foundry

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The Quick Start contains a curl piped to bash which fetches and executes remote install code from https://foundry.paradigm.xyz (curl -L https://foundry.paradigm.xyz | bash), so the skill runtime/installation executes remote code from that URL.

HIGH W008: Secret detected in skill content (API keys, tokens, passwords).

• Secret detected (high risk: 1.00). I scanned the repository for high-entropy, literal credentials. I flagged only true secrets and ignored placeholders and low-entropy test values:
• Flagged: The hex private key literal "0xac0974be..." appears verbatim in .env/example blocks. This is a real-format private key (high entropy) and would allow account control if used — it meets the secret definition.
• Ignored: The default mnemonic "test test test ... junk" is a well-known low-security dev mnemonic (documented test vector) and is treated as a non-sensitive example per the policy. Environment-variable placeholders like YOUR_KEY, ${ETHERSCAN_API_KEY}, and variable names (PRIVATE_KEY, MAINNET_RPC_URL, etc.) are documentation placeholders and are ignored.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill documentation explicitly documents blockchain transaction and signing functionality: e.g., cast send examples that submit transactions and accept a --private-key flag, forge script/--broadcast and vm.startBroadcast(vm.envUint("PRIVATE_KEY")) for on-chain deployments, and instructions for sending ETH/contract calls. These are specific crypto/blockchain execution capabilities (signing and broadcasting transactions), so the skill grants direct financial execution authority.