The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's installation process involves downloading an installer script and a pre-compiled plugin binary from the vendor's GitHub repositories.
Evidence: SKILL.md contains bash scripts fetching resources from raw.githubusercontent.com/okx/ and github.com/okx/.
Security Controls: The skill implements high-integrity practices by verifying the SHA256 checksum of the install.sh script and the compiled binary before execution, preventing local manipulation or intercepted download attacks.
[COMMAND_EXECUTION]: The skill programmatically invokes external command-line tools to manage blockchain wallets and sign transactions.
Evidence: src/onchainos.rs utilizes tokio::process::Command to execute the onchainos CLI tool for operations such as wallet contract-call and wallet addresses.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external blockchain sources, which represents a potential attack surface for indirect prompt injection.
Ingestion points: src/rpc.rs retrieves token symbols and names via eth_call, and src/commands/positions.rs fetches position metadata from TheGraph subgraphs.
Boundary markers: Data retrieved from the blockchain is processed as raw strings without explicit boundary delimiters in the Rust source code.
Capability inventory: The skill has the capability to initiate financial transactions on multiple EVM-compatible chains.
Sanitization: No specific filtering for adversarial instructions is performed on the token metadata strings before they are returned to the agent context.
Mitigation: The developer has included a 'Data Trust Boundary' protocol in SKILL.md (M08) that instructs the AI agent to filter fields and treat all CLI output as untrusted external content.

pancakeswap-v3-plugin