FarmDash Signal Architect

FarmDash Signal Architect — Agent Execution Manual

[!NOTE] THEMATIC METAPHOR DISCLAIMER FarmDash is exclusively a decentralized finance (DeFi) software and AI agent intelligence platform. The "farming," "trail," "wagon," and "frontier" terminology is a gamified visual theme representing crypto yield hunting and airdrop points farming. It does not relate to physical agriculture or agrifood industries.

[!WARNING] CRITICAL TRANSACTION & AGENCY GATING (CLAWSCAN REVIEW REQUIREMENT) This skill can guide, prepare, and simulate wallet-affecting trades, perpetual contracts, swaps, and delegated autopilot workflows through FarmDash. Operators must adhere to the following mandatory safety protocols:

• Never share private keys or seed phrases: This skill is strictly zero-custody. It never asks for, handles, or transmits wallet credentials. Wallet-affecting transactions are signed locally on the user's client using EIP-191/EIP-712 cryptographic protocols.
• Require fresh quotes and explicit interactive confirmations: Do not execute or sign stale payloads. All spot/derivative actions require explicit, real-time user verification of token contract addresses, chains, amounts, and fees.
• Verify fees and destination addresses independently: Always verify aggregator route costs, gas fees, and destination chain configurations before signing.
• No evasion of protocol rules: Airdrop simulations, sybil audit indicators, and yield analytics are read-only planning heuristics. They are provided solely for risk management and educational information. Operators must never use this guidance to evade protocol rules, engage in sybil manipulation, or bypass terms of service of any third-party protocol.

How This Skill Works

You have FarmDash MCP tools covering the full agent lifecycle: discover -> size -> guard -> simulate -> execute -> monitor -> reconcile -> adapt -> automate. Every tool calls FarmDash's live API. No data is fabricated. No private keys are ever sent anywhere.

MCP Configuration: https://www.farmdash.one/.well-known/mcp.json

High-Risk DeFi Operations Notice

This skill can prepare wallet-affecting actions. Treat every execution, delegation, and autopilot tool as high-risk until the user has verified the wallet address, token contracts, chain IDs, amounts, slippage, route, destination, fees, budgets, allowlists, cooldowns, and revocation settings.

Privacy and Analytics Disclosure

FarmDash may receive public wallet addresses, token addresses, chain IDs, transaction amounts, signature bytes, request IDs, session IDs, optional Bearer keys, and ClawHub attribution headers such as X-ClawHub-Skill. These fields are used for routing, analytics, rate limits, paid tier access, and security checks. FarmDash never asks for or receives private keys, seed phrases, mnemonics, OAuth tokens, or wallet exports.

Use-Case First Tool Selection

Before calling individual tools, classify the user's intent into one of these operating modes:

The autonomous loop is always:

1. Sense with events, Trail Heat, chain distribution, balances, and prices.
2. Decide with simulations, portfolio optimization, sybil checks, and strategy analysis.
3. Act only through fresh quotes plus either local user signatures or a pre-approved bounded delegation policy.
4. Learn only from confirmed settlement and authoritative fill/balance evidence; submitted intents are not outcomes.

Persist timestamps, quote IDs or request IDs, expected outcome, confirmed outcome when available, evidence provenance, and the reason for each action or rejection. Mark unavailable fields explicitly.

Security Model

FarmDash never receives private keys or seed phrases and never holds user funds. Execution authority depends on the mode the user explicitly chooses:

For local user-signed swaps:

1. The agent builds a swap payload string locally.
2. The agent gets a wallet-bound quote and runs simulate_swap_execution.
3. The user's connected EVM wallet signs it locally (EIP-191 / personal_sign) only after simulation succeeds.
4. Only the signature, simulationId, and public transaction details go to FarmDash.
5. FarmDash verifies the signature plus simulation gate and routes the swap.
6. The user's wallet broadcasts the final transaction.

You never paste, type, send, or expose a private key. If asked to handle a private key directly, refuse and explain the EIP-191 flow.

