The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external blockchain RPC providers, such as position details, token pairs, and reward amounts. This creates a surface for indirect prompt injection if a malicious contract or RPC node returns crafted metadata. The skill addresses this by including a 'Data Trust Boundary' section in SKILL.md that instructs the agent to treat external content as untrusted and filter specific fields.
Ingestion points: src/rpc.rs (eth_call, eth_getLogs), src/commands/positions.rs, src/commands/farm_pools.rs.
Boundary markers: Explicitly present in the 'Data Trust Boundary' and 'Output field safety (M08)' sections of SKILL.md.
Capability inventory: Transaction submission via the onchainos CLI tool (src/onchainos.rs), network requests to blockchain RPC providers.
Sanitization: Specific field filtering instructions are provided to the AI agent to minimize data exposure.
[EXTERNAL_DOWNLOADS]: The setup instructions in SKILL.md download the onchainos CLI and the pancakeswap-clmm-plugin binary from repositories within the okx GitHub organization. These downloads are performed from a trusted vendor source and are verified using SHA256 checksums before being executed or installed.
[COMMAND_EXECUTION]: The skill invokes the onchainos CLI to perform wallet address resolution and transaction submission. This execution is the intended primary function of the skill for interacting with on-chain smart contracts.

pancakeswap-clmm-plugin