mcpField
T
thoughtproof-reasoning-check

thoughtproof-reasoning-check

github.com/moonpayAgent workflows11239 installsUpdated 23w ago
Repository

About thoughtproof-reasoning-check

ThoughtProof Reasoning Check is a verification skill for agents preparing trades or other autonomous actions. It uses adversarial multi-model critique to return a signed ALLOW, BLOCK, or UNCERTAIN verdict, with objections and confidence.

Key features

  1. 01Adversarial multi-model critique
  2. 02ALLOW, BLOCK, or UNCERTAIN verdicts
  3. 03Confidence and objection details
  4. 04Stake-based request tiers

Use cases

  1. 01Check reasoning before a token swap
  2. 02Pause actions with material reasoning defects
  3. 03Escalate decisions with insufficient evidence

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/moonpay/skills --skill thoughtproof-reasoning-check

After installing, type/thoughtproof-reasoning-checkin Claude Code to confirm it loaded.

Details

From the project README

ThoughtProof Reasoning Check

Goal

Before executing a trade or autonomous action, verify that the agent's reasoning is sound. ThoughtProof runs adversarial multi-model critique and returns a signed verdict: ALLOW, BLOCK, or UNCERTAIN.

Prerequisites

  • A funded wallet on Base (for x402 payment): mp wallet create

Usage

Check reasoning quality before a trade:

curl -X POST https://api.thoughtproof.ai/v1/check \
  -H "Content-Type: application/json" \
  -d '{"claim": "Swap $2K USDC to ETH. ETH at $2180, 6% below 30d MA, RSI 34.", "stakeLevel": "medium", "domain": "financial"}'

First call returns 402 with x402 payment details (amount, recipient, asset on Base). Pay, then retry the same request to receive the verdict.

Response

{
  "verdict": "ALLOW",
  "confidence": 0.82,
  "objections": [],
  "durationMs": 55000
}

Decision logic

VerdictMeaningAction
ALLOWReasoning passed verificationProceed
BLOCKMaterial reasoning defects or unacceptable riskDo not execute
UNCERTAINSafe escalation state — insufficient clean evidence to justify ALLOW or BLOCKGather more context or require review

Pricing

ThoughtProof pricing is stake-based.

Stake LevelStandardLiteUse case
low$0.01$0.005Routine low-value actions
medium$0.02$0.01Standard trades and service payments
high$0.05—Large or sensitive actions
critical$0.10—High-value or highly risky actions

Notes:

  • lite is only available for low and medium stake.
  • high and critical always execute on the standard tier.
  • Billing follows the executed tier, not the requested tier.

Example workflow

MoonPay + ThoughtProof: payment layer + verification layer

MoonPay handles the payment execution. ThoughtProof verifies whether the decision to pay is well-reasoned before settlement.

Example:

  1. User says: "Swap $5K USDC for ETH because CT is bullish."
  2. Agent prepares the MoonPay swap request.
  3. Before calling moonpay-swap-tokens, the agent sends the reasoning to ThoughtProof.
  4. ThoughtProof returns:
    • ALLOW → proceed with swap
    • BLOCK → stop, surface objections
    • UNCERTAIN → escalate, ask for clarification or review
  5. If verification passes, the MoonPay skill executes the swap.

Example failure:

  • "Buy $50K of a token because influencers are bullish"
  • ThoughtProof → BLOCK
  • MoonPay execution never happens

Example pass:

  • "Rebalance portfolio back to target allocation with defined slippage and stop-loss constraints"
  • ThoughtProof → ALLOW
  • MoonPay executes

This gives the agent stack two layers:

  • MoonPay answers: can the transaction execute?
  • ThoughtProof answers: should the transaction execute?

Related skills

  • moonpay-swap-tokens — Execute the swap after verification passes

Comments

Markdown supported