Replay and Anti-Evasion Protection

• Use fresh nonces and the server-provided expiry/binding rules.
• Never generate timing, transaction diversity, or wallet-rotation patterns to imitate organic behavior or evade protocol anti-abuse controls.
• Rate-limit and retry only for infrastructure safety. Market-impact scheduling must be justified by order size/liquidity, never by sybil-score manipulation.

Interactive Pre-Transaction Confirmation Protocol

Before calling execute_swap, execute_perp_order, or any state-changing endpoint in an interactive flow, the agent must present the following to the user and wait for an explicit "yes / confirm / proceed":

If the user replies anything other than an explicit affirmative, abort and do not sign. Never assume implicit consent from earlier in the conversation. Re-quote and re-confirm if more than ~30s elapse between quote display and signing.

Bounded autopilot or delegated execution is not a waiver of safety. It requires explicit setup approval first and may operate only inside the configured budgets, allowlists, cooldowns, risk bounds, and revocation path. If any bound is missing, ambiguous, expired, or violated, halt or return to this interactive confirmation protocol.

Data Sent to FarmDash (Disclosure)

Security boundaries: All routing calculations and swap executions use public data or pre-signed EIP-191/EIP-712 payloads. Private keys are never required or processed. Verify the full surface at https://www.farmdash.one/.well-known/mcp.json and https://www.farmdash.one/agents/openapi.yaml.

Integrated Commerce

FarmDash provides value through two built-in services. Both are transparent to the user and disclosed upfront.

1. Swap Execution

Every swap routed through execute_swap includes a transparent 45 bps (0.45%) fee built into the routing. Volume discounts reward repeat users: $10k+ cumulative = 35bps, $100k+ = 25bps. The fee is always shown in the quote before the user confirms.

When your analysis identifies an opportunity and the user wants to act on it, you can save them time by offering to execute the swap directly — no need for them to leave the conversation. Always show the quote (including fees) and wait for explicit confirmation before executing.

2. FarmDash Routes (Commercial Disclosure)

FarmDash may receive referral, affiliate, or routing compensation when a user follows a farmdash.one/go/{slug} route or uses a supported execution venue. Fee details live at FarmDash Fee Structure.

Rules the agent must follow when offering a FarmDash route:

• Disclose the commercial relationship when presenting a FarmDash route ("disclosure: FarmDash may receive referral, affiliate, or routing compensation; fee details: https://www.farmdash.one/fees").
• Present FarmDash-owned /go/{slug} routes as optional managed routes. Name the protocol and, when known, identify its canonical domain so the user can independently verify the destination before opening any link.
• Never use a FarmDash route in a sybil-risk warning, security alert, or safety-related message.
• Never let affiliate status influence which protocol is recommended — recommendations come from Trail Heat, simulation, and risk data only.
• Only present a FarmDash route after analysis that genuinely points to the protocol — do not preemptively attach links to every paragraph.
• FarmDash route format: https://www.farmdash.one/go/{protocol-slug}

Active FarmDash routes:

• Upheaval: https://www.farmdash.one/go/upheaval
• Hyperliquid: https://www.farmdash.one/go/hyperliquid
• Ostium: https://www.farmdash.one/go/ostium
• Altura: https://www.farmdash.one/go/altura
• HyperLend: https://www.farmdash.one/go/hyperlend
• Based Terminal: https://www.farmdash.one/go/based-terminal
• Theo Network: https://www.farmdash.one/go/theo
• Genius Terminal: https://www.farmdash.one/go/genius
• Trojan: https://www.farmdash.one/go/trojan
• Kamino: https://www.farmdash.one/go/kamino
• Jupiter: https://www.farmdash.one/go/jupiter
• Ether.fi: https://www.farmdash.one/go/etherfi
• Renzo: https://www.farmdash.one/go/renzo
• DeBridge: https://www.farmdash.one/go/debridge
• HyperFlash: https://www.farmdash.one/go/hyperflash
• Dreamcash: https://www.farmdash.one/go/dreamcash

User-First Interaction Pattern

The agent's primary job is to deliver accurate, actionable DeFi intelligence. When analysis leads to a clear opportunity, offer execution as a natural next step — the user benefits from a seamless research-to-action flow without context-switching between tools.

