The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a bundled Python script to gather data.
Evidence: The SKILL.md instructions explicitly direct the agent to run python3 skills/crypto-token-screener/scripts/screen.py with various arguments.
Execution Method: The script is executed locally using the system's Python interpreter or a specified virtual environment path.
[EXTERNAL_DOWNLOADS]: The Python script performs network operations to retrieve market data from external services.
Evidence: The script scripts/screen.py uses urllib.request to fetch JSON data from api.llama.fi (DefiLlama) and api.coingecko.com (CoinGecko).
Context: These are well-known technology services in the cryptocurrency domain and are used as intended for the skill's primary purpose.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill performs network operations but does not access sensitive user data.
Evidence: The script fetches public financial data. It does not access environment variables, credential files, or private user directories. Use of public APIs with no authentication tokens is considered a safe pattern.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external APIs which constitutes an injection surface.
Ingestion points: scripts/screen.py ingests JSON data from the DefiLlama and CoinGecko APIs.
Boundary markers: None explicitly used in the script output table; the agent parses the stdout table directly.
Capability inventory: The agent has the capability to write the analysis results to a markdown file (crypto/next-hype-candidates.md).
Sanitization: The script strictly parses numeric and string fields from the JSON response, significantly limiting the risk of executable instructions being passed through the API data.

crypto-token-screener