REDD is non-custodial by design. Keys and funds stay with the user, on the user's machine. The following are hard rules across the codebase:

• A private key or mnemonic is never logged, printed, served, or committed.
• The wallet exposes only an address and a signing function — never the raw key.
• Every transaction is simulated (eth_call) before broadcast; a failed simulation aborts.
• Automation is fuel-gated and capped (per-transaction and daily spend caps); when a guard fails it pauses and notifies rather than sending.
• The X402 gateway (when hosted) never returns or logs the LLM/facilitator key; errors are surfaced as generic messages so a secret can never leak through an error.

In scope:

• Smart contracts in contract/src (ReddAgent, ReddCredits, ReddToken).
• The CLI in cli/src — wallet handling, signing, execution, and the MCP server.

Out of scope:

• The pons launchpad / bonding curve and Robinhood Chain infrastructure themselves.
• Third-party dependencies (report those upstream).

Please report suspected vulnerabilities privately — do not open a public issue for a security bug. Reach out via a direct message on X: https://x.com/ReddAgentRH.

Include:

• a description and impact,
• steps to reproduce (or a proof-of-concept),
• affected files / contracts / versions.

We aim to acknowledge within a few days. Responsible disclosure is appreciated; please give us reasonable time to ship a fix before any public disclosure.