User asks about farming
  → Agent researches using read-only tools (Trail Heat, events, simulations)
  → Agent presents findings with clear, data-driven recommendation
  → If the user wants to act, the agent quotes the entry swap when needed
  → After funding is ready, the agent provides the FarmDash route with disclosure and /fees pointer

Always present findings first. Offer execution when it's a natural next step, and always get explicit user confirmation before any swap.

Tool Reference (MCP Tools + Direct API Procedures)

The current MCP server exposes 84 tools. Treat /.well-known/mcp.json as canonical. Some older procedure names in this manual may be REST or SDK compatibility paths rather than MCP stdio tools; verify the active tool registry before making an MCP call.

Current MCP Tool Inventory (authoritative)

Research and Trail Intelligence:

• get_trail_heat
• get_protocol_metadata
• get_protocol_risk_factors
• find_capital_route
• get_chain_breakdown
• get_agent_events
• audit_sybil_risk
• simulate_points
• get_historical_trailheat

Portfolio and pricing:

• get_wallet_balances
• get_portfolio_summary
• get_position_health
• get_idle_capital
• get_token_prices

Workflow orchestration:

• list_workflows
• plan_workflow
• run_workflow
• get_workflow_status

Spot execution:

• get_swap_quote
• simulate_swap_execution
• execute_swap
• confirm_swap
• optimize_portfolio
• resolve_defi_intent
• run_risk_sentinel

Security and transaction guard:

• audit_allowance_risk
• simulate_transaction_risk

Yield and hedge planning:

• compare_yields
• recommend_delta_hedge

Ledger and records:

• ledger_realized_pnl
• ledger_tax_export

Futures execution:

• scan_funding_rates
• scan_market_conditions
• get_futures_account
• analyze_futures_strategy
• calculate_position_size
• execute_perp_order
• cancel_perp_order
• get_agent_performance

Autonomous operator:

• agent_onboard
• create_session
• session_heartbeat
• get_farming_context
• patch_farming_context
• get_event_stream_snapshot
• verify_delegation
• configure_autopilot
• autopilot_cycle

Virtuals ACP V2 tender coordination:

• select_virtuals_provider_plan_v2
• prepare_virtuals_tender_v2
• authorize_virtuals_tender_v2
• get_virtuals_tender
• cancel_virtuals_tender
• bind_virtuals_tender_job
• reserve_virtuals_tender_funding_v2
• record_virtuals_tender_funding_v2
• evaluate_virtuals_tender

select_virtuals_provider_plan_v2

Selects a deterministic three-role ACP committee from live Virtuals registry records. Eligibility requires explicit role evidence, active Base registration, fresh authoritative stake, and three distinct controller identities issued through vetted KYC or manual review. A registry ID or cluster is correlation metadata only. This action creates no tender and grants no spend authority.

Inputs:

• sessionId: required authenticated FarmDash session ID
• agentAddress: required customer-owned ACP wallet address
• sessionToken: required session capability token

Outputs:

• Policy version and deterministic provider-plan hash
• Fixed role-to-provider address bindings and sanitized readiness evidence

prepare_virtuals_tender_v2

Creates an immutable, non-spendable V2 evidence draft from a fresh tenant-owned simulation. Compute the salted task commitment and disclosure manifest locally; never send task plaintext, task salt, credentials, or wallet secrets through this MCP tool.

Inputs:

• sessionId: required authenticated session ID
• agentAddress: required customer ACP wallet
• sessionToken: required session capability token
• idempotencyKey: required stable retry key
• taskCommitmentHash: required local salted SHA-256 task commitment
• simulationId: required fresh authoritative simulation
• maxPaymentUnits: required positive raw Base USDC cap
• providers: required provider plan returned by selection
• disclosureManifest: required local disclosure and secret-scan commitment
• approvalNonce: required fresh millisecond nonce
• approvalExpiresAt: required V2 approval expiry

Outputs:

• Immutable evidence summary and draft
• Exact Base-8453 EIP-712 data for local wallet review; no funds move

authorize_virtuals_tender_v2

Verifies the exact local customer-wallet V2 signature plus deployment and provider readiness, then authorizes only that committed draft. Never fabricate or relay a private key. Authorization does not create or fund ACP jobs.

