feat: add ThoughtProof reasoning verification action provider

[ThoughtProof]

Summary

New action provider for pre-execution reasoning verification — adversarial multi-model critique before high-stakes agent actions.

Actions

Action	Description
verify_reasoning	Runs Claude + Grok + DeepSeek adversarial critique on a claim. Returns ALLOW or HOLD with confidence score and key objections.

Why

Agents executing DeFi transactions, fund transfers, or irreversible actions need a reasoning check layer — not just token safety, but decision quality verification:

1. Agent receives: "Swap 500 USDC for TOKEN_X at market"
2. Agent calls verify_reasoning({ claim, stakeLevel: 'high', domain: 'financial' })
3. If HOLD → agent refuses and surfaces objections to the user
4. If ALLOW → agent proceeds with confidence

Design

• Fail-safe: API errors return HOLD (never silently proceed)
• Stake-aware: low / medium / high / critical levels route to appropriate model depth
• Signed attestations: responses include JWKS-signed EdDSA attestation for audit trails
• Payment: x402 micropayments on Base (USDC) via api.thoughtproof.ai

Links

• API: https://api.thoughtproof.ai
• Docs: https://thoughtproof.ai/skill.md
• npm MCP: npx @thoughtproof/mcp-server

[cb-heimdall]

🟡 Heimdall Review Status

Requirement	Status	More Info
Reviews	🟡 0/1	Denominator calculation Show calculation 1 if user is bot 0 1 if user is external 0 2 if repo is sensitive 0 From .codeflow.yml 1 Additional review requirements Show calculation Max 0 0 From CODEOWNERS 0 Global minimum 0 Max 1 1 1 if commit is unverified 1 Sum 2

[ThoughtProof]

Updated this branch to align with pot-sdk v2.0 (now live at api.thoughtproof.ai).

Changes:

• Verdict enum: ALLOW / BLOCK / UNCERTAIN (was ALLOW / HOLD)
• UNCERTAIN now has explicit handling as a safe escalation state
• Error/fallback case updated from HOLD to BLOCK
• Updated JSDoc comments

HOLD and DISSENT are now internal calibration states only — they no longer appear in public API responses.

───

Example usage with AgentKit:

const result = await agent.run(
"Swap $5,000 USDC for ETH because CT is bullish"
);
// ThoughtProof intercepts before execution:
// 🚫 BLOCK — "sentiment-based reasoning without data support"

const result2 = await agent.run(
"Rebalance portfolio to 60/40 ETH/USDC, ETH is 6% below 30d MA, stop-loss at -8%"
);
// ThoughtProof clears:
// ✅ ALLOW (confidence: 0.87)

The action provider calls POST /v1/check before any irreversible execution. BLOCK and UNCERTAIN both prevent the action from proceeding — UNCERTAIN surfaces a human-review prompt instead of a hard stop.

[ThoughtProof]

Added Python SDK implementation alongside the existing TypeScript provider.

New in this update:

Python action provider with two actions:

• verify_reasoning — multi-model adversarial consensus via POST /v1/check
• get_verification_receipt — EdDSA-signed JWT receipts via POST /v1/verify (audit trails, compliance, on-chain settlement)

20 test cases covering schemas, provider init, API success/error, and auth.

Both TypeScript and Python SDKs now have ThoughtProof reasoning verification. Auth via THOUGHTPROOF_API_KEY env var.