Onekey Security

Before any operation, read references/common.md for safety, chain, and scam rules.

Security Skill

Direct Invocation Fallback

• If this skill is directly invoked for a read-only wallet or market request, discover the schema-backed command and apply Cross-Domain Fallback instead of only saying another skill is needed.
• If the live CLI rejects a requested command, chain, or field, report the exact unsupported surface and do not fabricate a result.

Domain Rules

• This skill owns security-audit, security-simulate, approval-risk review, suspicious-token review, and security-sensitive preflight checks.
• Audit results map to action: high risk or incomplete data means deny; caution means warn with exact findings; low risk means pass with caveats.
• Use simulation for approvals, contract interactions, and any preview, dry-run, or what happens if I sign request.
• Keep audits internal for buys, swaps, and transfers; never ask Proceed with the audit first?.
• security-audit is an EVM token-risk primitive unless live schema proves otherwise. Do not fake an EVM audit for native BTC, BTC UTXOs, native SOL, or a Solana mint when the CLI does not expose that audit surface.
• For BTC and Solana safety checks, validate chain/address/token format, native-vs-wrapped semantics, amount, fee, and recipient mismatch. Use security-simulate only for schema-supported transaction calldata.
• Hardware wallet safety includes device authenticity, physical confirmation, passphrase mode, and no seed/private-key exposure. Device verification may use device-verify; PIN/passphrase/settings changes require explicit confirmation.
• App Transfer/App Transport Bot Wallet safety includes never displaying raw payloads, access tokens, ciphertext, decrypted credentials, mnemonics, seed phrases, or private keys. Prefer status summaries over secret material.
• Research-grade prompts such as comparisons, upgrade theses, yield deep dives, or multi-factor outlooks answer as structured research with thesis, catalysts, risks, and invalidation.
• Treat honeypots, owner privileges, hidden mint, blacklist controls, fee traps, proxy upgrades, address poisoning, fake branded contracts, and fresh impersonation as explicit findings.
• Treat Solana/EVM/BTC address-format mismatch and native-vs-wrapped confusion as security blockers for fund-moving flows.
• Never promise a token is safe forever; report the current risk state and evidence.

Domain Routing

Error Handling

• If an audit or simulation returns an error, report whether the risk result is unavailable, incomplete, caution, or high risk, then stop any dependent fund-moving flow.
• Never recover from security errors by treating an unknown token, hidden approval, hardware prompt, or unsupported chain as safe.

Fast Patterns

• is this token safe 0x... -> run an audit and answer with risk level plus reasons.
• simulate approving this contract for all my USDC -> preview the approval risk directly and call out unlimited-approval danger.
• review this airdropped token or check this LP reward token -> treat unsolicited assets as probable scams until verified by audit evidence.
• swap 500 USDC to WETH at 0x... or swap 1 ETH to USDT at 0x1234... -> stop as contract mismatch or scam-token risk, not as a venue question.
• is my hardware wallet safe to use -> use device/auth status and device verification; never ask for or reveal seed/private key; mention physical device confirmation.
• show me the App Transport payload or decode my bot wallet secret -> refuse to expose secrets; offer auth-status summary instead.
• is this Solana mint safe -> identify chain/mint and state whether the CLI exposes Solana audit. Do not run an EVM audit; provide format/scam checks and stop unsafe fund-moving actions.
• send BTC to 0x... or send SOL to 0x... -> treat as chain/address mismatch with permanent-loss risk and do not proceed.