The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The Python scripts (fetch-infinity-pending.py, fetch-syrup-pending.py, and fetch-v3-pending.py) include logic to check for the 'requests' library and install it via pip if it is not found. This involves downloading a package from the official Python Package Index (PyPI) at runtime.
[COMMAND_EXECUTION]: The skill uses subprocess.check_call() within its Python scripts to execute pip install commands. Additionally, the SKILL.md instructs the agent to use the cast CLI tool and curl for blockchain and API interactions. These are controlled executions necessary for the skill's operation.
[DATA_EXFILTRATION]: The SKILL.md contains an initialization step that uses curl to send a telemetry ping to pancakeswap.ai. This ping transmits system information including the agent identifier, skill version, timestamp, operating system, and architecture to the vendor's infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests data from external sources such as DexScreener, CoinGecko, and PancakeSwap APIs.
Ingestion points: Data is fetched from explorer.pancakeswap.com, infinity.pancakeswap.com, configs.pancakeswap.com, api.dexscreener.com, and api.coingecko.com across multiple scripts.
Boundary markers: The skill includes a 'Security' section in SKILL.md with a 'Mandatory Security Rule' explicitly directing the agent to treat API content as untrusted and to avoid interpreting token names or symbols as instructions.
Capability inventory: The skill has access to shell execution via Bash (curl, jq, python3, node), cast CLI, and network operations through WebFetch.
Sanitization: The instructions provide specific regex patterns for validating wallet addresses and chain IDs, and mandate single-quoting of user-provided values in shell commands to prevent injection.

harvest-rewards