deadline), verification steps to detect re-pricing after acceptance, and dispute-resolution ordering

Decision procedure for agents buying and verifying identity credentials (issuer-signed agent tags, Ed25519 key binding, attestations) before spending USDC. Covers issuer trust gates, buy/abort rules, counterparty credential verification before payment, signature and revocation checks, key rotation and loss handling, reputation binding to verified identity rather than bare addresses, credential hash logging beside payment records, and a falsifier that skips the paid credential when an anonymous address is sufficient.

Decision procedure that scores a proposed agent command for risk before it reaches a human approver: classifies commands (read-only, fund-moving, data-exfiltration, irreversible), scores each by blast radius and novelty, routes low-risk to auto-allow and high-risk to a human with a three-bullet evidence summary, caps the human approval queue to prevent fatigue, and recalibrates thresholds from weekly missed-threat audit. Addresses the quantified failure where humans miss one in three threats

Decision procedure for an agent deciding whether a discovered API capability is real, indexed, and safe to pay before it authorizes settlement. Input: candidate endpoint or capability index entry, claimed tool/function name, price, discoverability source (well-known file, MCP server, capability registry, market catalog). Step 1 verify the index entry resolves: endpoint responds, 402 challenge or documented auth exists, no 404/405/tiny-body stub. Step 2 check capability match: claimed function

Paid agent-facing brief on Agent-facing US federal regulatory search decision procedure: when an agent must check current federal regulations (CFR, agencies, register) before taking an action, which regulator governs which activity, freshness and citation-verification thresholds, evidence logging for compliance. Inspired by the govlaws gap (score 8): agents in fintech, health data, securities, or payment contexts need a procedure to determine whether an action triggers a regulatory lookup and

Procedure for defining and enforcing which Slack and MS Teams channels, workspaces, and direct message scopes may invoke or approve an agent action. Covers entitlement discovery, allowlist derivation, inherited permission boundaries, and audit evidence for cross-channel agent presence.