The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it processes untrusted data from the web and user arguments.
Ingestion points: The skill accepts company names via $ARGUMENTS and collects extensive financial/competitive data using background research agents (Task tool) from external web sources.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potential commands embedded within the financial data it retrieves.
Capability inventory: The skill uses shell execution (python3 tools/financial_rigor.py), task agents for web research, and file-writing capabilities to save reports to the user's home directory (~/巴菲特Checklist-...md).
Sanitization: No explicit sanitization or filtering of the retrieved data is mentioned in the instructions. However, the risk is inherent to the skill's primary purpose of research and is considered a standard operational characteristic.
[COMMAND_EXECUTION]: The skill executes local scripts and system commands as part of its normal workflow.
Evidence: It runs python3 tools/financial_rigor.py for valuation verification and date to establish a baseline for its report. These commands are executed locally using project-provided tools and do not involve remote code execution.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill generates reports based on its analysis.
Evidence: Results are written to the local filesystem at paths like ~/巴菲特Checklist-[公司名].md. No sensitive files (e.g., credentials or SSH keys) are accessed, and no unauthorized network exfiltration was detected.

investment-checklist