Elytro - Ethereum Wallet

The name/description (EIP-4337 smart contract wallet) align with the runtime commands (calls to an elytro CLI such as `elytro account list`, `elytro tx send`, etc.). Requesting the `elytro` binary is appropriate. Minor inconsistency: the registry lists no install spec, but the SKILL.md frontmatter and text recommend installing `@elytro/cli` via npm and require Node >=24 — the registry metadata and SKILL.md disagree about whether installation is handled for the user.

The instructions mandate using Telegram Inline Buttons and a specific callback_data naming scheme for all confirmations and navigation. That is a strong UI/platform requirement that is not declared elsewhere (no required messaging credentials or platform integration described). The SKILL.md also directs the agent to run CLI commands 'silently in the background' and to never accept text confirmations — this enforces a particular integration pattern and automated callbacks, which increases risk if the environment doesn't actually support safe button handling. The skill does not describe how it expects private keys to be managed (local key files, hardware wallet, remote signer), though the CLI will likely access wallet material; that omission is important.

This is instruction-only (no code files). The SKILL.md suggests installing the official CLI via npm (`npm install -g @elytro/cli`) which is a traceable install mechanism if you verify the package and repository. The registry's top-level install spec is absent, producing a mild inconsistency: the agent/installer may not automatically install the required binary. No arbitrary download URLs or archive extraction are present in the skill files themselves.

The skill declares no required environment variables or credentials, which superficially looks safe. However, the elytro CLI will likely need access to private key material or a signer; the SKILL.md does not state how keys are provided or protected. Because the skill enforces automated actions and callback-driven approvals, there is a risk that transaction-signing credentials available to the CLI could be used without clear documentation of consent boundaries. No unrelated secrets are requested by the skill itself, but the lack of key-management guidance is a notable gap.

The skill is not always-enabled and allows model invocation (default), which is normal. The SKILL.md's callback_data conventions and emphasis on automated Inline Button flows enable programmatic triggering of actions across agents — combined with CLI access to wallets this increases the blast radius, but the skill does not request persistent system-level privileges or manipulate other skills' configs. Overall, no elevated platform-level privileges are requested, but the operational design encourages automation that a user should evaluate carefully.