The Agent Skills Directory

[COMMAND_EXECUTION]: The script 5minbtc-news.py executes local sub-scripts via subprocess.run to gather news from Telegram channels.
Evidence: The script dynamically constructs paths to telegram-treenews.py and telegram-cointelegraph.py within the skill's scripts/ directory and executes them using the Python interpreter.
[EXTERNAL_DOWNLOADS]: The skill fetches market data and news from several external domains to perform its analysis.
Evidence: It makes network requests to data-api.binance.vision for K-line data, api.coinbase.com for spot prices, and several news RSS feeds including CoinDesk and CoinTelegraph.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted news content from external websites, which could potentially contain malicious instructions targeting the AI agent.
Ingestion points: News headlines and bodies are scraped from multiple third-party crypto news websites in 5minbtc-news.py.
Boundary markers: While data is structured in JSON, the agent's instructions in SKILL.md prompt for a "comprehensive analysis" of the content without specific guardrails against embedded instructions.
Capability inventory: The skill has access to terminal commands, web tools, and local script execution.
Sanitization: No specialized sanitization is performed on the scraped news text before it is presented to the agent.
[CREDENTIALS_UNSAFE]: The script 5minbtc-news.py reads the user's shell configuration file to extract API keys.
Evidence: The function load_env() in 5minbtc-news.py opens ~/.bashrc and scans for lines starting with export that contain _API_KEY to populate the environment variables used by the tool.

5minbtc