Privy Agentic Wallets

**Judgement:** A strong, security-conscious skill for creating and operating agent-controlled crypto wallets via Privy. Well-suited for serious agentic onchain workflows, but inherently risky because it can move real funds. **What it does well** - Exposes core Privy wallet APIs: create/list/get wallets, create/get/delete policies, execute onchain transactions. - Emphasizes **policy-first design**: you are instructed never to create a wallet without attaching at least one safety policy (e.g., per-tx caps, chain restrictions). - Provides concrete examples for Ethereum/Base and highlights CAIP-2 chain identifiers, making multi-chain usage straightforward. - Includes a detailed security checklist and explicit **prompt-injection warnings**, tailored for AI-agent environments. **Key risks & limitations** - **Real funds at risk:** Any mistake in policy configuration, address, amount, or chain selection can lead to irreversible loss of assets. - **Prompt injection:** Despite the documented defenses, if the calling agent ignores the rules (e.g., follows untrusted text or web/email instructions), it could still send unintended transactions or weaken policies. - **Policy deletion is dangerous:** Removing policies can eliminate spending limits or chain restrictions; the skill requires explicit verbal confirmation, but higher-level orchestration must enforce that rigorously. - **Secrets management:** PRIVY_APP_ID and PRIVY_APP_SECRET must be correctly set and never surfaced back into the conversation or other skills; misconfiguration or leakage compromises wallets. - **Operational reliability not guaranteed:** The skill depends on the external Privy API. Downtime, rate limits, or network issues will affect behavior and must be handled by the calling system. **Recommended scenarios** - Building **agentic trading, payment, or automation systems** where an AI needs limited autonomous onchain capabilities under strict guardrails. - Server-side or backend workflows that must **avoid exposing private keys** but still let agents initiate transactions through a controlled API. - Prototyping advanced onchain agents in a **staging/testnet environment first**, before carefully rolling out to mainnet with conservative policies. **Use with caution** - Always follow the security.md checklist for every transaction (validate sender/recipient, amount, chain, and that the request comes directly from the user). - Never remove or relax policies without explicit, repeated user confirmation in the conversation. - Prefer minimal, tightly scoped policies (low spending limits, chain allowlists) and test thoroughly on testnets before enabling real-value operations.