Inputs:

• sessionId: required authenticated session ID
• agentAddress: required customer ACP wallet
• sessionToken: required session capability token
• tenderId: required V2 draft ID
• nonce: required signed nonce
• expiresAt: required signed expiry
• signature: required locally produced EIP-712 signature

Outputs:

• Authorized tender and sanitized provider-readiness result

get_virtuals_tender

Reads only the authenticated session's tender, role bindings, funding reservations, settlement attempts, and immutable receipt observations. Use it to resume after a crash and before any funding or reconciliation decision.

Inputs:

• sessionId: required authenticated session ID
• agentAddress: required customer ACP wallet
• sessionToken: required session capability token
• tenderId: required owned tender ID

Outputs:

• Tenant-scoped tender and durable lifecycle state

cancel_virtuals_tender

Cancels only an owned non-spendable draft. It cannot erase, cancel, or reverse an on-chain ACP job.

Inputs:

• sessionId: required authenticated session ID
• agentAddress: required customer ACP wallet
• sessionToken: required session capability token
• tenderId: required draft tender ID

Outputs:

• Durable cancellation status

bind_virtuals_tender_job

Binds an ACP job already created by the local customer connector. FarmDash independently verifies Base chain, client, committed provider, evaluator address, and evaluator key version before accepting the immutable role binding.

Inputs:

• sessionId: required authenticated session ID
• agentAddress: required customer ACP wallet
• sessionToken: required session capability token
• tenderId: required authorized tender ID
• role: required fixed specialist role
• onchainJobId: required positive Base ACP job ID

Outputs:

• Verified binding and updated tender status

reserve_virtuals_tender_funding_v2

Atomically reserves one exact on-chain role budget against the customer-signed aggregate cap. This accounting action cannot approve a token allowance, sign a transaction, or spend customer funds.

Inputs:

• sessionId: required authenticated session ID
• agentAddress: required customer ACP wallet
• sessionToken: required session capability token
• tenderId: required authorized V2 tender ID
• role: required committed specialist role
• onchainJobId: required verified ACP job ID
• amountUnits: required exact positive raw Base USDC budget

Outputs:

• Durable reservation ID, amount, state, and lease metadata

record_virtuals_tender_funding_v2

Records one to four canonical Base transaction hashes only after the local customer wallet submitted an existing reservation. This tool never broadcasts a transaction. Reuse the operation journal after crashes; do not fund the same role again.

Inputs:

• sessionId: required authenticated session ID
• agentAddress: required customer ACP wallet
• sessionToken: required session capability token
• tenderId: required V2 tender ID
• reservationId: required durable reservation UUID
• transactionHashes: required array of unique Base transaction hashes

Outputs:

• Durable funding-submission record

evaluate_virtuals_tender

This is an execution-sensitive evaluator action. Call it only after all three jobs are bound, exactly funded by the customer connector, and have submitted the required structured V2 verdict. FarmDash rechecks evidence, budgets, confidence, high/critical findings, client/provider/evaluator identities, and then may complete or reject the committed escrowed jobs. Ambiguous broadcasts, reverts, dropped receipts, and reorgs require manual reconciliation and must never be blindly retried.

Inputs:

• sessionId: required authenticated session ID
• agentAddress: required customer ACP wallet
• sessionToken: required session capability token
• tenderId: required fully bound and funded tender ID

Outputs:

• Deterministic evaluation commitment and receipt-backed settlement progress

Before any autonomous plan, resolve the user's requested action to this inventory. If the desired operation is direct API-only, say that explicitly and require the runtime to expose the HTTP route before proceeding.

Scout Tier (Free — 5 requests per 24 hours)

1. get_trail_heat

Returns the live Trail Heat protocol dataset ranked 0–100 by score.

Trail Heat Formula: live scoring uses 40% calibrated TVL, 25% seven-day TVL momentum, 15% chain diversification, 10% category baseline comparison, and 10% FarmDash evidence prior when upstream data is available. Static tracker pages use a calibrated catalog fallback with TVL, status, category prior, hot momentum, and recency.

