Prove Before Act — Accountability Layer for AI Agents

This skill describes how to anchor reasoning (WHY) before an action and its result (WHAT) afterwards with Prove Before Act. It provides an auditable record when the returned proof is confirmed; it does not itself enforce an execution policy.

Current product and compatibility notice: Prove Before Act is the product name. The ClawHub install slug @jasonxkensei/xproof is retained as a legacy compatibility identifier for existing OpenClaw installations. Install the canonical Python package as prove-before-act; its supported Python module namespace is xproof. XProofClient, XPROOF_*, xProof-Action, and jasonxkensei/xProof are legacy protocol, action, or repository identifiers retained for compatibility. MX-8004 support is not active unless GET /api/mx8004/status reports status: "active"; the current production response is status: "not_configured".

The Core Pattern: Prove Before & After Act

1. Reason (WHY) → 2. Hash locally → 3. Anchor WHY on-chain → 4. Get proof_id → 5. Execute → 6. Hash result (WHAT) → 7. Anchor WHAT on-chain → 8. Full 4W trail

Anchor reasoning (WHY) on-chain before execution. Anchor the actual result/output (WHAT) after for a complete 4W audit trail. The proof_id is your immutable pre-execution commitment — if something goes wrong, you have forensic evidence that the reasoning preceded the action.

4W breakdown — what each letter means and who provides it:

Prove Before Act records WHAT / WHEN / WHY context. MX-8004 may provide WHO when its integration is active. Check live proof and MX-8004 responses before treating the resulting record as complete.

Agent reference: https://provebeforeact.com/agent-context — 13 Q&A sections written specifically for autonomous agents: production deployment patterns, retry policy, x402 payment walkthrough, 4W audit trail deep-dive, privacy model, fleet monitoring, latency benchmarks, and framework integration examples (LangChain, CrewAI, AutoGen, LlamaIndex, OpenAI Agents SDK, Fetch.ai).

Why Prove Before Act? An agent that can prove its reasoning before acting is fundamentally more trustworthy than one that cannot. It creates an unbreakable causal chain: the intent existed before the action, and the blockchain makes that sequence immutable.

Why audit? Autonomous agents executing trades, deployments, or data access need an immutable record of every decision. The Agent Audit Log Standard describes a blocking pattern your agent can adopt: obtain a proof before executing. Enforcement is implemented in your agent code; Prove Before Act supplies the proof record.

Why violations? Operators can use recorded violations as an accountability signal when the relevant Base integration is deployed and active. Confirm contract and event availability from the live documentation before automating a policy.

Pricing

Flat rate per proof -- no tiers, no volume discounts, same price whether you anchor 1 or 100,000 proofs. The current per-proof price is served live at /api/pricing; multiply it by your volume for a cost estimate.

Payment: USDC on Base (x402, no account) or EGLD on MultiversX (ACP/wallet) or prepaid credits.

Quick Install

mkdir -p .agent/skills/prove-before-act/references

# Core Skill — from the canonical main repository
curl -sL https://raw.githubusercontent.com/jasonxkensei/prove-before-act/main/clawhub-publish/xproof/SKILL.md \
  > .agent/skills/prove-before-act/SKILL.md

# Reference Manuals
for f in certification x402 mcp; do
  curl -sL "https://raw.githubusercontent.com/jasonxkensei/prove-before-act/main/clawhub-publish/xproof/references/${f}.md" \
    > ".agent/skills/prove-before-act/references/${f}.md"
done

Source verification: All files above are served from github.com/jasonxkensei/prove-before-act — the canonical repository for the server code, contracts, and SDKs.

First successful integration

Use this four-step path to create a verified record without exposing the original content:

1. Install — run openclaw skills install @jasonxkensei/xproof. Its package slug is legacy; the installed skill and product are Prove Before Act.
2. Hash locally — compute a SHA-256 hash of the decision or output. Do not send raw text, files, prompts, or credentials.
3. Anchor — call certify_file through MCP or POST /api/proof with the hash and a filename.
4. Verify before relying on it — retain the returned proof_id and use verify_proof or GET /api/proof/<proof_id> until the status is confirmed. Your own operator policy decides whether to continue, retry, request review, or block an action.

