The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accepts an arbitrary file path via the LOG_FILE input and processes it using various system tools. This allows a user to point the skill at sensitive system files (e.g., /etc/shadow, .env files, or SSH keys) instead of actual logs. If the agent has the necessary permissions, the skill will read and display the content of these files, leading to sensitive data exposure.
[COMMAND_EXECUTION]: The skill executes multiple shell commands (awk, grep, sed, head) and Python scripts using the user-provided LOG_FILE and TOP_N inputs. While the script attempts to use quotes for safety, the lack of strict validation on the LOG_FILE path allows for potential tool-specific exploitation (e.g., passing arguments to awk or grep via the filename parameter).
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Content is read from external files specified by the user in LOG_FILE across all workflow steps (Steps 2 through 6).
Boundary markers: None. The content of the logs is processed directly and then printed to the agent's output buffer.
Capability inventory: The skill uses shell tools and Python to generate reports, including "Example block commands" based on log data.
Sanitization: There is no sanitization of the log content before it is processed by regex or printed. An attacker who can write to a log file could inject content designed to mislead the agent or influence subsequent actions if the agent processes the output of this skill.

log-analysis