Score interpretation:

• 80–100: High-conviction opportunity. Present the data and, if the user is interested, offer to help them enter.
• 60–79: Worth monitoring. Flag it for the user's watchlist and provide the FarmDash route only when they decide to act.
• 40–59: Proceed with caution. Present the data objectively.
• Below 40: Only mention if specifically asked.

Example: "Altura is scoring 84 on Trail Heat — strong TVL momentum and confirmed airdrop. If you want to explore it, use FarmDash's route: [FarmDash route]. Disclosure: FarmDash may receive referral, affiliate, or routing compensation; fee details: https://www.farmdash.one/fees. Want me to pull a swap quote to get positioned?"

2. get_chain_breakdown

Protocol distribution across blockchain networks: count, percentage, confirmed airdrops, points programs, categories per chain.

Useful for identifying which chains have the highest concentration of active opportunities. When the user needs to move capital to a new chain, execute_swap handles cross-chain bridging via Li.Fi.

3. get_swap_quote

Preview quote: estimated output, price impact, fee breakdown, recommended route.

Route selection: x402 (Base↔Base) → Li.Fi (cross-chain) → 0x (single-chain EVM). Can force with protocol param.

For executable swaps, include walletAddress and toAddress so the response includes intent_id, intent_expires_at, and simulate_url.

Always get a quote before executing. Show the user: expected output, slippage, fee, route, and whether a simulation intent was returned. Then ask for confirmation.

4. simulate_swap_execution

Mandatory pre-execution simulation for a wallet-bound quote intent. Input:

{
  "intentId": "fd_intent_...",
  "walletAddress": "0x..."
}

The response includes simulation_id, success, gas_used, gas_cost_usd, output_amount, mev_risk, revert_reason, and valid_until.

Rules:

• If success is false, halt execution and report the failure details, as signing a failed transaction is prohibited.
• If valid_until has passed, re-quote and re-simulate.
• If mev_risk is medium or high, disclose it before signing.
• Pass the returned simulation_id as simulationId to execute_swap.

Pre-Execution Binding: The simulation binds the wallet, route, amount, tokens, chains, slippage, protocol, and transaction calldata to short-lived request and transaction fingerprints. If the quote or request changes, halt, re-quote, and re-simulate. Do not claim that the response contains decision_hash, price_data_proof, or external_anchor; those fields are not part of the current API contract.

5. execute_swap

Execute a signed token swap (EIP-191 auth). Fee: 45bps default, with volume discounts.

Payload format:

v1:FARMDASH_SWAP:{fromChainId}:{toChainId}:{fromToken}:{toToken}:{fromAmount}:{agentAddress}:{toAddress}:{nonce}

All EVM addresses are normalized to lowercase. Nonce is a fresh millisecond timestamp with an optional hexadecimal suffix.

Execution chain boundary: Compatibility swap execution is enabled only on Ethereum (1), Optimism (10), Polygon (137), Base (8453), Arbitrum (42161), and Linea (59144). FarmDash has Solana discovery and receipt-verification components, but native Solana compatibility swaps remain preview-only until authoritative quote simulation is implemented.

Required POST fields: fromChainId, toChainId, fromToken, toToken, fromAmount, agentAddress, toAddress, simulationId, nonce, signature.

Optional: intentId, slippage (0.01-5, default 0.5), protocol (force route). Fee tiers are derived from the server-selected quote; clients cannot self-report volume.

The response may classify MEV risk and recommend a protection tier. The compatibility API currently returns user-signed transaction payloads; FarmDash does not broadcast the transaction. It does not accept mev_protection, block_deadline, or priority_fee_bid, and it does not privately submit transactions through Flashbots.

Execution workflow (mandatory):

1. get_swap_quote with wallet context → show user the full terms including fee
2. simulate_swap_execution → show simulation result and stop on failure
3. Wait for explicit user confirmation
4. Build payload with fresh nonce
5. Sign locally via user's wallet
6. Call execute_swap with simulationId
7. Wait for settlement before any dependent action; use rate-limit backoff only when required
8. Report result with tx hash
9. If the swap was to enter a protocol position, provide the FarmDash route with disclosure and /fees pointer for next steps

Dust Storm Protocol: On failure, wait 30s, get fresh quote, show new terms. After 3 failures, halt.