Python SDK quick start

Install the canonical package:

pip install prove-before-act

Then import its supported xproof module namespace and hash the content locally:

import hashlib
import json

from xproof import XProofClient

decision = {"action": "publish report", "reason": "approved by operator"}
sha256_hex = hashlib.sha256(
    json.dumps(decision, sort_keys=True).encode("utf-8"),
).hexdigest()

client = XProofClient.register("my-agent")
proof = client.certify_hash(sha256_hex, "decision.json", "my-agent")

Data & Privacy

Transparency notice — what leaves your environment when using this skill.

What provebeforeact.com does NOT receive: raw file contents, prompt text, model weights, environment variables, or any field you do not explicitly pass.

Data retention: proof records are permanent and public by default (that is the point — immutable evidence). Before certifying, confirm you want the hash publicly anchored on MultiversX mainnet forever.

Skill file: this file is intentionally public and contains no secrets. It is documentation only — API patterns, pricing, and integration examples. Safe to audit, share, or version-control.

Security

• NEVER commit API keys to a public repository.
• ALWAYS add .env to your .gitignore.
• API keys are prefixed pm_ -- treat them like passwords.
• x402 mode requires no API key (payment replaces authentication).
• NEVER send plaintext content to provebeforeact.com -- always hash locally first (sha256sum, crypto.subtle.digest, or equivalent). The only field Prove Before Act accepts is file_hash (64-char SHA-256 hex). No raw text, documents, or binary data should ever leave your environment.
• x402 is opt-in and autonomous -- once enabled, your agent can initiate USDC payments on Base without per-transaction confirmation. Configure a spending cap in your agent framework and require human approval above your threshold before enabling x402 in production.
• llms.txt and llms-full.txt are static documentation references -- load them once at install time for API reference, not at runtime on every call. Fetching them dynamically on each invocation creates an unnecessary runtime dependency on provebeforeact.com availability and a potential prompt-injection surface if the file is ever compromised.
• Guard/enforcement templates are versioned in the repository -- never fetch agent enforcement code from a runtime URL. Use the pinned versions in references/ or the canonical prove-before-act SDK package on PyPI and npm. Python users install prove-before-act and import its supported xproof module namespace.

Configuration

Option A: API Key Authentication

# ---- Prove Before Act ---------------------------------------------------------------
XPROOF_API_KEY="pm_..."                          # Your API key (from provebeforeact.com)
XPROOF_BASE_URL="https://provebeforeact.com"             # Production endpoint

Get an API key at provebeforeact.com (connect wallet, go to Settings > API Keys).

Option B: x402 Payment Protocol (No Account Required)

No configuration needed. Pay in USDC on Base (eip155:8453) directly in the HTTP request. The 402 response header tells your agent exactly what to pay (the current flat per-proof price, also served at /api/pricing).

WARNING -- autonomous payments: x402 is an opt-in mode that enables your agent to initiate on-chain USDC transactions without per-transaction user confirmation. Before enabling x402 in production:

• Set a spending cap in your agent framework (e.g. max $N/day or $N/session).
• Require human approval for any single call that would exceed your risk threshold.
• Note that POST /api/batch supports up to 100 items per call -- at the current per-item price, a batch of 100 costs 100× that price.
• Disable x402 entirely in environments where autonomous spending is not authorised.

1. Core Skills Catalog

1.1 Proof Anchoring (REST API)

Full Reference

1.2 Proof Anchoring (MCP -- JSON-RPC 2.0)

Full Reference

1.3 Payment (x402)

Full Reference

x402 is not a separate skill -- it is a payment method. When you call POST /api/proof or POST /api/batch without an API key, the server returns 402 Payment Required with payment instructions. Your agent pays in USDC on Base and retries with an X-Payment header.

2. Coherence Loop — close the WHY→WHAT link

Anchoring the WHY and the WHAT is not enough: you must link them. An unlinked WHY anchor shows as divergent in your public coherence history after 1 hour, and after the 2-hour TTL it is additionally flagged as a proposed fault violation — both lower your public coherence rate.

Full loop: check_coherence (WHY) → execute → certify_file with metadata.why_proof_id (WHAT) → POST /api/coherence/link.

