The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process significant amounts of untrusted data from the web (industry reports, company profiles, and financial forecasts) to synthesize investment reports.
Ingestion points: Extensive use of web search tools to gather data for logical chain verification and global company scans.
Boundary markers: Absent. The instructions do not specify delimiters or tell the model to ignore potential instructions hidden within the research material.
Capability inventory: The agent has access to web search, shell execution (via bash to run audit scripts), and file writing (to save the final report).
Sanitization: The skill mitigates risks by requiring the agent to "cross-check financial data," use "exact arithmetic tools for valuation," and perform a 15% random sample audit using a local script.
[COMMAND_EXECUTION]: The skill provides explicit instructions for the agent to execute shell commands to run a local auditing tool.
Evidence: Instructions to run python3 ~/ai-berkshire/tools/report_audit.py with arguments like --report and --results to verify the generated report's integrity.

industry-research