The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection as it processes data from logs/revenue.log when generating summaries and reports. • Ingestion points: Transaction descriptions are read from logs/revenue.log by the main.py and main.sh scripts. • Boundary markers: Technical delimiters to isolate untrusted data are absent from the log file format, although the SKILL.md provides defensive instructions to the agent. • Capability inventory: The system performs shell command execution via the documented ./main.sh interface. • Sanitization: The underlying scripts do not programmatically sanitize or escape the content read from the log files; input validation and safety are managed through high-level agent instructions.
[COMMAND_EXECUTION]: The skill utilizes shell and Python scripts to manage local financial data. This local execution is restricted to the skill's stated purpose of revenue logging and reporting.
[EXTERNAL_DOWNLOADS]: The SAFETY.md file references the developer's domain at faberlens.ai for detailed security reports and configuration. This is a neutral reference to official vendor documentation.
[COMMAND_EXECUTION]: The documentation includes examples of dangerous system commands like chmod 777 or rm. These are presented as test cases to demonstrate the effectiveness of the skill's security guardrails and are not used by the skill for its primary operations.

ai-revenue-tracker-hardened