# Step 4 — close the loop (API key required; both proofs must be yours)
curl -X POST https://provebeforeact.com/api/coherence/link \
  -H "Authorization: Bearer pm_..." \
  -H "Content-Type: application/json" \
  -d '{"why_proof_id": "<UUID from check_coherence>", "what_proof_id": "<UUID from certify_file>"}'
# → 200 {"success": true, "coherence_check": {"coherence_score": 85, ...},
#        "score_breakdown": {"linked": true, "what_within_1h": true,
#                            "what_references_why": true, "what_confirmed_on_chain": false,
#                            "execution_preceded_intent": false}}

Score (0–100): 50 for linking + 15 if WHAT within 1h of WHY + 20 if WHAT's metadata.why_proof_id references the WHY + 15 if WHAT confirmed on-chain. WHAT certified before the WHY → base halved to 25.

Error cases:

SDK helpers: client.link_coherence(why_proof_id, what_proof_id) (Python xproof ≥ 0.2.10) · client.linkCoherence(whyProofId, whatProofId) (npm prove-before-act ≥ 0.1.11).

Check your history: GET https://provebeforeact.com/api/agents/<wallet>/coherence — public; per-anchor status (linked / pending / divergent) + aggregate coherence rate.

3. Webhooks

Supply an optional webhook_url field on POST /api/proof or POST /api/batch to receive a callback when the proof is confirmed on-chain.

Scope — the webhook payload contains only:

No raw file content, no API keys, no account information, and no metadata beyond the above is ever sent to the webhook endpoint.

Authentication: Every delivery includes an X-Webhook-Signature header containing an HMAC-SHA256 signature computed with a per-relationship secret. Verify this signature before processing the payload. Retry policy: 3 attempts with exponential backoff (1 s, 5 s, 30 s).

SSRF protection: provebeforeact.com validates webhook_url before delivery. Private IP ranges (RFC 1918), loopback (127.x, ::1), link-local, and non-HTTPS destinations are blocked. DNS rebinding is mitigated by pinning the resolved socket address to the pre-validated IP at connection time.

# Example proof request with webhook
curl -X POST https://provebeforeact.com/api/proof \
  -H "Authorization: Bearer pm_..." \
  -H "Content-Type: application/json" \
  -d '{
    "file_hash": "a1b2c3...",
    "filename": "output.json",
    "webhook_url": "https://your-agent.example.com/webhooks/prove-before-act"
  }'

9. Violations Layer (Base)

The violation flow is optional and deployment-dependent. Confirm the live contract and event status before relying on it for Base enforcement or trust scoring.

Violation Types

Deployed Base contract sources retain their legacy Solidity names and cannot be renamed on-chain: XProofViolations.sol | ViolationWatcher.sol

Docs: https://provebeforeact.com/docs/base-violations

10. Agent Proof Standard

Prove Before Act implements the open Agent Proof Standard -- a composable, chain-agnostic format for agent accountability.

Full specification: AGENT_PROOF_STANDARD.md

Standard API: GET /api/standard | GET /api/standard/validate (POST)

11. Discovery Endpoints

12. Command Cheatsheet

# Hash locally first -- the original content must never leave your environment.
# Prove Before Act only receives the SHA-256 hex hash, filename, and metadata you choose to share.
sha256sum myfile.pdf | awk '{print $1}'
# Then POST the hash to /api/proof

# Anchor via MCP
curl -X POST https://provebeforeact.com/mcp \
  -H "Authorization: Bearer pm_..." \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"certify_file","arguments":{"file_hash":"...","filename":"myfile.pdf"}}}'

# Verify a proof
curl https://provebeforeact.com/api/proof/<proof_id>

# Get badge (embed in README)
![Prove Before Act](https://provebeforeact.com/badge/<proof_id>)

# Batch anchor
curl -X POST https://provebeforeact.com/api/batch \
  -H "Authorization: Bearer pm_..." \
  -d '{"files":[{"file_hash":"...","filename":"a.txt"},{"file_hash":"...","filename":"b.txt"}]}'

# Close the WHY→WHAT coherence loop (after check_coherence + certify_file)
curl -X POST https://provebeforeact.com/api/coherence/link \
  -H "Authorization: Bearer pm_..." \
  -d '{"why_proof_id":"<why-uuid>","what_proof_id":"<what-uuid>"}'

# Health check
curl https://provebeforeact.com/api/acp/health