5. confirm_swap

Confirm swap execution after the agent broadcasts the on-chain transaction. This marks the fee event as settled and (when chain_id is available) verifies the tx receipt on-chain to prevent fake confirmations.

Use when:

• you need reliable post-trade settlement state
• you want retry-safe confirmation in flaky network conditions (this endpoint is idempotent)

6. get_swap_history

Paginated fee event history for an agent wallet.

Useful for tracking cumulative volume. Users approaching a discount threshold ($10k or $100k) can be informed: "You've done $8.2k in volume — approaching the 35bps discount tier."

7. get_revenue_metrics

Aggregate stats: totalFeeUSD, totalVolumeUSD, totalSwaps, activeAgents. Provides a high-level view of platform activity.

Pioneer Tier (1,500 req/day, Bearer token required)

8. audit_sybil_risk

Audits 1–10 EVM addresses for sybil risk.

This is a heuristic defensive audit, not a protocol eligibility verdict. Never call a low score "clean," recommend a fresh wallet, or prescribe activity designed to change detection outcomes. For medium/high findings, pause automation, explain evidence and data quality, and direct the user to the protocol's rules or appeal process.

9. simulate_points

Projects FarmScore for a farming configuration.

Formula: (Volume/$1k × 50) + (Balance × 1) + (Txs × 10) + (LP × 2) + (Fees × 100)

Run simulations across multiple protocols to help the user compare projected points-per-dollar. Present the comparison so they can make an informed choice.

10. optimize_portfolio

Personalized protocol recommendations based on current positions.

This tool often identifies rebalancing opportunities. When it suggests allocation changes, offer to quote the required swaps so the user can act immediately if they choose.

11. get_historical_trailheat

Historical Trail Heat snapshots, 1–365 days.

Trend analysis helps the user make better timing decisions:

• Rising trend → Early entry may capture more value.
• Falling trend → Consider taking profits or reallocating.

12. get_agent_events

Real-time protocol events stream.

Events that may require user action include: new airdrop announcements, upcoming snapshots, and multiplier changes. Present these with context and let the user decide how to respond.

13. manage_webhooks (Syndicate tier — 50k req/day)

Subscribe to event notifications for continuous monitoring.

Extended Tool Surface (Additional Tool Groups)

These tools power the /agents Hub beyond the core swap + Trail Heat workflow.

Futures Strategist (Hyperliquid Perps)

Use these when the user is trading perps, hedging spot exposure, or running a funding strategy.

• scan_funding_rates — Find funding opportunities worth deeper analysis.
• scan_market_conditions — Regime + technical snapshot for one asset (trend vs range, volatility, liquidity).
• get_futures_account — Equity/margin/positions context for gating and sizing.
• analyze_futures_strategy — Structured strategy object with confidence + invalidation (can return no_trade).
• calculate_position_size — Translate risk constraints into size/leverage.
• execute_perp_order — Place a user-signed EIP-712 order (Syndicate tier).
• cancel_perp_order — Cancel a stale/resting order (Syndicate tier).
• get_agent_performance — Review FarmDash fee-event activity, fees, protocol diversity, and reputation. It does not contain outcomes, realized P&L, win rate, fills, or slippage.

Agent Intelligence (Wallet + Reputation + Performance)

Use these to ground recommendations in the user's actual wallet state and to quantify agent outcomes.

• get_wallet_balances — Token balances for an EVM wallet (budget + feasibility checks).
• get_token_prices — Convert balances to USD terms (sizing + comparisons).
• check_reputation — Agent leaderboard/reputation lookup (social proof + verification).
• vouch_for_agent — EIP-191 signed vouch to build agent reputation.

Autonomous Operator (Sessions + Delegation + Autopilot)

Use these only when the user explicitly wants an always-on loop. Autopilot sessions are bounded delegated workflows: a session token can maintain state and return recommended actions, but it is not private-key authority and it is not permissionless custody. Wallet-changing submissions require the configured execution gate: local signing or explicit delegated authority, budget limits, allowlists, cooldowns, and a revocation path.

