The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from news RSS feeds which could theoretically attempt to influence automated trading decisions. \n 1. Ingestion points: rwa_alpha.py via the fetch_news_headlines function. \n 2. Boundary markers: A mandatory 'Live Trading Confirmation Protocol' in SKILL.md requires explicit typed user confirmation for all live transactions. \n 3. Capability inventory: On-chain wallet access and token swapping via the onchainos CLI. \n 4. Sanitization: Signals are matched against a strict internal playbook using regex and a constrained LLM classification prompt. \n- [COMMAND_EXECUTION]: The skill performs local command execution to interact with blockchain infrastructure. Evidence in rwa_alpha.py shows the script calling subprocess.run on the onchainos CLI tool. \n- [EXTERNAL_DOWNLOADS]: The skill connects to several external services at runtime to fetch market intelligence. Evidence in rwa_alpha.py and plugin.yaml includes requests to news.google.com, gamma-api.polymarket.com, and api.anthropic.com. \n- [DYNAMIC_EXECUTION]: The script includes functionality to modify its own configuration file during setup. Evidence in rwa_alpha.py shows the _save_config_to_disk function writing user settings back into config.py.

rwa-alpha