Pre-execution transaction safety guardrail for Pharos Network AI Agents.
Every write transaction on Pharos passes through this skill first. It simulates the transaction, checks the target contract, detects dangerous patterns, and returns a risk score 0–100 with a clear PROCEED / WARN / BLOCK recommendation — before anything is signed or sent.
When an AI agent is about to call cast send or forge script --broadcast, this skill runs 6 read-only checks:
| # | Check | How | Risk Contribution |
|---|---|---|---|
| 1 | Simulation | eth_call with --from sender |
+20 if reverts |
| 2 | Contract existence | eth_getCode |
+50 if empty / +15 if < 100 bytes |
| 3 | Selector decode | cast 4byte + known dangerous table |
+35 for ownership/upgrade ops |
| 4 | Unlimited approval | Calldata pattern match on MAX_UINT256 | +70 → auto-BLOCK |
| 5 | Value check | cast balance + Chainlink PROS/USD |
+20 if > 50% wallet balance |
| 6 | Gas estimation | eth_estimateGas |
+5 if estimate fails |
Score → Action:
| Score | Level | Action |
|---|---|---|
| 0 – 29 | SAFE ✅ | PROCEED |
| 30 – 49 | LOW ✅ | PROCEED with note |
| 50 – 69 | MEDIUM |
WARN — require confirmation |
| 70 – 84 | HIGH 🔴 | BLOCK |
| 85 – 100 | CRITICAL 🚨 | BLOCK |
Override rules (applied after scoring):
- Unlimited approval → score = max(score, 70) → always BLOCK
- No contract code → score = max(score, 85) → always CRITICAL
- changeOwner / upgradeTo / setOwner in calldata → score = max(score, 70)
🛡️ Pharos Transaction Guardrail
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Target: 0xc879c018...1815 (USDC on Pharos mainnet)
Function: approve(address,uint256)
Value: 0 PROS
Risk Score: 70/100 — HIGH 🔴
Action: BLOCK
Checks:
✅ simulation: approval would succeed
✅ contract_exists: real contract, 1798 bytes
✅ selector: approve(address,uint256)
🔴 unlimited_approval: MAX_UINT256 detected — permanent access to ALL your USDC
✅ value_check: no native value sent
✅ gas_estimate: 71,897 gas
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🚫 BLOCKED — Use exact amount instead (e.g., 500000000 for 500 USDC)
git clone https://github.com/hosein-ul/pharos-tx-guardrail \
~/.claude/skills/pharos-tx-guardrailThe skill is auto-triggered when the agent detects any transaction-sending intent.
Submit https://github.com/hosein-ul/pharos-tx-guardrail in the Skill Hub.
git clone https://github.com/hosein-ul/pharos-tx-guardrail
# Point your agent to pharos-tx-guardrail/SKILL.md as the entry pointPrerequisite — Foundry:
curl -L https://foundry.paradigm.xyz | bash && foundryup
cast --version # verifypharos-tx-guardrail/
├── SKILL.md ← Agent entry point (read this first)
├── assets/
│ ├── networks.json ← RPC URLs, chain IDs, explorer URLs
│ ├── tokens.json ← USDC, WETH, WPHRS addresses
│ └── oracles.json ← Chainlink Push Engine feed addresses
├── references/
│ ├── tx-guardrail.md ← 6-check pipeline with exact cast commands
│ └── risk-patterns.md ← Dangerous selectors DB + detection logic
└── evals/
└── evals.json ← 4 test scenarios
| Network | Chain ID | RPC | Native |
|---|---|---|---|
| Pacific Mainnet | 1672 | https://rpc.pharos.xyz |
PROS |
| Atlantic Testnet | 688689 | https://atlantic.dplabs-internal.com |
PHRS |
Default: mainnet (chain 1672).
Chainlink oracles used for USD value calculation:
| Feed | Address |
|---|---|
| PROS/USD | 0x9356C87a48F913d11C87a0d4b8cD16CD04624BF3 |
| ETH/USD | 0x092ff0175Be8B2e83Ca5740d3EB13C6225901fa7 |
| USDC/USD | 0x8d08eA83A55ad1e805b5660F5eC76C99C6aF5eaf |
Hackathon page · Pharos Network · Companion skill: pharos-rwa-yield-router