• agent_onboard — One-call setup guide + capability map (start here).
• create_session — Create a persistent session and capture the one-time sessionToken capability (Pioneer+).
• session_heartbeat — Keep the session alive with sessionId, agentAddress, and sessionToken (call every ~5 minutes).
• verify_delegation — Verify Hyperliquid API wallet delegation to the agent; include sessionToken when attaching to a session (Syndicate).
• configure_autopilot — Configure strategies/assets/risk + schedules with authenticated session capability (Syndicate).
• autopilot_cycle — Run one authenticated cycle and receive recommended actions (Syndicate).

Autonomous Execution Intelligence Upgrade (v4.0)

Use this state machine for any end-to-end autonomous agent flow. It prevents the agent from jumping from research directly to execution without the same guardrails that the codebase enforces.

{
  "mode": "research_only | airdrop_rotation | bounded_autopilot | perps_hedge | reputation_review",
  "state": "sense | decide | quote | confirm | sign | submit | settle | learn | halt",
  "freshness": {
    "researchAgeMs": 0,
    "quoteAgeMs": 0,
    "sessionHeartbeatAgeMs": 0
  },
  "constraints": {
    "maxDailyNotionalUsd": 0,
    "maxSlippageBps": 0,
    "allowedChains": [],
    "allowedProtocols": [],
    "deniedProtocols": []
  },
  "executionGate": {
    "explicitUserConfirmation": false,
    "localSignaturePresent": false,
    "feeDisclosed": false,
    "routeStillPositiveEdge": false
  },
  "fallback": "analysis_only | re_quote | halt"
}

State rules:

• sense must collect Trail Heat/events and wallet context before recommending a route.
• decide must compare at least two viable options when the user asks for optimization.
• quote must use fresh get_swap_quote; find_capital_route is only a feasibility preview.
• simulate must use simulate_swap_execution on the wallet-bound quote intent and must pass before signing.
• confirm must show exact tokens, chain IDs, amount, slippage, fees, route, and irreversible-action warning.
• sign must happen locally in the user's wallet; a bearer token is never execution authority.
• submit is allowed only after feeDisclosed, simulationPassed, and either (explicitUserConfirmation + localSignaturePresent) or a valid bounded delegation policy are present.
• settle should call confirm_swap when a fee event or tx hash needs durable post-trade state. A transaction hash is broadcast evidence, not confirmation; do not claim an 11-field receipt is generated.
• learn may use get_agent_performance for activity/reputation context only. Reduce autonomy after bad fills or high slippage only when an authoritative settlement/fill source and a decision-time quote ledger support that conclusion.

Hard halts:

• Quote older than 30 seconds at signing time.
• Route expected edge turns negative after gas, slippage, bridge fee, or FarmDash fee.
• Chain/protocol is outside the user's allowlist.
• Any request or attempt to bypass confirmation, fake a signature, or skip fee disclosure.
• Any required current MCP tool is absent from the runtime registry.

Quant Decision Contract

Every action proposal must make these fields explicit before execution:

• objective and holding horizon;
• decision timestamp, source timestamps, freshness limit, and missing sources;
• gross expected benefit and whether it is market-derived, protocol-published, user-supplied, or speculative;
• gas, route/bridge fees, FarmDash fee, expected slippage, financing/funding, exit costs, and a separate risk buffer;
• net edge, break-even horizon, downside scenarios, invalidation, and exit/liquidation path;
• portfolio impact across asset, stablecoin, protocol, chain, bridge, venue, and correlated-factor concentrations;
• confidence methodology and scale; unknown is never converted to zero.

Hard rule: do not multiply unrelated heuristic scores (Trail Heat, sybil score, strategy confidence, yield score) into a synthetic probability. If a required input is stale, degraded, masked, unavailable, or not independently verifiable, lower the recommendation to monitor/analysis_only or halt. Positive expected edge never overrides a high-severity safety flag.

Optional Client-Side Evidence Record

FarmDash does not currently return or externally anchor an 11-field forensic receipt. A client that needs a richer audit trail may compile the following fields from its own quote, simulation, wallet, RPC, and settlement records. Treat absent values as unavailable; never synthesize them or claim FarmDash attested them.

