The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted log data from external systems, creating a surface for indirect prompt injection if the parsed output is consumed by an LLM without adequate sandboxing.
Ingestion points: Reads Windows EVTX logs, Linux auth.log, and web access logs in SKILL.md (Steps 2-4) and scripts/agent.py (parse methods).
Boundary markers: The instructions do not define boundary markers or explicit "ignore" instructions for the ingested log content.
Capability inventory: The skill performs file writes to the /cases/ directory and executes multiple shell commands for parsing, correlation, and report generation.
Sanitization: While the agent uses regular expressions for parsing, it does not explicitly sanitize log content to prevent it from being interpreted as instructions if fed back to an agent context.
[PRIVILEGE_ESCALATION]: The skill utilizes elevated privileges to install necessary forensic utilities.
Evidence: SKILL.md includes the command sudo apt-get install libevtx-utils to install system-level log export tools.
[EXTERNAL_DOWNLOADS]: The skill installs external dependencies required for log analysis.
Evidence: SKILL.md executes pip install python-evtx and sudo apt-get install libevtx-utils. These are well-known and standard tools for the described forensic tasks.
[COMMAND_EXECUTION]: The skill executes various shell commands and Python scripts to process forensic data.
Evidence: SKILL.md employs find, grep, awk, ausearch, and Python heredocs (python3 << 'PYEOF') to extract and correlate log events.

performing-log-analysis-for-forensic-investigation