The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive agent session logs, project history, and configuration files located in directories like ~/.claude/projects/, ~/.hermes/, and ~/.codex/. This data access is required for the skill to perform its primary function of token usage and cost auditing. No evidence of data transmission to external servers was found.
[COMMAND_EXECUTION]: The fleet.py script utilizes subprocess.run and os.startfile to invoke system handlers (like open or xdg-open) for displaying the generated HTML dashboard in the user's default browser. This execution is limited to opening the locally generated report.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from session logs and markdown files which could potentially contain untrusted instructions. The current implementation mitigates this by focusing on metadata extraction (token counts, model names, timestamps) and performing no dangerous operations on the message content itself.
[SAFE]: The skill implements security best practices, such as binding the local dashboard server to 127.0.0.1 and including a Host header check to prevent DNS rebinding attacks. It also includes specific logic to avoid command injection vulnerabilities when opening browser windows on Windows systems.

fleet-auditor