The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands for its own internal scripts (bun yield_trace.ts and bun boundary_pnl.ts) and testing harness. These are restricted to local scripts provided within the skill package.
[EXTERNAL_DOWNLOADS]: The skill fetches blockchain data from well-known and trusted services including Blockscout (base.blockscout.com, eth.blockscout.com, arbitrum.blockscout.com, optimism.blockscout.com) and DefiLlama (coins.llama.fi). These are standard data sources for DeFi applications and are used neutrally for historical pricing and block timestamping.
[COMMAND_EXECUTION]: The skill uses public RPC endpoints for Base, Ethereum, Arbitrum, and Optimism to perform eth_call operations. These calls are limited to standard read-only methods like balanceOf and convertToAssets to value wallet holdings.
[SAFE]: The skill implements a robust 'RPC Sanity Guard' (assertIdleRpcSanity) which throws an error and exits if live data significantly diverges from ledger data, preventing the output of incorrect financial figures due to flaky network providers.
[PROMPT_INJECTION]: No prompt injection or behavior override patterns were found. The instructions are focused on technical accounting logic and conservation formulas.
[DATA_EXFILTRATION]: No sensitive local files (e.g., .ssh, .aws, .env) are accessed. The network activity is restricted to public blockchain APIs and RPCs for the intended purpose of the skill.

defi-pnl