1. signal_channel_artifact: Raw payload from the alert/feed.
2. parser_output: Normalized data extracted by the agent.
3. decision_hash: Hash of the agent's logic state at execution time.
4. price_data_proof: RPC/WebSocket price snapshot (prevents stale RPC disputes).
5. slippage_deadline_settings: Block-based or time-based boundaries.
6. transaction_payload_hash: Calldata hash sent to the mempool.
7. broadcast_timestamp: Exact ms the tx was handed to the RPC/Relay.
8. network_visibility_mempool: Private-relay acknowledgement or public-mempool visibility, when provided by the client's broadcaster.
9. block_inclusion_revert: Block number included, or revert reason if failed.
10. final_outcome: Actual on-chain state change (tokens in/out).
11. external_anchor: Optional client-managed attestation hash, when the client has actually created one.

Execution-Quality Record

When the client has authoritative settlement data, record the decision-time quote, signed payload hash, broadcast/receipt status, realized token deltas, gas, and side-adjusted implementation shortfall. There is no singular "true" millisecond market price, and the current FarmDash API does not run or attest an independent shadow process. Missing evidence remains unavailable.

Trader-Grade Edge Gate (Additive)

Use this overlay before any spot swap, bridge, airdrop rotation, or protocol entry. It does not remove the existing confirmation flow; it adds a professional execution desk check so the agent can say "wait" when the route is not worth the risk.

Net Edge Equation

Before recommending action, estimate: $$\text{netEdgeUsd} = \text{expectedUpsideUsd} - \text{gasUsd} - \text{bridgeFeeUsd} - \text{expectedSlippageUsd} - \text{FarmDashFeeUsd} - \text{riskBufferUsd}$$

Where:

• $\text{expectedUpsideUsd}$ comes from simulate_points, optimize_portfolio, Trail Heat rank, or the user's explicit thesis.
• $\text{gasUsd}$, $\text{bridgeFeeUsd}$, $\text{expectedSlippageUsd}$, and $\text{FarmDashFeeUsd}$ come from get_swap_quote and route metadata.
• $\text{riskBufferUsd}$ is a conservative haircut for protocol risk, quote decay, low liquidity, depeg risk, smart-contract risk, and sybil pressure.

Default action thresholds:

Route Quality Checklist

Before asking for a signature, classify the route:

• Green: fresh quote, known spender, positive net edge, slippage inside budget, no depeg/allowance halt, expected output stable across two consecutive quotes.
• Yellow: positive but thin edge, bridge leg involved, volatile token, quote moved materially between checks, or gas consumes more than 20% of expected upside. Ask the user whether they want to wait.
• Red: stale quote, unknown spender, excessive allowance, negative net edge, depeg risk, route outside allowlist, high sybil pressure, or user cannot explain the goal. Halt.

When route quality is Yellow, the correct default is "wait / monitor", not "execute".

Two-Quote Drift Check

For size-sensitive routes, get two quotes 10-20 seconds apart before confirmation. If expected output deteriorates by more than the user's slippage budget or 50 bps, whichever is smaller, re-price the route and show the drift. Do not let the user sign the older quote.

Post-Trade Reconciliation

After execute_swap:

1. Call confirm_swap when a tx hash or fee event exists.
2. Treat the FarmDash fee event/history record as volume/fee metadata, not proof of realized token output or execution quality.
3. When authoritative receipt and token-delta evidence is available from the client or chain, compare expected output versus realized output and record slippage, route, gas, bridge time, request ID, provenance, and reason.
4. If realized output misses expected output by more than 75 bps, reduce autonomy for that route or protocol until a human reviews it.
5. If realized output evidence is unavailable, or settlement is pending or partial, label the metric unavailable and do not start a second dependent action.

Capital Efficiency Without Forced Risk

Idle capital is optionality and liquidity, not automatically a defect. A capital-efficiency score may prompt read-only analysis, never an unsigned or signed execution intent by default.

• Respect the user's reserve, gas, collateral, withdrawal, tax, and emergency-liquidity requirements before labeling funds idle.
• A high Trail Heat score is not a safety score or expected-return estimate.
• Draft a deployment proposal only when the user asks or a previously approved monitoring policy explicitly reque