The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches financial data from multiple external sources including api.polygon.io, api.notion.com, finnhub.io, and sec.gov. It also contains instructions to git clone a repository from github.com/hamelsmu/hamelnb.git to enable a stateful Jupyter kernel for iterative exploration.
[INDIRECT_PROMPT_INJECTION]: The skill ingests substantial amounts of untrusted natural language data from external sources, which could contain malicious instructions.
Ingestion points: fetch_financials.py extracts Business, Risk Factors, and MD&A sections from SEC 10-K filings and earnings call transcripts from Finnhub.
Boundary markers: No delimiters or warnings to ignore instructions are utilized when processing these text blocks.
Capability inventory: The skill has extensive Python execution capabilities via local analysis scripts and the hamelnb Jupyter integration.
Sanitization: Natural language data from external sources is not sanitized or filtered before being processed by the agent.
[COMMAND_EXECUTION]: The skill frequently executes shell commands to run financial analysis scripts (e.g., calculate_reproduction_value.py, analyze_roic_wacc_spread.py) and manage the jupyter-live-kernel environment.
[CREDENTIALS_UNSAFE]: The skill attempts to retrieve API keys for financial services by reading sensitive local configuration files, including ~/.hermes/.env and internal config/api_keys.json files within the skill directory.

data-science