The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by aggregating external financial news.
Ingestion points: scripts/fetch_news.py retrieves news headlines and descriptions from multiple external RSS feeds (e.g., Reuters, WSJ, Bloomberg) and user-added custom feeds.
Boundary markers: The skill uses standard Markdown headers (e.g., ## Headlines) to separate data in the LLM prompt in scripts/summarize.py. It does not use robust delimiters or per-item isolation for untrusted content.
Capability inventory: The skill possesses the capability to send messages to WhatsApp and Telegram channels via the openclaw CLI and Lobster workflows (scripts/briefing.py, workflows/briefing.yaml).
Sanitization: scripts/summarize.py includes a HARDENED_SYSTEM_PROMPT that instructs the LLM to treat headlines as untrusted input. However, there is no mechanical escaping or sanitization of the headline text before it is interpolated into the prompt template.
[EXTERNAL_DOWNLOADS]: The skill performs legitimate network operations to fetch market data and news.
It fetches RSS feeds from various financial news providers defined in config/config.json.
It makes API calls to local tailnet endpoints (kalliope, gx10) for LLM processing using urllib.request in scripts/utils.py.
It utilizes the is.gd service to shorten URLs in scripts/summarize.py via a GET request.
[COMMAND_EXECUTION]: The skill uses subprocess.run for several utility tasks.
scripts/briefing.py calls the openclaw message tool to deliver briefings.
scripts/earnings.py and scripts/fetch_news.py invoke openbb-quote to retrieve market data.
scripts/translate_portfolio.py calls openclaw agent to translate headlines.
These calls are generally controlled and use hardcoded command paths or known CLI tools.

finance-news